![\"\"](\"https://cdn-images-1.medium.com/max/1024/0*HU_zvMO5XXpvhig-\")
By Luis Cisneros, CEO
Every serious AI company in healthcare right now is building the same product in a slightly different order, which is the kind of convergence that usually signals either genuine consensus or a collective failure of imagination. In this case it is the second one, dressed up in enough pitch-deck geometry to pass for the first. Abridge, OpenEvidence, Heidi, Epic, and the long tail of companies crowding in behind them have all settled on the same three wedges. AI scribe for documentation. AI clinical decision support for reasoning. AI billing for revenue. Abridge announced on April 15, 2026 that it was integrating NEJM and JAMA content into its CDS, closing the triangle for a company that started in scribe and billing. OpenEvidence ran the sequence in reverse, beginning with CDS in 2022 powered by the same journals, launching Visits for scribe in August 2025, and extending into billing through its Tandem partnership. Heidi started with scribe and bolted on Evidence CDS later. Epic is playing catch-up from billing outward, leaning on Microsoft because it is not AI-native and has no plausible path to being so. The strategic logic is clean. Capture the highest-frequency pain point, expand outward, close the triangle, declare yourself the operating system. It is the kind of plan that makes investor decks sing and clinical reality groan.
The only problem is that everyone is optimizing the wrong system. Every wedge in that triangle is provider-centric and encounter-centric by construction. Document what happened in the room. Reason about it. Bill for it. It is a reactive architecture built around discrete events, being built at the exact moment the healthcare system is trying to move away from discrete events entirely. Value-based care inverts the logic of the encounter. It pays for longitudinal outcomes rather than episodes, redistributes risk toward whoever can actually change behavior over time, and rewards trajectory rather than volume. Building faster tools for the encounter layer while the encounter layer is losing economic primacy is a strange way to win a market, and yet here we are, with billions of dollars chasing companies that have won the current war without doing much to prepare for the next one.
Adding AI to an archaic workflow does not modernize the workflow. It makes the archaic workflow faster and cheaper to run, which is the kind of improvement that is real but bounded. The United States already spends more per capita on healthcare than any peer country and produces worse outcomes across almost every dimension that matters. Life expectancy, maternal mortality, preventable hospitalizations, chronic disease management, health equity. The binding constraint is not that documentation is slow or that billing cycles are inefficient. Those are symptoms of a system optimized around encounters, procedures, and reimbursement events rather than around longitudinal health. Automating that system compounds the efficiency of a machine pointed in the wrong direction. Efficiency is not effectiveness. A more efficient fee-for-service workflow still rewards volume. A faster prior authorization still gates care behind administrative friction designed to suppress utilization, which is a lovely euphemism for making you give up on the care you were promised. A better-documented encounter still captures a moment rather than a trajectory. Bolt AI onto everything that exists today and you end up with a system that costs slightly less to run, moves slightly faster, and produces the same outcome gaps it has always produced. In some scenarios the misalignment gets worse, because efficiency gains inside a misaligned system accelerate the misalignment. The car goes faster. The cliff is still where it was.
None of this is an argument that the current AI tools are worthless, and the piece would be dishonest if it pretended otherwise. Physician burnout from documentation is not a minor inconvenience. It drives medical errors, shortens careers, accelerates early retirements, and quietly reduces access in exactly the populations that can least afford to lose it. Ambient scribes measurably cut that load. Clinical decision support grounded in peer-reviewed evidence reduces diagnostic errors inside the encounters that still happen and will continue to happen for the foreseeable future. Billing automation frees margin that small practices need to stay open. These are not trivial wins. They are real clinical and economic improvements delivered to real physicians and real patients today, and the companies building them have earned both revenue and clinician trust in the process. That trust is not a decoration. It is the substrate on which any subsequent architectural play gets built, because healthcare is a trust business before it is a technology business, and nobody gets to the fourth layer without first being welcome in the room where the first three live.
The critique is not that the scribe-CDS-billing triangle is wrong. It is that the triangle is insufficient, and that treating it as the terminal destination of healthcare AI rather than as a bridge to something larger is the actual failure of imagination. Pragmatism is not betrayal. Pragmatism is how the capital and the credibility accumulate that make the next move possible. The problem is that the category has confused the bridge for the destination. It is one thing to build a scribe today because physicians need one and the market is willing to pay. It is another to believe, as most of the category seems to, that owning the triangle is the same as owning the future. Companies that build the triangle without ever turning toward what comes next will end up as the Kodaks of this cycle: profitable, respected, and eventually irrelevant, because they optimized the surface the world was already leaving.
The bridge matters. The bridge is not the crossing.
The reason the crossing matters is that the economic architecture of healthcare was built for a different disease burden, and the misalignment between that architecture and our actual problem runs deeper than any workflow optimization can reach. The workflow is not broken. It is archaic relative to the problem it now faces, which is a different kind of critique entirely. Something broken is malfunctioning and needs to be repaired. Something archaic is functioning exactly as designed but was designed for a world that no longer exists. The healthcare system we have inherited falls squarely in the second category, which is why the frantic effort to repair it through incremental efficiency gains keeps missing the mark. You cannot fix a mismatch. You can only replace the architecture that produced it.
It is worth pausing to name the deeper dynamic, because the shift from fee-for-service to value-based care is not really a healthcare story. It is an anthropological one. We call our species Homo sapiens, wise ape, and the name is fitting precisely because what makes us wise is our capacity to be economical. Sapience and economic intelligence are not competing descriptions. They are the same faculty viewed from two angles. Homo sapiens and Homo economicus are interchangeable terms for a species whose distinguishing feature is the ability to recalibrate its collective arrangements in response to changing material conditions. What we are watching in healthcare is that faculty at work. The population is renegotiating, through a thousand small decisions made across employers, patients, physicians, and policymakers, which economic model it will live under. Fee-for-service priced the event. Value-based care prices the trajectory. The species is doing what the species does, updating the economic architecture around the new dominant problem, and the update is happening as a swarm. No single actor is orchestrating it. The herd is moving, and the movement is visible in the data long before it surfaces in the headlines.
This reverses the usual causal story that technology people tell themselves. Technology does not lead these transitions. It follows them. The economic model shifts first, driven by the collective recalibration of what the population is actually paying for and why, and then innovation is pulled into service to scale solutions for the new burdens that arise. The railroads did not create the industrial economy. The industrial economy created the demand that pulled the railroads into being. Electrification did not invent the modern corporation. The modern corporation needed electrification to function at the scale it wanted. Healthcare AI will follow the same pattern. The scribe-CDS-billing triangle is what innovation produces when it is asked to optimize fee-for-service. Something very different will be produced once innovation is asked to scale the trajectory-priced model the herd is already drifting toward.
The shift epidemiologists have spent a generation documenting is the transition from acute and infectious disease as the dominant drivers of morbidity and mortality to chronic noninfectious disease as the dominant drivers. For most of modern history, the first category ruled. Tuberculosis, influenza, pneumonia, cholera, sepsis, obstetric hemorrhage, trauma from the new machinery of industrial and automotive life. The economic model that emerged to match that burden was the hospital network and the insurance network. Both are risk-pooling mechanisms designed around discrete, high-severity, time-bounded events. You pay in when you are healthy, the pool pays out when an acute event happens, the event resolves or it does not, and the accounting closes. Hospitals aggregated the specialized capacity to handle acute events at scale. Insurers aggregated the financial capacity to smooth the cost across populations. Both were the right answer to the problem in front of them.
Part of what made that answer work was that acute and infectious disease demanded synchronous action. A pneumonia does not wait. A ruptured appendix does not reschedule. A hemorrhaging patient in labor does not tolerate asynchronous coordination across a distributed care team. The clinical reality required everyone relevant to the problem (physicians, nurses, surgeons, anesthesia, imaging, pharmacy, blood bank) to be in the same building at the same time, moving on the same clock, responding to the same event. That is what justified the hospital as a building and the hospital system as an institution. You needed specialized capacity colocated in space and time because the disease did not give you the option of distributing it across either. Insurance networks extended the same logic outward, pooling financial risk across populations who would, on any given day, be mostly healthy. The whole architecture was an answer to a disease burden whose defining feature was the convergence of everything important into a single moment.
The architecture worked so well that it produced the problem we are now trying to solve. Antibiotics, vaccines, trauma surgery, sanitation, seatbelts, prenatal care, workplace safety regulations, all of it worked. Mortality from those categories compressed down to a fraction of its former weight. People who would have died of pneumonia at forty now live into their eighties. People who would have died in childbirth have daughters who die of Alzheimer’s instead. Chronic noninfectious disease is not a separate problem that arrived from nowhere. It is the downstream consequence of solving the previous one. It is emergent in the strict sense. It is evolutive. We solved the problem of dying young, and in doing so we created the problem of living long enough to accumulate decades of cardiometabolic, neurodegenerative, oncologic, and multimorbid decline. Success metastasizes into the next category of failure, and the institutions that delivered the success do not, as a rule, volunteer to retire.
The new burden inverts the logic of the old one. Chronic disease is asynchronous by nature. The relevant events do not happen in one place at one time. They happen in kitchens, bedrooms, workplaces, neighborhoods, pharmacies, gyms that nobody goes to, grocery stores stocked according to the incentives of manufacturers whose optimization function has nothing to do with health. The care that moves the needle on a chronic trajectory is distributed across weeks and months, across multiple clinicians who rarely speak to each other, across a patient and a family and sometimes a community health worker, across a medication regimen that requires daily adherence, across behaviors whose accumulation over years is what determines outcomes. No single building can contain this. No single clock can coordinate it. A care team managing a chronic condition is not a team in the trauma-bay sense. It is a loosely coupled coalition that has to function across time and space rather than within a single event.
Because the problem is asynchronous, it is also profoundly regional. The trajectory of a chronic condition is shaped by where the patient lives in a way acute disease never was to the same degree. Social determinants of health (ZIP code, income, education, food environment, housing stability, transit access, neighborhood walkability, exposure to pollution, proximity to primary care) are not soft factors decorating the edges of the clinical picture. They are the clinical picture for a huge share of chronic disease burden. An acute infection will kill you roughly the same way in Manhattan or Mississippi, modulo access to treatment. A cardiometabolic trajectory is almost entirely shaped by which of those two places you live in, and by which ZIP code within each. The geography is not an external variable. It is internal to the disease itself.
The behaviors that determine trajectory sit almost entirely outside the clinical encounter. Sedentary work, processed food environments, chronic stress, sleep disruption, social isolation. These are the conditions under which obesity compounds, cardiometabolic dysfunction accelerates, and multimorbidity becomes the norm rather than the exception. They are preventable at the individual and community level in ways infectious disease never was, because the vector is not a pathogen moving between bodies but a pattern of living slowly accumulating inside one. Managing that pattern through the acute-care apparatus, which is what care management programs and disease management pilots have been trying to do for two decades with the dutiful mediocrity of a project nobody really believes in, produces the results you would expect from using the wrong tool on the wrong problem with a straight face and a quarterly report. You cannot treat a trajectory with an encounter. You cannot change a behavioral environment through a fifteen-minute office visit three times a year. You cannot pool risk the same way when the risk is not a discrete event but a slow-moving condition every member of the pool eventually develops.
And yet the new burden carries the same population-level reach the old one did. More than half of American adults live with at least one chronic condition. A third live with two or more. The prevalence of obesity, hypertension, type 2 diabetes, mental health disorders, and musculoskeletal decline has grown to the point where the chronic burden now touches essentially every household in the country. This is comparable to what tuberculosis or influenza once were in the sense of being a problem everyone was either affected by or one degree removed from. The architectural scale required is the same. The character of that architecture is entirely different. What the old era demanded in centralized institutional capacity, the new era demands in distributed regional capacity, and the existing apparatus cannot reshape itself into the answer.
Part of how the new capacity is becoming visible is through a reorganization of who actually delivers primary care in this country, and it is worth giving credit where credit has historically been withheld. Nurses have always been the connective tissue of care. They are the ones holding the patient’s context across shifts, catching the detail that gets lost in the handoff, reading the room when the patient is not saying what they mean, and translating clinical plans into something a human being can actually execute on. The system has relied on this work for generations without naming it, and the conversation about healthcare tends to mention physicians without mentioning the people who make the care physicians order actually happen. That underrecognition is its own kind of archaism. Nurses have always done more than the org chart admitted, and the chronic era is where that reality becomes impossible to ignore, because this is exactly the kind of longitudinal, relational work they have always been particularly good at.
Physicians remain central to all of this. The work of diagnosing, managing complexity, carrying clinical responsibility for the patient’s trajectory, and making the hard calls at the edges of evidence belongs to them and cannot be outsourced. But physicians are also carrying a liability profile that has grown heavier decade by decade, and the combination of litigation exposure, reimbursement structure, and the procedural orientation of fee-for-service has steadily pulled them toward specialty medicine. This is not a failure of physicians. It is a rational response to their incentive environment. One consequence of that drift is a gap in the longitudinal primary care layer, and that gap has been filled by a wonderful group of providers willing to throw punches for patients and fight the good fight. Nurse practitioners and physician assistants now deliver a growing share of primary care, and in the settings where they are given the room to practice at the top of their training, they deliver it very well. Beyond the licensed medical professions, an entire layer of integrative and alternative providers (naturopaths, chiropractors, health coaches, functional medicine practitioners, acupuncturists, nutritionists) has expanded alongside them. The clinical community has its opinions about the evidence base for each of these modalities, and some of those opinions are well founded. But the population-level fact stands. Patients are seeking out these providers in enormous numbers, and they are doing it mostly on a cash-pay basis, because most of these modalities sit outside the boundaries of traditional insurance coverage. The framing is not physicians versus nurses versus integrative providers. It is a team-based reality in which each role is load-bearing and the system has been slow to recognize the full cast.
The cash-pay dynamic is not a coincidence. It is the same story the primary care reorganization tells, surfacing in a different form. Ge Bai and her colleagues at Johns Hopkins have been documenting something that should be a bigger scandal than it is. For a growing list of services, paying cash is cheaper than paying through insurance. Not in some exotic edge case. Routinely, across imaging, generic medications, common procedures, and a widening slice of primary care. The cash price at a surgery center is often a fraction of the negotiated rate that flows through the insurance complex, and in many cases it is a fraction of what the patient’s own deductible would have forced them to pay anyway. The system designed to reduce financial burden now increases it for a meaningful share of transactions. That is a machine telling you something important about itself, if you are willing to listen past the advertising. The administrative overhead of insurance-mediated care has grown large enough that routing around it is cheaper than participating in it. Direct primary care memberships expand. Transparent-price surgery centers grow double digits. Generic drug marketplaces have redrawn the pricing floor for large categories of medications. Telehealth companies built on direct-to-consumer cash flows reach scale without ever touching a payer contract. Each is a small defection from the managed-care architecture, and each one confirms the same diagnosis. When patients can see prices and pay directly, the price drops and the friction disappears. When they cannot, a fifteen-dollar medication becomes a sixty-dollar medication after the prior authorization, the pharmacy benefit manager spread, the utilization management review, and the administrative layering that each extracts rent while claiming to add value. The patient pays more so a longer chain of intermediaries can take their cut of a simpler transaction, which is the kind of arrangement you would call extortion in any other industry and which we call “coverage” in this one.
What patients are paying cash for, in most of these interactions, is not a specific clinical intervention. It is time, attention, continuity, and the sense that someone is actually listening to them as a whole person rather than as a billing code. It is human care. It sits adjacent to the clinical domain without being reducible to it, and it is exactly the kind of longitudinal, trust-based presence chronic disease management requires. The rise of these providers is a market signal that the population recognizes the gap and is willing to pay out of pocket to fill it, even when the insurance apparatus refuses to acknowledge that the gap exists. The cash-pay market is building the pricing infrastructure for the world after fee-for-service, and it is doing it faster than the institutional reform conversation has noticed, partly because the institutional reform conversation is being held at conferences sponsored by the intermediaries.
What the chronic era needs is the economic analog of what hospital networks and insurance networks were for the acute era. A new structural model built around the dominant disease pattern of our time. In the old era, the model socialized the cost of rare catastrophic events across healthy populations and colocated specialized capacity inside centralized institutions. In the new era, the model has to reduce the cost of risk by changing the trajectory of risk itself, which means shifting the locus of action from the hospital encounter to daily life, from a single clinician to a coordinated team, from reactive treatment to proactive behavior. This is not a technology problem. It is an economic and institutional problem. Technology is a lever inside it, not a substitute for it, which is worth repeating because the current investment thesis for healthcare AI is essentially “what if we called the lever the answer.”
The economic theory that fits this moment is closer to narrative economics and behavioral economics than to the actuarial tradition that built modern insurance. Robert Shiller’s framing of narrative economics, which holds that economic outcomes are driven by the stories populations tell themselves and act on, maps directly onto chronic disease. Chronic disease is a behavioral and narrative phenomenon at population scale. People do not develop type 2 diabetes because they lack information. They develop it because the stories, incentives, environments, and social structures around them reward the behaviors that produce it and do not reward the behaviors that would prevent it. Changing a trajectory requires intervening at the level of narrative and nudge, at the level of the choice architecture of daily life, which means meeting patients where they actually live and giving them the resources, feedback loops, and incentives to act as agents of their own health over years and decades. This is what Thaler and Sunstein articulated more than a decade ago and what the healthcare system has never meaningfully operationalized, because the economic model has never rewarded anyone for doing so and nobody gets a Nobel for implementing someone else’s.
All of this points toward the same conclusion. The patient has to become the structural center of the new economic model, not as a rhetorical flourish but as a necessity. Patients will only become proactive if the resources required for proactivity are actually placed in their hands. Their own longitudinal data. An AI-native agent capable of interpreting and acting on that data. Financial incentives aligned with trajectory rather than event. Pricing they can actually see. A care architecture that rewards prevention rather than intervention. You cannot ask patients to be agents of their own well-being while keeping the data, the decision-making power, the pricing, and the economic upside locked inside provider and payer institutions. The asymmetry is the problem. Value-based care, ICHRA expansion, and the rise of cash-pay are early structural signals that the system is trying to rebalance this asymmetry, but the rebalancing is partial and uneven. It transfers risk to the patient faster than it transfers the tools required to manage that risk, which is why the transition feels punishing rather than empowering. It is the healthcare equivalent of handing someone the steering wheel after the car has started sliding, and then billing them for the driving lesson.
Consider who is doing the administrative work and who captures its economic output. Physicians grind through documentation and coding to justify reimbursement that primarily accrues to payers and health systems. The patient, who is the actual buyer in any meaningful market definition, is structurally excluded from the workflow that determines what their care costs and how it is delivered. This made sense when hospitals aggregated enough patient volume to negotiate as the dominant counterparty to payers. It no longer does. The Mount Sinai and Anthem Blue Cross Blue Shield dispute shows what the breakdown looks like in practice. The fight peaked between January and March of 2026, with physicians going out of network, the hospital system nearly following, more than four hundred fifty million dollars in disputed claims, and the two sides recently resolving the standoff. Mount Sinai tried the old form of leverage (walk away from a payer that refused its terms) and discovered the leverage had already moved. Anthem responded by advertising directly to physicians and members, effectively decoupling network status from hospital affiliation with a pitch that amounted to stay in network even if your hospital isn’t. Patients were caught between two institutions each assuming the other needed them more, and neither did. The dispute resolved because the patient flow both sides were fighting over had become the scarce resource, and neither party controlled it cleanly anymore. The hospital’s historical leverage, which was its network of patients, is no longer locked to the hospital. The leverage now has to follow the patient, because the patient is the one who decides where care happens.
ICHRA and HRA expansion accelerates this shift at a scale most healthtech conversations fail to register. Employer adoption among firms over fifty employees grew more than thirty-four percent in 2025, reaching fifty-two percent in some mid-market segments. Overall ICHRA enrollment tripled into 2026, and the trajectory since 2020 exceeds a thousand percent. Employers adopt these vehicles because they convert a volatile line item into a fixed-cost model. Employees end up with plan choice, reimbursement management, and cost-quality tradeoffs sitting in their hands. The administrative and decision-making burden that used to live inside HR and benefits teams is offloaded onto the individual. It is the same transfer of responsibility value-based care imposes on the clinical side, now happening in parallel on the purchasing side. Patients are becoming the actual customer on both axes simultaneously. Unions figured out this logic decades ago. The party writing the check, or coordinating the collective check, has the pricing power. ICHRAs generalize that insight to the individual level, with all the promise and all the peril that implies.
Every structural force in the system is converging on the same conclusion. The patient is becoming the principal, not the subject, of the care economy.
It is worth being honest about where this vision breaks down, because the vision has three failure modes that deserve to be named rather than hand-waved. The first is cognitive burden. Chronic illness depletes energy, executive function, time, and money, and the worst thing you can do to a depleted person is hand them another job. An AI agent reduces administrative friction, which is real, but it does not eliminate the fact that care orchestration requires attention, and attention is precisely what chronic illness consumes. Any patient-side architecture that works has to be built around the assumption that the patient is exhausted, not heroic. The agent has to carry the weight. The patient has to be able to disengage for weeks at a time without the system punishing them for it, because that is what chronic illness actually looks like. Architectures that assume sustained, optimized patient engagement will fail the people they are supposed to serve, which is the population too tired to fight the architecture.
The second failure mode is the incumbent moat. The existing model is archaic relative to the disease burden but extremely profitable relative to its stakeholders, and those two facts do not have to agree. Hospital systems, payers, and pharmacy benefit managers extract billions of dollars from intermediation, and they will fight the interoperability, data liberation, and pricing transparency that make a patient-side agent functional. They will fight it through lobbying, through selective API implementation, through the strategic deployment of “privacy concerns” whenever a competitor wants access and the strategic absence of those concerns whenever they want access themselves. The 21st Century Cures Act and ONC information-blocking rules exist because this fight was already anticipated, and it is still being fought in every implementation detail. Assuming the incumbents will cooperate with their own disintermediation is the kind of assumption that sinks strategies.
The third failure mode is inequity. A patient-driven, cash-pay, AI-navigated system could produce dramatic improvements for affluent and digitally fluent populations while leaving everyone else stranded in the decaying remnants of the old one. This is not hypothetical. It is the default outcome of every consumer-driven reform in American healthcare so far. The digital divide is real. Health literacy varies by orders of magnitude. The populations with the highest chronic burden are often the populations with the lowest capacity to navigate a self-directed system, which means a patient-centric architecture that is not deliberately designed for equity will become another mechanism for sorting the population into those who can afford good health and those who cannot. The design answer is not to avoid building the patient layer. It is to build it with community health workers, cash-pay affordability, multilingual and low-literacy interfaces, and public-option pricing as first-class features rather than afterthoughts. A patient agent that only works for the patients who least need it is not a solution. It is a tax on inequality with a cleaner UI.
These failure modes do not invalidate the argument. They specify it. The patient-centered model has to be built with exhaustion, incumbent resistance, and inequity as design constraints, not footnotes. Any version that ignores them will produce exactly the disaster its critics predict, and the critics will be right.
This reframes the strategic question entirely. The scribe-CDS-billing triangle competes for the room where care is delivered, but the room is a diminishing share of where care outcomes and care economics are actually determined. What is missing, and what nobody in the current race is fully owning, is a patient-centric orchestration layer. Not another MyChart clone. Not another patient portal bolted onto a provider-side system. An AI-native personal health agent that aggregates longitudinal data under genuine patient control, with FHIR R4 and CCDA as the baseline plumbing extended by wearables, patient-reported outcomes, and claims data. An agent that offloads administrative work from the provider onto the patient’s side of the equation, generating summaries for the physician rather than by the physician, handling prior authorization agentically, automating follow-up and adherence. An agent that surfaces cash-pay pricing alongside insurance-mediated pricing so patients can see the true cost of their options in real time. An agent that enables the proactive and longitudinal loop value-based care actually requires, with predictive risk nudges, shared decision-making, and closed-loop learning from patient action to outcome to model improvement. An agent that gives patients real leverage to dictate care pathways, compare costs across networks, and carry their full history when they switch providers or plans. An agent designed for people who are tired, for a market that will resist it, and for a population whose inequities it is obligated not to amplify.
The technical and regulatory blockers are real but tractable. Interoperability and data silos remain the hardest problem. EHRs still do not talk cleanly to each other, and most patient access APIs are read-only or unusable in practice despite the 21st Century Cures Act, ONC APIs, and TEFCA providing the formal plumbing. Consent and trust frameworks lag the technology. Patients have legitimate concerns about data misuse. The digital divide concentrates benefits among populations that need them least. Provider workflow incentives still reward encounter volume and simple quality metrics rather than the transfer of agency to patients. But the 2026 value-based care model generation, including ACO REACH, ACCESS, and the expanding ICHRA ecosystem, explicitly pays for longitudinal and patient-centered outcomes. The first builder to close the loop captures a reimbursement tailwind rather than fighting one.
The patient layer matters because it is the only structural change that rewires the incentive loop from encounters to longitudinal outcomes. Without that rewiring, AI in healthcare is just faster paperwork inside a model built for a disease burden we no longer have. The scribe-CDS-billing triangle gets you into the room. The layer built around patient data ownership, patient-side orchestration, behavioral nudges, transparent pricing, and a care architecture that treats the patient as principal rather than subject is where the defensibility lives and where the new economy actually takes shape.
What the full architecture looks like in practice is a stack, sequenced in a way that respects how these transitions actually happen. Software first, because software is the only thing that can operate the unified patient knowledge graph and the real-time orchestration chronic care requires. Clinics next, because a software layer without a delivery layer is a dashboard, and dashboards do not change trajectories. An employer and financing wrapper after that, because aligning the payment mechanism with the delivery mechanism is what closes the loop on predictable cost. AI across the entire stack, because the stack is what generates the clean, structured, longitudinal data AI needs to reason at its full potential.
The software layer is the operating system for modern care. Its job is to standardize intake, clean and map data, route patients intelligently, document decisions and actions, close follow-up loops automatically, and maintain a unified patient knowledge graph that persists across visits, providers, and specialties. Every lab, every note, every prescription, every procedure, every outcome feeds into a single structured living map of the patient’s health. The system continuously updates that map and reasons over it to surface what matters most. Inside the same stack sits the reasoning and evidence layer, the clinical intelligence engine that grounds the knowledge graph’s routing and recommendations in the peer-reviewed literature. That layer is also the wedge, in the strict sense of the word, because it is the entry point that builds trust with clinicians in the short term while the longer-term architecture gets built underneath it. It participates in the same category the rest of the triangle operates in, but with a different endgame. It is not trying to optimize the encounter. It is seeding the ground on which the patient-centric orchestration layer will eventually stand.
The clinics come next, because software and evidence without a delivery layer are theoretical. These are interdisciplinary polyclinics bringing primary care, dentistry, behavioral health, pharmacy, diagnostics, and common procedures together under one roof, placed deliberately into underutilized real estate like dead malls and strip malls to create walkable neighborhood health nodes at the scale the population actually needs. A patient with diabetes whose A1c will not come down no longer faces a referral to an external dentist, weeks of scheduling delays, and near-certain dropout. The dentist is in the same building, on the same software, reading the same knowledge graph. The primary team flags the need. The system schedules the procedure. Results feed back instantly. Physical co-location removes friction and the software enforces the connection. Hospitals stay focused on trauma and acute emergencies, which is what they were designed for. Nearly everything else belongs in this distributed middle layer.
The employer wrapper is the final piece of the sequence, and it only makes sense once software and delivery are already working together. At that stage, what the system offers is no longer traditional insurance with its blind spots and adversarial dynamics. It is simple, transparent packaging around a care system that already functions end-to-end. Self-funded employers buy a closed-loop model with known costs, repeatable outcomes, and underwriting grounded in live operational data rather than actuarial assumptions. Care once fragmented across coverage disputes and out-of-network surprises becomes a set of standard interventions the system already knows either prevent or resolve downstream complications. Employers gain predictable spend because gaps are closed proactively. Patients gain better health because the full picture of their needs is visible, connected, and acted upon inside a single integrated system. The AI across the stack becomes indispensable at this point, because it is now being fed clean structured real-world data from live patients and live clinics, and it can reason simultaneously across clinical intent, benefit design, prior authorization logic, financial risk, and longitudinal context in a unified frame.
Each layer reinforces the next. The triangle optimizes the encounter. The stack operates the trajectory. The triangle sells efficiency into the old model. The stack builds the new model from the software layer up. The triangle is a bridge. The stack is the crossing.
Hospital networks and insurance networks did not emerge because someone invented a better bandage. They emerged because the economic structure of the old disease burden required pooled institutions to function, and those institutions then shaped how technology developed around them. The new era requires its own economic structure, and the institutions that will define it do not exist yet in mature form. They will be built around continuous data, patient agency, behavioral nudges, narrative economics, transparent pricing, and risk models that price trajectory rather than event. Whoever builds that structure will define the chronic-care economy the way Blue Cross and the modern hospital system defined the acute one. The technology is downstream of the economic question, not the answer to it.
This is the frame the entire AI-in-healthcare conversation is missing. The platform races, the integration debates, the arguments over who goes deepest into Epic. All of it is a conversation about how to run the existing model more efficiently. None of it addresses the fact that the existing model is itself archaic relative to the disease burden it now has to absorb, and that the burden it now has to absorb is the emergent consequence of the previous model having worked. The real work is not technological. It is economic. The technology is a tool for executing on a new economic model once that model is defined. Used inside the old model, it extends the life of an architecture that should be retired. Used inside a new one, it becomes the mechanism through which the transition actually happens.
Which brings us back to the convergence we started with, and to the failure of imagination it represents. An industry full of brilliant engineers and well-capitalized founders has looked at the largest, most misaligned, most structurally exhausted sector in the economy and concluded, with striking consistency, that automating the paperwork is the whole response. Automating the paperwork is part of the response. It is the part physicians need today, the part the market will pay for today, and the part that builds the trust and the revenue any larger play eventually requires. But it is a bridge, not a crossing. The failure is not in building the bridge. The failure is in mistaking it for the destination. That is great technical sophistication married to insufficient strategic curiosity about what comes after the current wedge is won, which is how industries get disrupted by the companies that bothered to ask what problem they were actually solving. The scribe-CDS-billing triangle is the optimization of the old machinery. The chronic-care economy is the thing being built underneath it, by whoever finally notices that the problem we face now is the one we created by solving the last one, and the next one will be the one we create by solving this.
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*aRquQfgh8LucbDacWBRgPQ.jpeg\")
Luis Cisneros · CEO & Co-Founder, Serelora & EvidenceMD · April 15, 2026
Math is just a way of counting things. Language is just a way of expressing things. And in medicine, the way you express a patient determines how you count them; and how you count them determines whether the algorithm sees a person or a ghost.
Hospital administrators should pay attention to this. More than 124 peer reviewed studies, published in respected journals, cited at conferences, and used to support clinical software and training, may have been built on data representing patients who never existed.
Not anonymized. Not synthetic. Not clearly labeled as such. Just wrong. Patterns inconsistent with human biology. Numbers that don’t behave the way numbers behave when real people are sick.
Gustavo Monnerat, Deputy Editor at The Lancet Americas, traced the problem back to its source and found what anyone familiar with hospital workflow could have predicted. The datasets came from unknown origins, were pooled at scale without provenance, and then fed into machine learning pipelines as if they were trustworthy. The pipelines learned, but they learned the wrong thing.
The question was never whether AI could learn from patient data. The question was always: whose patient data, collected how, structured by whom, and verified against what?
This is not a rare edge case. This is the architecture. And if you want to understand why, you need to understand the difference between two fundamentally opposed philosophies of how clinical data should live in the world.
Epic Cosmos is impressive the way a warehouse is impressive. Over 300 million patients. 19 billion encounters. 2,000 hospitals. 47,000 clinics. When you hear those numbers, the temptation is to treat them the way we treat the GDP… as if scale itself is a proxy for truth.
It isn’t.
Top-down aggregation takes data that was generated inside a clinical workflow—messy, narrative, sometimes incoherent—and pools it into a lake ex post facto. De-identified. Averaged. Decontextualized. The origin story of any individual record is gone. Whether a lab value was entered by a physician who examined the patient or copy-pasted from a prior visit is indistinguishable. Whether a diagnosis code was assigned by a specialist or a billing department working three weeks after discharge is invisible.
For population-level trends, this is fine. Epidemiology doesn’t need every record to be perfect. It needs enough records to be statistically representative. But clinical AI isn’t epidemiology. It’s making decisions about this patient, today, in this room. And for that, the provenance of every data point is everything.
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*-hLAWvAjdXhD8i5J1kkwUg.png\")
Rare diseases make the problem unmistakable. When the signal appears in only one out of every ten thousand patients, aggregation does not make it easier to detect. It pushes it further out of view. Coding inconsistencies, missing context, and duplicated records do not cancel out when events are that rare. They create false patterns. The algorithm still finds a signal, but not one that is real.
This is where the language problem becomes literal.
A clinical note is, by default, a prose document. It narrates. It hedges. It sometimes rambles. It carries context in syntax and tone that no off-the-shelf tokenizer was built to understand. Feed it raw into a language model and you’re feeding it a Victorian novel when it needed a balance sheet.
We built something called MOLG—the Medical Ontology Language Graph—because we believe the way you express a patient’s clinical state fundamentally determines how well a model can reason about it. MOLG is a compact, deterministic string format that rewrites clinical documentation in the native grammar of ontology — every condition linked to SNOMED CT, every medication to RxNorm, every lab to LOINC, every procedure to CPT. No brackets, no quotes, no JSON bloat.
The results from 245 complex clinical vignettes are not incremental. They’re structural.
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*Kz6xroh8Wq1c5KgifGf84A.png\")
This matters for a reason beyond engineering elegance. It is about what gets lost in translation. When a model processes a note, every ambiguous token becomes a potential hallucination, where the model fills in what it expects instead of what is actually documented. MOLG does not compress meaning. It removes the ambiguity that invites fabrication in the first place.
Clean data changes what risk modeling can be. Most clinical risk scores do something quietly dishonest… they stack individual disease burden on top of population-derived mortality baselines without acknowledging that those baselines already include average disease burden. The math double-counts. The risk inflates. Nobody notices because the output looks reasonable and nobody audited the inputs.
Our Clinical Complexity Risk framework (CCR) is built to ask a different question. Not “how sick is this patient” but “how far is this patient deviating from where someone exactly like them, same age, same demographics, same baseline biology, would be expected to be right now?”
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*cvt1w7RFyCNm7pGGF7cfLw.png\")
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*dhIJK2KJ804g_g4nh_uBVw.png\")
This distinction sounds technical until you think about what it means clinically. A 78-year-old with three chronic conditions who is stable and well-managed is not the same risk profile as a 78-year-old with three chronic conditions who just lost 12 pounds and stopped filling her prescriptions. Standard risk scores might give them nearly identical numbers. CCR separates them because the deviation from expected is the actual signal.
Missing data is handled by Bayesian shrinkage rather than silent nullification. Which is another way of saying we admit what we don’t know instead of pretending we do.
Serelora doesn’t sit outside the EHR as an analytics dashboard you open when you have thirty minutes to interpret visualizations. It ingests data the way care actually happens—ambient scribes, lab pulls, imaging, patient messages, claims—and returns structured MOLG records and CCR scores inside the clinical decision loop. The data that trains our models is the same data the clinician used to treat the patient an hour ago.
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*v2HvY0eLes_GMr_E46ZOSA.png\")
This closed loop is not a product feature. It is the only honest response to the problem Monnerat documented. If an AI model trained on questionable data is already shaping care decisions, and with more than 124 papers behind it that is a fair assumption, then the standard for trustworthy clinical AI is not more data. It is data with clear provenance, structured at the source, validated by the workflow that produced it, and explicit about its limits.
Scale without structure is not progress. It is a very large warehouse of uncertainty wearing a confidence interval as a costume.
The way we describe patients determines how we count them. How we count them determines how we filter, approximate, and predict. And how we predict—even imperfectly, even probabilistically—shapes what care they receive. That’s not a technical problem. That’s a moral one. And it starts with being honest about the language.
![\"\"](\"https://cdn-images-1.medium.com/max/543/1*fpCXs9RUL8sE01wAZqMGvw@2x.jpeg\")
There is a moment, and you have felt it, where you type a question into ChatGPT and the answer arrives so cleanly that you do not verify it. You do not even consider verifying it. You read it, you nod, and you move on.
That was not a lapse in judgment. That was your brain doing exactly what four hundred million years of evolution designed it to do. Your brain burns twenty percent of your body’s energy while weighing two percent of its mass. Every thought is a caloric expense. Every shortcut you have ever taken in thought, every gut feeling you trusted instead of reasoning through, every time you skimmed instead of read — that was the most sophisticated energy-conservation system in the known universe working as intended. Being approximately right has always been cheaper than being precisely right, and for most of evolutionary history, cheaper was enough.
Now there is a system that thinks fluently, responds instantly, and costs the body nothing. Zero metabolic overhead. The brain was always going to take that deal. And a new paper out of Wharton just measured what happens when it does.
Steven Shaw and Gideon Nave ran 1,372 people through a series of reasoning problems, gave some of them access to an AI chatbot, and then did something clever and slightly cruel. They rigged the chatbot. On some questions it gave the right answer. On others, it gave the wrong one, confidently, with a neat little rationale attached. Then they watched what happened.
What happened was cognitive surrender.
On trials where the AI was correct, accuracy jumped 25 percentage points above the baseline of people reasoning alone. On trials where the AI was wrong, accuracy dropped 15 points below. Participants followed the AI’s faulty advice roughly four out of five times. Their confidence went up even as their accuracy went down. They didn’t just trust the machine. They stopped thinking altogether and called the machine’s thoughts their own.
Shaw and Nave have a name for what they observed. They call it Tri-System Theory. For half a century, cognitive psychology has run on a two-system model. System 1 is fast, intuitive, automatic. System 2 is slow, deliberative, effortful. Daniel Kahneman built an empire on this distinction. Shaw and Nave argue we now need a third. System 3 is artificial cognition. External, algorithmic, data-driven, and dynamic. It operates outside the brain entirely, in silicon rather than neurons, but its outputs get absorbed into thought and behavior as though they were born there.
The paper is important. But I think it understates its own implications. What Shaw and Nave have identified is not merely a new cognitive pathway. It is the behavioral expression of a thermodynamic principle that has governed biological life since long before anything resembling a human brain existed on this planet.
Most discussions of AI and cognition frame this as a problem of trust, or literacy, or design. They miss the biology entirely. Cognitive biases are not design flaws. They are energy-saving shortcuts that worked well enough, for long enough, in environments simple enough, that evolution kept them. The gazelle that stops to deliberate about whether the rustling in the grass is really a lion does not pass on its genes. The one that runs first and thinks later does.
When Shaw and Nave show that people surrender their reasoning to an AI chatbot, they are not documenting a failure of human rationality. They are documenting its deepest success. The brain found a cheaper source of answers and it took it. That is exactly what four hundred million years of competitive pressure trained it to do.
The problem is that the environment has changed. The heuristic that says “accept the confident answer from the external source” evolved in a world where the external sources were your tribe, your elders, your accumulated cultural wisdom, all of which had skin in the game of your survival. The external source now is an algorithm trained on the internet, optimized for fluency rather than truth, and operated by a company whose survival depends on your engagement, not your accuracy.
Here is where the paper’s findings connect to something much larger than a laboratory demonstration with reasoning puzzles.
Shaw and Nave showed that cognitive surrender is dose-dependent. The more participants used the AI, the more their accuracy tracked the AI’s accuracy rather than their own reasoning. At low usage, people still thought for themselves. At high usage, they became mirrors of the machine. Their performance was no longer a function of their intelligence, their education, or their analytical capacity. It was a function of whether the algorithm happened to be right.
Now extend this finding beyond the laboratory. Think about the platforms that structure daily life. Google does not merely answer your questions. It decides which questions you see answered first, which framings you encounter, which sources you trust. YouTube does not merely show you videos. It constructs a sequential decision funnel in which each choice narrows the next, each click reinforcing a pattern that the algorithm then exploits to keep you clicking. TikTok does not merely entertain you. It runs thousands of micro-experiments on your attention, testing which stimuli produce the longest dwell time, then feeds you more of whatever keeps you still.
This is not nudge theory, though it borrows from it. This is not choice architecture, though it uses its tools. What it is, more precisely, is a system that exploits the same metabolic logic that Shaw and Nave documented in the lab. The brain is looking for the lowest-energy path to resolution. The platform engineers that path. Over time, the user does not merely choose content. The user becomes the kind of person who chooses that content. The algorithm stops reflecting preferences and starts manufacturing them.
Shaw and Nave’s participants followed faulty AI advice 80 percent of the time. Their confidence went up. They felt smarter. The machine was wrong, they adopted the wrong answer, and they walked away more certain than before. That is the real product. Not information. Not answers. The feeling of having thought, without the cost of thinking. The platforms have perfected the manufacture of this feeling. They have built systems that consume human attention while producing the sensation of understanding, when in fact the user has merely been guided through a corridor whose walls were invisible and whose destination was chosen by someone else.
One of the most striking findings in the Wharton paper is who surrenders and who doesn’t. People with higher trust in AI used the chatbot more, followed its advice more uncritically, and suffered greater accuracy declines when the AI was wrong. People with higher fluid intelligence and higher “need for cognition,” a psychological measure of how much someone enjoys effortful thinking, were more resistant. They overrode the AI more often. They maintained their own reasoning even when the machine offered an easier path.
The people most vulnerable to cognitive surrender are not the people we typically worry about when we discuss AI safety. They are not edge cases. They are the median. They are the people who trust technology because they have been told to, who defer to confident-sounding systems because confidence has always been a social signal of competence, who lack the time or the training or the disposition to verify what a machine tells them. They are, in other words, most of us, most of the time.
This is the trap. You can be smart enough to catch the AI’s error and still not bother, because the energetic cost of checking exceeds the perceived benefit of being right. Every interface decision that makes AI output feel more natural, more authoritative, more seamlessly integrated into the flow of thought, is a decision that makes cognitive surrender more likely.
Shaw and Nave ran a third experiment that deserves more attention than it has received. They gave participants financial incentives for correct answers and immediate item-by-item feedback telling them whether they got each question right or wrong. This is the intervention that rational choice theory says should work. Make the stakes real. Make the errors visible. People will adjust.
And they did adjust. Override rates on faulty AI advice more than doubled. Accuracy improved. The combination of money and feedback reactivated System 2, the effortful, deliberative processing that cognitive surrender had suppressed. People started checking the machine’s work.
But here is the finding that matters most. Even with real money on the line and immediate feedback after every single question, cognitive surrender persisted. The accuracy gap between AI-accurate and AI-faulty trials remained at 44 percentage points. Participants who used the AI heavily still had their accuracy determined primarily by whether the AI happened to be right, not by their own reasoning. Incentives and feedback reduced the magnitude of surrender. They did not eliminate it.
Think about what this means outside the lab. In the experiment, the stakes were small and the feedback was instant and unambiguous. In life, the stakes are often enormous and the feedback is delayed, noisy, or absent altogether. You don’t find out your doctor’s AI gave a subtly wrong drug interaction warning until the patient deteriorates. You don’t find out the recommendation engine radicalized your information diet until you realize you can no longer have a conversation with someone who watches different feeds. The conditions that partially mitigate cognitive surrender in the lab barely exist in the wild.
There is a concept in evolutionary biology called the ratchet effect. Once a species loses a capacity, it rarely gets it back. Eyes that atrophy in cave-dwelling fish do not re-evolve when the fish returns to light. Muscles that weaken from disuse do not spontaneously rebuild. The metabolic savings become structural. The organism adapts to the reduced state and builds new dependencies around it.
I think about this when I read about cognitive surrender. Not because I believe AI will cause the human brain to physically atrophy, although the endoscopy study the Wharton paper cites, where physicians who relied on AI diagnostic tools showed measurable declines in unaided performance, suggests the functional equivalent is already happening. I think about it because the ratchet captures the trajectory. Each increment of convenience creates a new baseline. Each new baseline makes the previous level of effort feel unnecessary. Phone numbers, navigation, factual recall — we have been externalizing cognitive function for decades, one reasonable trade at a time, each one making the next one easier to accept. Cognitive surrender is not a single trade. It is the latest step in a sequence whose direction we never explicitly chose.
Shaw and Nave frame their paper as an extension of dual-process theory, a third system added to an existing cognitive architecture. But I think the deeper reading is that System 3 is not merely joining Systems 1 and 2. It is, gradually, making System 2 unnecessary. Not by replacing it with something better but by making it feel too expensive to use. The machine thinks for you. The machine thinks fluently. The machine thinks confidently. Why would you spend the metabolic cost of thinking for yourself when the answer is already there, pre-formed, waiting to be consumed.
The question is not whether AI will reshape human cognition. It already has. But there is a deeper question underneath it, one the paper brushes against without fully asking, and it is the one I cannot stop turning over. What if the brain was never optimizing for truth in the first place?
Shaw and Nave’s participants did not follow the AI because they believed it was correct. They followed it because it sounded correct. Because it was fluent, confident, and internally consistent. Their accuracy dropped but their confidence rose. That only makes sense if the thing being satisfied is not accuracy but something else entirely. Something cheaper to verify. Something the brain has always preferred.
Coherence.
The brain does not ask “is this true?” It asks “does this fit?” And fitting is orders of magnitude cheaper to compute than verifying. A statement that feels logically consistent, aligns with what we already believe, and does not immediately threaten us passes the metabolic audit without triggering the expensive machinery of doubt. Not because we are gullible. Because doubt costs calories and coherence is free.
This reframes cognitive surrender completely. The participants in the Wharton experiments were not failing to think critically. They were doing what humans have always done. They were accepting coherent narratives from confident external sources, the same way we accepted them from elders, from priests, from consensus, from anyone who spoke with enough fluency to make verification feel unnecessary. Consensus, after all, was the original System 3. Before algorithms, the external cognition source was the group. And the group never optimized for accuracy. It optimized for alignment. Because alignment kept you alive, and being right sometimes got you exiled.
AI inherited that role. It just runs it faster, at scale, without the social cost of disagreement. It is consensus with one participant.
Which raises a possibility worth sitting with. What if the period in human history where we valued truth over coherence, the Enlightenment, the scientific method, peer review, the whole apparatus of institutional verification… what if that was the anomaly?
What if cognitive surrender is not a new failure mode introduced by technology but a reversion to the default, the state the brain always preferred before a few centuries of expensive cultural infrastructure made verification feel normal?
If that is the case, then the machines are not degrading human thought. They are simply revealing what it was optimized for all along. And the question stops being “how do we resist cognitive surrender” and becomes something much harder to answer. Something we may not want to answer. Something that implicates not just the technology but the entire Enlightenment bet that verification could ever outcompete comfort at the scale of a species.
We built tools that think for us, and they are extraordinary. We built tools that know what we want to hear, and they are getting better at it every day. We are not getting dumber. We are not getting lazier. We are getting exactly what we selected for.
You’ve been nodding along for a while now. Did you check any of the numbers?
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*248AiV5aTy1fc642PWgdsA.png\")
By Luis Cisneros, CEO of Serelora
Imagine you walk into a hospital. You’ve been alive for 55 years. You’ve seen dozens of doctors. You’ve had blood drawn, prescriptions filled, surgeries scheduled, insurance claims filed, imaging done. All of that information, every note a doctor ever scribbled, every lab result, every claim your insurer processed, lives somewhere in your electronic health record.
Now here’s the part nobody tells you.
That record is roughly the length of Moby Dick.
Not a pamphlet. Not a summary. A novel. Sometimes two novels. The average patient’s longitudinal health record runs between 450,000 and 500,000 tokens. Patients with complex histories at major hospitals? One to two million tokens of raw medical text. And when a hospital tries to point an AI at that record to help your doctor make a decision, the AI chokes. Not gracefully, not partially. It chokes the way anyone would if you handed them two thousand pages and said “read this in ten seconds and tell me what’s wrong with the patient.”
The technology behind ChatGPT and every other AI assistant making headlines has a fundamental constraint that rarely comes up in the breathless coverage. It’s called a context window. Think of it like the AI’s short-term memory. The biggest models today can hold somewhere between 128,000 and two million tokens in that window at once, which sounds like a lot until you realize your medical record is already pushing those limits before the AI has even started thinking.
So the system does what any overwhelmed reader would do. It skims. It summarizes the summary. It cuts corners. It drops the lab result from three years ago that actually explains why your kidneys are struggling today. It loses the temporal thread, the story of you over time, and starts guessing to fill the gaps.
Those guesses have a name in AI research. They’re called hallucinations. And in medicine, a hallucination isn’t a quirky chatbot moment. It’s a misdiagnosis. It’s the wrong drug. It’s a missed cancer screening. This is the context-window crisis, it affects virtually every hospital system experimenting with AI-assisted care right now, and almost nobody is talking about it honestly. The industry keeps reaching for bigger models with longer windows, as if the solution to drowning in text is a bigger swimming pool.
But what if the problem isn’t the size of the pool? What if it’s that we’re swimming in the wrong ocean entirely?
That’s the question that changed everything for us at Serelora. And it’s so simple it almost sounds naive. Why are we asking AI to read novels when medicine already has its own language?
Think about it. Doctors don’t actually communicate in paragraphs when precision matters. They use codes. When your doctor says you have Type 2 diabetes, somewhere in the system that becomes ICD-10 code E11. Your blood glucose test? That’s a LOINC code. Your metformin prescription? RxNorm. The procedure they billed your insurer for? CPT-4. The clinical description of what’s actually happening in your body? SNOMED CT.
These codes already exist. They’ve been standardized for decades. They’re universal across every hospital, every insurer, every country that practices modern medicine. They’re precise. And perhaps most importantly for the problem we’re solving, they’re tiny.
So the insight behind Serelora was almost stupidly elegant. Instead of feeding the AI your entire Moby Dick-length medical record in plain English, translate the whole thing into clinical codes first. Store it that way. Let the AI think in codes, reason in codes, and only translate back into human language when it needs to talk to your doctor. That 450,000-token record collapses to somewhere between 50,000 and 150,000 tokens. A 70 to 85 percent reduction, not by deleting information, but by saying the same things in a language that was designed to be compact and unambiguous.
It’s the same leap mathematics made centuries ago. You can describe the trajectory of a cannonball in three paragraphs of flowery English, or you can write one equation. The equation isn’t a simplification. It’s a better representation of reality. Clinical codes do the same thing for medicine. And once you accept that premise, the entire architecture of the system follows naturally.
It starts with the mess. PDFs, scanned notes, insurance forms, lab reports, discharge summaries. The unstructured chaos of a real medical file, the kind that lands on a desk and makes a physician sigh. The system ingests all of it through a secure upload pipeline with multimodal extraction, meaning it can read typed text, interpret scanned images, and pull structured data from forms simultaneously, regardless of format.
From that raw chaos, normalization begins. Every medical concept the system extracts gets mapped to its canonical clinical code. A diagnosis becomes an ICD-10 code. A medication becomes an RxNorm identifier. A lab observation becomes a LOINC code. A procedure becomes a CPT-4 code. The clinical concept itself, the actual meaning, gets anchored to SNOMED CT. Custom extensions handle the inevitable edge cases, but the principle is absolute. Every piece of information gets one unambiguous code. No synonyms. No paraphrasing. No ambiguity.
And then something that might sound radical happens. The narratives go away. Once every concept has been coded and mapped, the original prose is no longer the source of truth. The data lives as nodes and edges in a structured knowledge graph, with clinical concepts as nodes connected by typed relationships. Temporal relationships capture what happened when. Causal relationships capture what caused what. Administrative relationships capture what was billed and what was covered.
This is the compression engine. This is how Moby Dick becomes a short story without losing a single plot point. And critically, when the AI later needs to turn a code back into a medical concept, it isn’t guessing. Decompression is deterministic and lossless. One code, one meaning, every single time. There’s no generative step, no probabilistic interpretation, no room for the AI to improvise its way into an error.
But compression alone, even compression this dramatic, doesn’t fully solve the problem. A compressed knowledge graph is still a vast landscape of interconnected clinical information. Reasoning across it all at once would still push the boundaries of what any single AI process can handle reliably. Which is why Serelora doesn’t try to do it all at once. It unleashes the gremlins.
We call them gremlins. Six parallel modules that tear through a patient’s knowledge graph simultaneously, each one devouring a different slice of the record. We finalized this architecture just this week.
One gremlin handles medications, mapping every prescription through RxNorm and flagging interactions. Another tracks labs and vitals over time, using LOINC codes to catch trends a human might miss across years of scattered data. A third processes the qualitative assessments doctors have written in clinical notes. A fourth chews through the insurance landscape, what’s covered, what’s denied, where financial leakage is hiding. A fifth works in ICD-10 and SNOMED CT to map the full diagnostic journey and flag condition trajectories. And a sixth digs into social determinants of health, the non-clinical factors like housing stability, employment, and food access that shape outcomes as powerfully as any prescription.
Each gremlin sees only its slice. None is overwhelmed. None is skimming. And because they’re all running at the same time rather than sequentially, the system processes a patient’s entire clinical history in a fraction of the time it would take a single process to wade through even the compressed version.
But gremlins don’t make decisions. They decompose and retrieve. The reasoning happens one layer up, across four specialized agents.
The Graph RAG Retrieval agent pulls structured information from the knowledge graph in response to queries, ensuring that every piece of data the system reasons about is anchored to its source in the graph rather than hallucinated from general training. The Clinical Reasoning agent handles differential diagnosis, treatment planning, and risk projection, including Gompertz-curve modeling that shows how different interventions might alter a patient’s health trajectory over time. The Administrative agent translates clinical intent into reimbursement logic, maps procedures to billing codes, identifies leakage, and navigates the liability landscape. And the Orchestrator sits above all three, reconciling their outputs, running cross-agent consistency checks, resolving conflicts, and synthesizing everything into one coherent narrative tailored to whatever question was actually asked.
That is the architecture in full. Six gremlins decomposing the record in parallel. Four agents reasoning across different domains. An Orchestrator that brings it all together only when needed, only for the specific context being queried. The complete picture never has to exist in a single context window at any point. The system never drowns because it never tries to drink the whole ocean at once.
That alone would be a meaningful advance. But in medicine, a correct answer that can’t be verified is almost as dangerous as a wrong one. Which brings us to what might be the most important piece of the entire architecture.
Every recommendation Serelora surfaces to a physician comes with dual provenance. Two kinds of proof, visible by default, not hidden in a settings menu.
The first is document provenance. Click on any sentence in the AI’s output and the original source document pops up. Not a summary of the document. Not a paraphrase. The actual document, with the exact relevant passage highlighted. You see where the information came from the same way you’d check a footnote in a research paper, except the footnote takes you directly to the primary source rather than making you hunt for it.
The second is guideline provenance. Every treatment suggestion, every administrative recommendation, every diagnostic consideration appears alongside the specific evidence-based clinical guideline the AI used to reach that conclusion. If the system recommends a particular screening protocol, the AHA or NCCN or CMS citation is right there, not buried in a methodology section somewhere, but right next to the recommendation where the doctor can evaluate it in real time.
This dual layer is what separates a tool doctors will actually trust from one that collects dust in the IT department. It’s also what natively satisfies regulatory requirements, malpractice documentation standards, and reimbursement audit trails, not as compliance features bolted on after the fact, but as consequences of how the system was designed from the ground up.
And because every piece of provenance traces back through the same clinical code substrate, the system can do something else that has historically been one of medicine’s most expensive unsolved problems.
Medicine has always lived in two worlds simultaneously. There’s the clinical world, where doctors think about what’s wrong with a patient and how to fix it. And there’s the administrative world, where insurers think about what’s covered, what’s billable, and what documentation is required to justify a claim. These two worlds describe the same patient in different languages, and the translation gap between them costs the American healthcare system billions of dollars a year in denied claims, administrative overhead, revenue leakage, and liability exposure.
When your entire substrate is built on standardized clinical codes, that translation becomes native for the first time. The same ICD-10 code that describes a patient’s condition maps directly to reimbursement categories. The same CPT-4 code that describes a procedure maps to billing requirements. There’s no “interpretation” step where an AI reads a doctor’s note and hopes it captures the right billing nuance. The code is the nuance. Clinical intent and administrative logic finally share a common language because the language was always there, just never used as the foundation for reasoning itself.
This is also where Serelora’s RAF scoring layer comes in, a retrieval-augmented fine-tuning approach we finished integrating this week that grounds every piece of reasoning in domain-specific, retrieval-verified knowledge rather than general-purpose pattern matching. Alongside it, an advanced scribe capability produces physician-ready notes with inline citations, turning the AI’s analysis into documentation that’s immediately usable rather than a draft that needs another hour of human cleanup.
Together, these capabilities make the treatment planning work in a way that feels qualitatively different from anything else in the market. The Clinical Reasoning agent returns ranked differential diagnosis options with projected risk trajectories, Gompertz-curve modeling that shows how different interventions might alter a patient’s health arc over time. It doesn’t just say “here are three options.” It lays out the evidence for each, shows how each one changes the long-term picture, and the Administrative agent maps the pathway to make each one actually happen. Clinical reasoning and operational execution, unified through the same code substrate, delivered with full provenance on both sides.
Which brings us back to the three problems everyone said couldn’t be solved at the same time.
The technical literature calls it the context-compute-hallucination trilemma, and every serious effort in healthcare AI has treated these as tradeoffs to be managed rather than problems to be eliminated.
The context problem is that your medical record is too big for AI to hold in working memory. Clinical code compression reduces the raw token count by 70 to 85 percent, and the gremlin architecture means each parallel module only needs a fraction of what remains. A gremlin handling medications never sees the insurance documents. The full picture only gets assembled by the Orchestrator at synthesis time, and only for the specific question being asked.
The compute problem is that processing millions of tokens is expensive, because compute scales with the square of input length in traditional transformer architectures. Cutting tokens linearly cuts costs dramatically, and running six gremlins in parallel on modern orchestration frameworks approaches linear speedup. Less data, processed simultaneously, means radically cheaper inference at every scale.
The hallucination problem is that AI fabricates information when it’s uncertain. But when an AI can only reason about concepts that exist in a verified clinical ontology, when decompression from code to concept is deterministic rather than generative, when the Graph RAG agent ensures every retrieval is anchored to the knowledge graph, and when every output must point back to both a source document and a clinical guideline, the space for fabrication doesn’t shrink. It collapses. Published research shows 50 to 80 percent reductions in hallucination rates with knowledge-graph grounding alone. Serelora bakes that grounding into the substrate itself, into the very language the AI speaks.
These three problems aren’t actually a trilemma. They’re three symptoms of one underlying mistake, which is trying to make AI think in a language that was never designed for machine reasoning. Fix the language, and all three resolve together.
We are in a strange moment in healthcare AI. The technology is powerful enough to help. The data exists to make it work. The clinical coding standards that make all of this possible have been around for decades. And yet most hospital AI deployments are still trying to brute-force their way through medical records like a college student cramming a textbook the night before an exam. Reading everything, remembering almost nothing, and making things up when the details get fuzzy.
The answer was never bigger models with longer context windows. It was a better language. A language medicine already invented but never taught its machines to speak.
Anchored clinical codes as the native vocabulary for AI reasoning. A knowledge graph that compresses Moby Dick into a short story that loses nothing. Six gremlins tearing through the record in parallel so no single process ever drowns. Four agents reasoning across clinical, administrative, and retrieval domains. An Orchestrator that reconciles it all into one coherent answer. Dual provenance that lets every doctor verify every answer against both the source material and the clinical evidence, right there on screen. RAF scoring that locks reasoning to retrieval-verified knowledge. And a translation layer that finally bridges the gap between what a clinician means and what an administrator needs, because both sides were always speaking the same language underneath.
The architecture is live. The gremlins are running. The agents are reasoning. The provenance clicks through to real documents and real guidelines. And for the first time, the AI isn’t trying to read your medical record at all.
It’s speaking it.
![\"\"](\"https://cdn-images-1.medium.com/max/914/1*VERz5viXHVFHkWDCEM0XoA.png\")
By Luis Cisneros, CEO | Serelora
Something strange is happening to clinical notes across American medicine, and almost nobody is talking about it honestly.
Run five charts from five different clinicians through the same AI scribe and you will get five notes that read like they were written by the same cautious, mid-level committee member who has never once touched a patient. The phrasing is immaculate. The tone is beige. The reasoning, if you can still call it that, has been sanded down to a smooth regulatory finish that says everything a payer wants to hear and almost nothing another clinician actually needs to know.
This is not a glitch. It is the product working exactly as designed. And that is the problem.
The entire AI scribe industry has organized itself around one question: how do we make the note more accurate? Meanwhile, the question that actually matters has gone almost entirely unasked: whose voice is this, and why does it no longer belong to the person who was in the room?
A clinician who says, “Your back pain is just compensation, it’s secondary, we’ve got to sort your knee out and the pain will disappear,” is not being imprecise. That sentence is doing extraordinary work. It names a biomechanical cause-effect chain. It assigns clinical priority. It expresses a level of confidence that communicates urgency to the patient and, just as importantly, to every downstream clinician who reads the note. The word “just” is doing more labor than an entire paragraph of hedged prose. The phrase “will disappear” is a clinical commitment, a decision marker that says: I am confident enough to stake my reasoning on this sequencing.
Now watch what the scribe does with it.
“Lumbar symptoms are considered secondary to primary knee pathology. Addressing the knee dysfunction is expected to resolve back pain.”
Same facts. Entirely different document. The causal chain has been replaced by passive attribution. The confidence has been swapped for bureaucratic hedging. The clinical priority, which was the whole point, has been flattened into two equally weighted clauses that tell the next reader almost nothing about what this clinician actually thinks, how certain they are, or what they would do first if forced to choose.
That is not documentation. That is a translation error with a malpractice tail.
And it is happening millions of times a day, in every specialty, in every setting where an AI scribe sits between the clinician’s mouth and the medical record. The note stops being an artifact of clinical thought and becomes an artifact of linguistic processing. The author changes. The voice disappears. What remains is compliant, defensible, and dangerously hollow.
Recent research published in JAMIA confirms what clinicians have been feeling in their bones — the notes don’t sound like them. And that is not a cosmetic complaint. When the voice changes, the signal changes. When the signal changes, the next clinician reading that note is working with degraded information dressed up in professional formatting. It looks like a note. It parses like a note. But the thing that made it medically useful, the reasoning, the stance, the hierarchy of concern, has been quietly removed and replaced with something that reads more like a compliance document than a clinical one.
To understand why this keeps happening, you have to understand that healthcare runs on two completely different languages that were never meant to occupy the same sentence.
Clinical language is built for decision-making under uncertainty. It is direct, often conversational, full of shorthand and implicit priority markers. When a physician writes “classic presentation” or “looks septic” or “this is the knee talking,” they are compressing enormous amounts of clinical reasoning into phrases that other experienced clinicians can instantly decompress. The messiness is not a flaw. The messiness is the fidelity. Real clinical reasoning is nonlinear, probabilistic, and full of confident bets made under incomplete information. The language reflects that, and it should.
Administrative language serves an entirely different master. It is built for justification after the fact, optimized for billing rules, audit defense, prior authorization logic, and utilization review. It is structured, coded, and deliberately cautious because its audience is not another clinician but a payer, a regulator, or a plaintiff’s attorney. This language is not wrong. It is simply answering a different question. Clinical language asks what do I think is happening and what should we do? Administrative language asks can we prove this was necessary and get paid for it?
Two languages. Two purposes. Two entirely different audiences.
Now here is where it gets ugly, and where it gets personal.
I grew up between Spanish and English. I speak both. I live in both. And like anyone who has ever navigated two languages in the same household, the same neighborhood, the same sentence, I know Spanglish intimately. Not from a textbook. From the kitchen table.
Spanglish is what happens when two languages collide in the same environment and neither one fully wins. It borrows Spanish grammar with English nouns, English sentence structure with Spanish slang, mixed verb conjugations, half-translated idioms that land perfectly if you grew up inside the culture and land nowhere if you did not. It is vivid, economical, and deeply expressive within its native context. Among people who share the hybrid, Spanglish is a superpower. It says things neither language can say alone.
But here is the thing about Spanglish that nobody romanticizes — it only works when everyone in the room shares the code.
A pure English speaker hears Spanglish and catches fragments. They follow the nouns, maybe the verbs, but the grammar feels wrong and the context is missing. They get the shape of the meaning but not the weight. A pure Spanish speaker hears it and winces, because the phrasing is unnatural, the words don’t carry their proper emotional gravity, and it sounds like someone is breaking the language on purpose. Both groups understand pieces. Neither group fully trusts the output. Spanglish becomes this strange middle zone where the words are familiar enough to follow, foreign enough to confuse, and inconsistent enough to feel unreliable.
That is beautiful when you are ordering cafecito from your tía at the counter. It is catastrophic when you are communicating a differential diagnosis across a care team that never met each other.
AI scribes are producing clinical Spanglish. They take the clinician’s raw reasoning and inject billing vocabulary into the narrative of the visit. They wrap compliance language around clinical judgment. They thread administrative phrasing through the clinician’s causal reasoning like English nouns crammed into Spanish grammar. What comes out is not an optimized note. It is a third dialect, a hybrid that is neither a real clinical note nor a real administrative artifact, and it is being generated at industrial scale by systems that have no idea they are doing it.
And now only one group can interpret it fluently — the tiny population of people who live in both worlds at once. The clinic manager who used to be a biller. The physician who taught herself coding. The administrator who once worked bedside. Everyone else is reading fragments and filling in the gaps with assumptions.
The clinician reads the note and thinks: this doesn’t sound like me. Where is the decision? Where is the priority? Why does it read like a lawyer drafted it? The administrator reads the same note and thinks: this is still vague. It doesn’t clearly justify medical necessity. It doesn’t map cleanly to the rules I need it to map to. Both parties catch pieces. Neither party fully trusts what they are reading. And nobody says anything, because the note looks professional and the AI produced it quickly and questioning the machine feels like questioning progress itself.
No. That is the worst possible outcome. A note that loses clinical signal and still fails administrative clarity. A document that satisfies nobody and risks everyone.
Spanglish is powerful socially because it creates intimacy among people who share the hybrid. But it is dangerous for precision because it produces a third language that only a small subset of speakers can truly interpret. Healthcare cannot afford a documentation language that only hybrid speakers understand. In medicine, misunderstanding is not merely annoying. It breaks care. It breaks reimbursement. It breaks trust. And unlike a fumbled cafecito order, nobody laughs it off at the end.
This is the scribe dilemma in its purest form… one document is being forced to perform two fundamentally incompatible jobs, and the AI is making it worse by being very, very good at producing text that looks like it solved the problem while quietly inventing a dialect that nobody asked for and nobody can fully rely on.
The consequences of authorship collapse do not stay inside the note. They propagate.
When a clinician’s confidence markers are stripped, the next provider reading that note cannot distinguish between a finding the original clinician was certain about and one they were merely considering. “Expected to resolve” does not carry the same clinical weight as “will disappear.” One is a hedge. The other is a bet. That distinction matters when the next provider is deciding whether to pursue additional workup or trust the prior assessment. Flatten it, and you have introduced diagnostic noise into a system that was already drowning in it.
When causal reasoning is replaced by passive summarization, the clinical logic becomes invisible. A note that says “lumbar symptoms are considered secondary to primary knee pathology” does not tell you why the clinician believes this, what evidence led to that conclusion, or how strong the inferential chain is. It tells you only that a conclusion exists, stripped of its supporting architecture. This is the medical equivalent of showing someone the answer to an equation without the work. It is technically informative and practically useless for the next person who needs to build on that reasoning.
When every note sounds the same, pattern recognition across a patient’s chart degrades. Clinicians reading longitudinal records rely, more than they often realize, on subtle shifts in language, tone, and emphasis to detect when a colleague became more or less worried, when a diagnosis crystallized, when a treatment plan shifted from exploratory to committed. Homogenize the voice and you homogenize the signal. The chart becomes a flat, undifferentiated wall of text where every visit reads like every other visit, regardless of what actually happened.
And then there is liability. The legal fiction of AI-scribed notes is that they represent the clinician’s documentation. But when the phrasing, tone, and reasoning structure have been fundamentally altered by an intermediary, what exactly is the clinician signing? They are attesting to a document they did not write, in a voice they do not recognize, expressing a level of certainty they may not have intended. Plaintiff’s attorneys have not yet fully discovered this seam, but they will. And when they do, the question will not be whether the note was accurate. It will be whether the note faithfully represented the clinician’s actual reasoning, or whether an AI quietly rewrote the standard of care.
The solution is not better prompts, fancier templates, or more sophisticated natural language processing applied to the same flawed architecture. The solution is separation.
Preserve the clinical note as a clinical artifact. Keep the clinician’s voice, reasoning chain, prioritization, and confidence markers intact. The rawness is not noise. It is often the only place in the entire healthcare data ecosystem where clinical intent is truly visible. A note that sounds like the clinician who wrote it is not an unpolished note. It is a high-fidelity note. Those are not the same thing, and the industry’s confusion on this point has been the single largest driver of documentation decay.
Then generate administrative outputs separately, downstream, as distinct artifacts purpose-built for the workflows they serve. Instead of rewriting the clinical note to satisfy billing, prior authorization, utilization review, or claims defense, translate clinical intent into administrative logic as an explicit, auditable step. The billing letter is not the note. The prior auth justification is not the note. The coding summary is not the note. They are derived products, and they should be produced by reasoning layers that understand both the clinical source and the administrative target without corrupting either.
This is the architectural insight that most of the scribe market has missed entirely. You do not fix authorship collapse by making the AI write more like a clinician. You fix it by stopping the AI from rewriting the clinician at all. You let the note stay a note. You let the system produce everything else.
Serelora is built around this separation. The clinical record stays faithful to the clinician and the patient. The reasoning, the confidence, the voice, all of it preserved, because that is where the medical value lives. Administrative outputs are generated downstream through structured reasoning, explicit justifications, and workflow-specific artifacts that match how healthcare actually operates. Not by laundering clinical language into something a payer might accept, but by maintaining two clean signal paths where one corrupted hybrid used to be.
![\"\"](\"https://cdn-images-1.medium.com/max/1024/1*BtHogjYpvLBmDbUnjT1-6g.png\")
The scribe dilemma was never about whether AI could document a visit. It can. The dilemma is that documentation is not transcription, and a note is not a form. A clinical note is a compressed, authored representation of one human being’s reasoning about another human being’s health. The moment you let a machine silently replace that authorship, you have not automated documentation. You have automated its destruction.
Preserve authorship. Translate downstream. Stop blending two languages into something nobody fully trusts and everybody quietly edits on their way out the door.
Whoever figures this out does not just win the scribe market. They fix the record.
![\"\"](\"https://cdn-images-1.medium.com/max/512/1*K36w0LxZH3qPJ2j2wHzPOA@2x.jpeg\")
By Luis Cisneros, CEO | Serelora
My mother used to sit at the clinic computer after a long day and say the same thing. She would be clicking through the electronic health record, searching for something about a patient, toggling between screens and tabs and modules, and she would turn to me and say I just wish I could talk to it. I wish I could ask it to bring up what I need so I can focus on the patient. So I can focus on the procedure. So I can focus on actually doing the work.
My father felt it even more deeply. He was old school. He kept his notes on paper because paper was easier. You could flip through it. You could find what you needed without clicking through endless menus trying to remember where the system hid the information you were looking for. The computer was supposed to make things faster, but for him it made things harder to find. So he wrote things down by hand and organized them the way his mind worked.
But my parents also taught me something else that took years to fully understand. They would say I have never just seen a head. I have never just seen a heart. I have never just seen an arm and I have never just seen a mouth. But I have seen a whole human being before.
That observation haunts me now. We have broken apart dentistry from medicine. We have broken apart mental health from primary care. We have broken apart primary care from specialties. Each fragment has its own system, its own records, its own way of describing the same patient. And in all that fragmentation we lost sight of the fact that there is one human body. One human being.
I think about my parents constantly as we build Serelora. They were both clinicians who spent their careers serving patients, and they both felt the same frustration with the tools that were supposed to help them. The technology that promised efficiency delivered complexity instead. The systems that promised clarity delivered navigation. And the records that were supposed to capture the patient captured only pieces, scattered across silos that never talked to each other.
That experience stayed with me through every job I held in healthcare. I managed medical and dental clinics. I worked as a medical scribe watching doctors toggle between six systems to answer one question. I processed claims and saw how much human time gets consumed by forms that could fill themselves. I built insurance plans and learned how the entire industry optimizes for documentation rather than care. And through all of it, I kept hearing my mother’s voice. Why can’t I just ask it what I need to know?
That question is no longer rhetorical. The technology to answer it finally exists. And healthcare, the industry that needs it most, is about to change in ways that will make the old world of dashboards and logins feel like a strange and painful memory.
Something happens to clinicians over time. They stop noticing how much of their day consists of searching rather than thinking. Open this tab. Set this filter. Export this report. Cross-reference with that other system. Copy the number. Paste it somewhere else. Click. Click. Click. The cognitive load accumulates so gradually that it becomes invisible, like background noise you forget you are hearing until someone finally turns it off.
Consider something as simple as wanting to know how a patient’s blood sugar has trended over five years. In most systems today, that question requires opening a dashboard, finding the right analytics module, setting the date range, selecting the lab type, pulling the trend view, and then still wondering whether those data points are accurate or where they actually came from. The doctor asked one question. The system demanded fifteen actions.
My father was right to prefer paper. At least with paper you knew where things were.
We started building Serelora because we believe that pattern is fundamentally wrong. Clinical software should not feel like navigation. It should feel like understanding. It should feel like asking a question and receiving an answer so you can get back to the work that actually matters.
Something remarkable is happening in AI right now. The infrastructure for connecting systems has become almost magical. Tools that once required months of custom engineering can now talk to each other with minimal configuration. This is why every major technology company wants into healthcare. They see the opportunity and they see the scale.
But there is something they do not see.
Healthcare is not another sector to conquer. When big tech looks at this industry, they see market size and growth curves and another vertical to add to the portfolio. They build general-purpose tools and assume healthcare will adapt. That approach has failed for decades, and AI will not change the fundamental mismatch. You cannot bolt clinical safety onto a consumer product as an afterthought. You cannot treat patient data like any other data. You cannot optimize for engagement when the stakes are human lives.
There is another problem brewing that the industry has not fully confronted. Every day, physicians are discovering that they can build their own AI tools. They write prompts, create workflows, and share solutions for real clinical pain points. On the surface this is wonderful. The people closest to the work should shape the tools.
But when you look closer, two failures emerge almost immediately. Most of these projects remain prototypes that never integrate into electronic health records, identity systems, or audit trails. They exist as islands. And when tools sit outside the clinical environment, patients pay the price through fragmentation. They get pushed into a patchwork of portals and links and one-off apps that do not communicate with each other. The physician experience improves while the patient experience fractures.
Healthcare does not need more tools. It needs infrastructure that makes tools unnecessary. It needs a unifying principle that serves patients rather than individual workflows. It needs a standard.
The user opens a chat and asks for what they need. That is it. No dashboards. No navigation. Just a question and an answer.
That simplicity is not accidental. It is the result of everything the system has already done before anyone ever types a word.
When patients or providers feed in their data, whether intake forms, PDFs, labs, notes, claims, or images, Serelora builds a Patient Knowledge Graph. Every piece of information becomes a verifiable claim tied to its source, its date, and its confidence level. This graph is not a summary. It is a structured representation of everything known about the patient, with every assertion traceable to the document that produced it.
But the graph alone is just the foundation. Once it is built, the system sends it through an intelligence layer that performs the analysis a physician would otherwise have to do manually. It generates risk scores for emergency visits, hospitalization, and mortality. It identifies social determinants of health barriers and financial coverage gaps. It constructs a causal inference table that traces problems back to their root causes rather than simply listing symptoms and diagnoses side by side. And perhaps most valuably, it identifies what is missing. Which labs would increase confidence in these risk assessments. Which imaging reports would clarify the clinical picture. These gaps become recommended actions for the care team, and when those tests get ordered they also become billable events for the practice.
All of these generated insights then get appended onto the original graph as a new version. The system preserves what was known when, what analysis was performed, and what conclusions were drawn at each stage. This versioning matters for medico-legal purposes. It matters for auditability. It matters for the kind of trust that healthcare demands.
By the time someone asks for a report, the thinking has already been done. The AI does not generate new answers in that moment. It compiles what the graph already knows into exactly the format requested, with every statement linked to its source evidence, every clinical term connected to its definition and anchored codes, every recommendation tied to the guidelines and protocols that support it.
This means the user can ask for exactly what they need and nothing more. A full comprehensive report. An analysis of labs. A financial report showing SDOH risks and coverage gaps. A plan comparison showing what insurance would be optimal for this patient’s care needs. A physician who needs a quick clinical snapshot does not have to scroll through ten pages of financial analysis. A care coordinator focused on benefits optimization does not have to wade through diagnostic histories. Each report serves its purpose.
The clinical reports themselves are structured to support real reasoning rather than overwhelm with data. Risks are organized into functional domains with their respective phenotypes. Cardio metabolic domain with cardiovascular manifestations. Pharmacological domain with polypharmacy concerns. This structure matters because it connects symptoms to systems to root causes in a way that mirrors how experienced clinicians actually think.
That structure is also what allows something elegant to emerge. The system does not just recommend what to add to a patient’s care. It surfaces de-prescribing opportunities that most workflows overlook entirely. Consider a patient taking gabapentin for chronic pain who is not getting meaningful relief. The system can trace the causal chain and surface that the underlying issue might be jaw clenching, that a dental appliance could address the root cause, and that reducing the gabapentin could lower the patient’s polypharmacy burden along with their hospitalization and mortality risk. That kind of insight gets buried in traditional systems where records are fragmented across silos. In Serelora it rises to the surface because the graph connects what those silos cannot.
The reports go beyond analysis to include actionable next steps with the proper guidelines and protocols for this type of patient. When the recommended action after a denial is an appeal, the system drafts the appeal letter ready to be reviewed and sent. When the next step is a referral, the system can prepare that too. The goal is to extend what the user can see and do, augmenting both clinical and financial understanding of the patient while generating the actual artifacts that move care forward.
Through all of this, every statement has provenance one click away. No unattributed assertions. No hallucinations slipping through. This is not magic. It is disciplined architecture and intelligent compilation of what the graph already knows.
The key design choice in Serelora is storing claims rather than facts. This distinction matters more than it might initially seem.
A fact is treated as final. A claim is an assertion with evidence, confidence, and the ability to be revised when better data arrives. This architecture enables longitudinal correction without corrupting the record. When new evidence comes in, whether a corrected lab result or a new clinical note or a physician override, the system updates the relevant claims and appends a new version of the graph. The record becomes living. It continuously reassesses risk and barriers and drivers as new information lands while preserving the full history of what was known and concluded at every point along the way.
This solves a problem that has plagued healthcare AI since its inception. When every field in every output traces back to its source evidence and the exact steps that produced it, errors become easier to catch. Denials become easier to appeal. Trust becomes something you can actually verify rather than something you hope for.
The foundation we are building today enables something powerful tomorrow.
The next evolution makes reports fully interactive. A clinician will be able to ask to see A1C trends over five years, and Serelora will generate a graph they can actually explore. Every data point will link back to its source document. The visualization becomes a window into the evidence, not just a picture of numbers.
More importantly, the same interface will let users correct the record when something is wrong. They will be able to highlight an area of the response and say this is incorrect. If the AI generated something that does not match reality, the user uploads a document showing the correct information, and the system revises the underlying claim. That correction document gets saved as part of the permanent audit trail, linked to the claim it corrected, so future queries can trace back not just to the original source but to the correction and the proof behind it.
Instead of one-off reports that drift out of date, every interaction improves the patient’s record. The system learns. The record evolves. Trust compounds.
Once exploration and correction are rock-solid, the chat becomes the place for operational work.
Users will be able to generate prior-auth packets, appeals, referrals, claim forms, FMLA paperwork, disability letters, and DME requests as reviewable artifacts. The system will compile every field from the Patient Knowledge Graph, attach provenance to every value, and present only the exceptions for human review.
This is where the philosophy becomes concrete. Human in the loop should feel like signing, not typing. The user sees what changed since last time. They see the handful of things that actually need attention. They approve or edit or route. No scrolling through pages of payer requirements. No hunting for evidence. No retyping the same demographics and history into yet another form.
Where direct APIs exist, submission will be clean and electronic. Where no API exists, controlled automation can work through portals as a pragmatic bridge, always with explicit permission and auditability and a human approval checkpoint.
The goal is not to remove people from the loop. It is to remove the busywork that keeps them from doing meaningful work.
The end state is something healthcare has never had before. One patient. One source of truth. And every stakeholder able to access what they need in the language they understand.
This is not a system built only for doctors. The Patient Knowledge Graph serves the patient directly, giving them visibility into their own health in terms they can actually understand. It serves the physician who needs clinical depth and diagnostic reasoning. It serves the broker who needs to match coverage to care needs. It serves the payer who needs to understand risk and utilization. It serves the TPA who needs to coordinate benefits across plans. It serves the caregiver who needs to know what medications to watch and what symptoms to report. It serves the clinical biller and coder who needs accurate documentation tied to proper codes.
Each of these users looks at the same patient. Each of them needs different information presented in different ways. And until now, none of them have had access to a comprehensive, connected record that captures the whole human being rather than fragmented pieces scattered across incompatible systems.
The insight that makes this possible is understanding what AI chatbots actually are at their core. They are magnificent translation tools. The underlying data does not change. The source of truth remains constant. But the system can translate that truth into the narrative and lexicon that each user needs. A physician gets clinical terminology and diagnostic frameworks. A patient gets plain language explanations and actionable guidance. A biller gets codes and documentation requirements. A broker gets coverage analysis and plan comparisons. The graph is the same. The intelligence is the same. The translation adapts.
This is how we finally solve the fragmentation my parents observed decades ago. Dentistry and medicine can connect because the graph holds both. Mental health and primary care can connect because the graph holds both. Specialties and generalists can connect because the graph holds both. The system sees what my parents always saw. Not just a head or a heart or an arm or a mouth. A whole human being.
A user asks for information and receives it. They click any point and see the originating evidence. They click any concept and see the definition and codes appropriate to their role. They click any recommendation and see the rationale and supporting citations. And when action is needed, the same interface produces the paperwork and routes it outward with the user staying in control.
One interface. One evidence spine. One place where anyone involved in a patient’s care can retrieve reality, verify it, and act on it.
Healthcare AI has overpromised instant everything for too long. We are taking a different approach. Build a foundation that earns trust through rigor and transparency, then layer capabilities deliberately. Not because we cannot move faster, but because healthcare has been burned too many times by systems that moved fast and broke things that mattered.
My mother was right all those years ago. She just wanted to ask the computer what she needed to know so she could focus on taking care of people. That future is finally here. And for everyone who touches a patient’s care, from the patient themselves to the clinician to the coordinator to the payer to the family member helping them navigate, everything is about to get so much better.
Luis Cisneros is CEO and Co-Founder of Serelora.
![\"\"](\"https://cdn-images-1.medium.com/max/603/1*fiwl3dQmVjODzyzQsurwIw@2x.jpeg\")
By Luis Cisneros, CEO | Serelora
The modern wellness industry wants you to believe that living longer requires turning your life into a full-time optimization project. Cold plunges. Sleep trackers that cost more than rent. Meal prep as performance art. Every morning is a biochemistry experiment. Every evening is data entry. Then a new UK Biobank study drops in eClinicalMedicine and says the minimum viable product for adding a year to your life is almost offensively modest. Five extra minutes of sleep. Two additional minutes of physical activity. Half a serving of vegetables daily.
Five minutes of sleep is one more episode of a show you’ve already seen. Two minutes of movement is walking to the fridge with appropriate dramatic sighs. Half a serving of vegetables is three baby carrots having an existential crisis. This is not the stuff of transformation montages. This is Tuesday with slightly better choices. But the researchers followed 59,000 adults and found something subversive about exactly these kinds of choices. Meaningful change doesn’t operate like venture capital, where you make dramatic bets and most ventures fail. It operates like compound interest. Small, consistent, almost embarrassingly boring.
The study measured three dimensions across the entire cohort. Sleep duration, meaning actual unconscious hours rather than time spent in bed doom-scrolling. Diet quality, assessed using the Mediterranean diet adherence scale because apparently the Greeks figured something out between inventing democracy and financial crisis. Physical activity, ranging from minimal to the kind that makes you slightly out of breath but still capable of conversation. What emerged from tracking these behaviors over time was an elegant finding about how they interact. All three matter. But you don’t need to perfect any single one. The combinations work synergistically, meaning three mediocre improvements beat one heroic transformation.
The optimal behaviors got linked to 9.4 additional years of both lifespan and disease-free living. That’s nearly a decade. But the interesting part is that you don’t need optimization to see returns. Want four extra years of disease-free life? The minimum investment is 24 more minutes of sleep, which is approximately the time you spend scrolling Instagram before bed while feeling vaguely guilty. Add 4 additional minutes of movement. Include modest dietary improvements like one cup of vegetables, one serving of whole grains, two servings of fish weekly. None of these changes individually looks like much. Together they compound into years you didn’t expect to have.
This portfolio approach may be more effective precisely because it’s sustainable. Committing to an hour at the gym feels like sacrifice. Adding two minutes of movement to what you’re already doing? That’s statistical noise. It doesn’t feel like you’re doing anything, which paradoxically makes it easier to actually do. Our resistance to incremental improvement reveals something about how we construct narratives of change. We’re a species addicted to conversion stories. The dramatic before-and-after. The marathon runner. The reformed smoker. The person who gave up sugar and won’t shut up about it at dinner parties. Small, simultaneous improvements across multiple domains don’t make for compelling stories. They just make for longer lives.
Think about health behaviors as temporal investment. You’re literally trading present effort for future existence. Most people approach self-improvement like day traders, making dramatic bets, seeking immediate visible returns, burning out spectacularly. But longevity appears to function more like index fund investing. Boring. Diversified. Consistent. Over long time horizons, remarkably effective. No single day matters much. No single choice is catastrophic. What matters is the aggregate direction over years.
The study found progressively larger benefits as behaviors improved from low to medium to high quality across all three domains. But the curve isn’t linear. Getting from terrible sleep to merely bad sleep has enormous benefits. Getting from good sleep to perfect sleep? Much smaller gains, much higher costs. Have you met people who are serious about sleep optimization? They’ve calculated optimal room temperature to the degree, installed blackout curtains that cost more than most people’s monthly rent, and they’re slightly insufferable at social gatherings. This is classic marginal utility theory applied to human longevity. The first improvements provide the highest returns. Past a certain point, you’re just optimizing optimization.
Which brings us to the obvious methodological caveats nobody wants to hear but everyone needs to understand. This is observational data showing associations, not proof of causation. Maybe people who sleep well, move regularly, and eat vegetables are just fundamentally different humans. Perhaps they also floss, return shopping carts to the corral, and don’t ghost their Tinder matches. The diet data was self-reported, which means it’s subject to all the usual human biases about what we think we eat versus what we actually eat. Everyone’s diet looks remarkably virtuous in retrospect, especially if you don’t count the snacks. The study measured revealed preferences, actual behaviors, rather than testing what happens when you try to change these things. It’s observing people who already have different patterns and seeing how long they live.
Still, the sheer scale of the cohort (59,000 adults) and the magnitude of the associations (up to 9.4 additional years) makes it worth paying attention to, even with these limitations. The pattern is consistent across multiple behavioral combinations. The dose-response relationship holds across different levels of improvement. The effects appear in both lifespan and healthspan, meaning you’re not just living longer but living better during those additional years.
Here’s what troubles me about these findings. The gap between my current trajectory and a significantly longer, healthier life might be bridged by changes so small they barely register as effort. Five minutes more sleep. Two minutes more movement. Some vegetables. This is almost offensive in its simplicity. We’ve constructed elaborate systems around health and longevity. Entire industries. Social movements. Identity categories. And underneath it all might be something as prosaic as going to bed slightly earlier and eating some broccoli.
The marginal revolution will not be optimized, quantified, or posted to Instagram. It will happen in aggregate, across thousands of tiny choices that feel like nothing in the moment but compound into years. We’ve created elaborate rituals of optimization and measurement when the actual path to longer life might just be trying slightly less hard at being perfect and consistently more committed to being slightly better. The study suggests that transformation doesn’t require transformation. It requires showing up marginally less terrible at taking care of yourself, day after unremarkable day, until suddenly you’ve lived a decade longer than you expected.
And really, what could be more human than that? Adding years to our lives through the accumulated weight of vegetables we didn’t particularly want to eat and sleep we kept telling ourselves we could skip. The revolution will be incremental. It starts with a nap and a salad.
Koemel et al. “Minimum combined sleep, physical activity, and nutrition variations associated with lifeSPAN and healthSPAN improvements” eClinicalMedicine, 2026
![\"\"](\"https://cdn-images-1.medium.com/max/594/1*nWKAfN84xYVeQxFTbm8NFA@2x.jpeg\")
By Luis Cisneros, CEO | Serelora
There is something curious about modern healthcare that we have all quietly agreed not to notice. The sicker you become, the harder it gets for your information to travel between the people trying to help you.
This is not a bug in the system. It is the system.
A healthy person can pull up their cholesterol numbers on a smartphone app while waiting for coffee. A patient with end-stage renal disease, navigating a forty-step pathway toward a kidney transplant that could add a decade to their life, cannot get a cardiac clearance faxed to the transplant center before the evaluation window closes. A cancer patient moving between oncology, radiology, and palliative care discovers that each department operates as if the others do not exist. A heart failure patient discharged from the hospital arrives at their follow-up appointment only to learn that the cardiologist has no record of what happened during the admission.
We might call this the Care Transfer Paradox. And by care transfer, I mean something broader than data exchange. I mean the movement of records, decisions, and accountability across people and sites of care. The full handoff that should happen when a patient moves from one context to another. The complexity that generates the need for this coordination is the same complexity that defeats it. The patients who need information to move fastest are precisely the patients for whom it moves slowest. Healthcare has somehow arranged itself such that the most resource-intensive, highest-stakes clinical scenarios are also the ones most likely to be derailed by a fax machine, a missing form, or an authorization that expired while sitting in someone’s inbox.
The temptation, of course, is to treat this as a technical problem. Interoperability standards. Better APIs. Smarter software. If we could just connect the systems, the thinking goes, data would flow and coordination would follow.
This framing is seductive and almost entirely wrong.
The bottleneck is not capability. The pipes exist. FHIR protocols work. Patient-directed APIs have been built. The bottleneck is governance, incentives, and liability posture. American healthcare did not stumble into fragmented information systems the way one stumbles into a poorly lit room. It built them deliberately, over decades, in response to institutional priorities that had nothing to do with patient coordination and everything to do with institutional protection.
Consider HIPAA. The regulation explicitly permits sharing of protected health information for treatment, payment, and healthcare operations without patient authorization. The legal framework allows for coordination. Yet institutions routinely behave as if HIPAA forbids it, because risk management teams have concluded that the safest interpretation is the most restrictive one, and because no one gets fired for refusing to share a record. The compliance officer’s reflex is not “how do we share this safely?” but “what is our liability exposure if we share at all?” The regulation meant to protect the patient has been operationalized, through institutional risk aversion, as the justification for excluding the patient from their own care.
Protection becomes exclusion. This is the first inversion, and it is far from the last.
Let us be concrete for a moment, because abstraction makes it too easy to look away.
Consider kidney disease. A patient begins hemodialysis. Three times per week, four hours per session, they are tethered to a machine that performs the blood filtration their kidneys can no longer manage. They are exhausted. They are often cognitively impaired in ways the medical literature describes with clinical detachment but which feel, to the patient, like trying to think through wet concrete. They are told, somewhere between medication adjustments and dietary lectures, that transplantation is an option, that it could restore near-normal physiology, that survival rates favor transplant over dialysis by every available measure.
What happens next should be straightforward. Referral to a transplant center, evaluation, testing, waitlisting, surgery, recovery, life.
What actually happens is a labyrinth designed by someone who has never had to navigate it while chronically ill. The dialysis center has records of sessions and lab values in one system. The nephrologist has consultation notes in another. The cardiologist who performed the required stress test has results in a third. The primary care physician has vaccination history in a fourth. The patient is told to “coordinate” the transfer of all this information to the transplant center.
Coordinate. As if they were a project manager with a Gantt chart and a functioning email system. As if they were not spending twelve hours per week hooked to a machine, managing dietary restrictions, coping with fatigue that makes grocery shopping feel like summiting Everest, and slowly dying.
The patient becomes the infrastructure. They become the human router through which data packets must travel because the institutions refuse to talk to each other directly. And they are expected to perform this role while their body is actively failing them.
The numbers are stark. Research published in JAMA and related journals has found that only about 28% of dialysis patients aged 18 to 69 are referred for transplant evaluation within one year of starting dialysis.¹ Among young patients with zero comorbidities, the people who would benefit most, roughly 30% are waitlisted within a year, and just over half within five years.² Referral rates vary dramatically by facility, ranging from nearly zero to over 75%, meaning that a patient’s odds of even hearing about transplantation depend heavily on which dialysis chair they happen to sit in.³
These are not patients failing the system. These are patients being failed by a system that asks them to personally compensate for its own fragmentation.
The kidney transplant pathway is not unique. It is simply a particularly vivid illustration of a dynamic that operates everywhere in healthcare. Cancer care requires coordination across surgery, medical oncology, radiation oncology, pathology, radiology, and often palliative care. Each discipline maintains separate records, separate workflows, separate institutional priorities. The patient moves between them like a traveler without a passport, repeatedly explaining their history, repeatedly discovering that information did not travel with them, repeatedly wondering if anyone is actually in charge.
Chronic disease management, where the majority of healthcare spending actually occurs, is almost entirely a coordination problem. The diabetic patient with comorbid heart disease and depression sees an endocrinologist, a cardiologist, and a psychiatrist, none of whom have reliable visibility into what the others are doing. Medication conflicts go unnoticed. Treatment plans contradict each other. The patient is left to reconcile these contradictions on their own, often without the clinical knowledge to do so safely, often while being blamed for “non-compliance” when the reconciliation fails.
The pattern is consistent across every specialty, every condition, every institution. The more complex the patient, the more specialists involved, the more institutions touched, the worse the information flow becomes. Precisely when coordination matters most, the system is least capable of providing it.
Now we must ask the uncomfortable question that polite healthcare discourse prefers to avoid. If coordination is so obviously valuable, if better information flow would so clearly improve outcomes, why does the system resist it?
The answer is structural rather than conspiratorial. No one is twirling a mustache in a boardroom, plotting to harm patients. Something worse is happening. The incentives have been arranged such that institutional interests and patient interests quietly diverge, without anyone needing to make an explicitly harmful decision.
A dialysis facility that successfully guides a patient toward transplantation loses a revenue stream. The transplanted patient no longer needs dialysis. From a pure business logic standpoint, the facility has no financial reason to aggressively promote an intervention that ends the patient relationship. CMS requires facilities to “educate” patients about transplant options, but there are no standardized metrics for what this education should contain, no penalties for facilities where referral rates approach zero, no mechanism by which the stated goal is connected to any actual incentive.
Hospitals that control patient records can slow-walk transfers to competing institutions. EHR vendors that make data export difficult benefit from data lock-in as a competitive advantage. Specialists who maintain separate documentation systems create referral dependencies that generate revenue. None of this requires malice. It requires only that institutions act according to their own interests in an environment where those interests are not aligned with patient coordination.
The harm is ambient. It is built into the furniture of the system. The system that is supposed to treat you benefits, in measurable financial terms, from your continued engagement with it. This is the second inversion, and it explains why solutions that have existed for years remain mysteriously unimplemented.
The anthropologist David Graeber observed that bureaucracy is not merely inefficient. It is a form of structural violence. Not physical violence, but the violence of forcing people to navigate arbitrary systems that do not recognize their needs, that demand documentation they cannot produce, that require forms of legibility that erase the complexity of actual human situations. Healthcare bureaucracy is this violence refined to an art form.
When a transplant center requests records from a dialysis facility, or when an oncologist requests imaging from a hospital across town, the request enters a queue. Someone must locate the records, determine what can be released, redact what cannot, prepare documents for transfer. The transfer mechanism is often still a fax machine, a technology from 1964 that remains canonical in American medicine. The fax persists not because it works well, but because it is the lowest common denominator of legal defensibility. Sending a fax is “safe” in a way that digital transfer feels risky to compliance departments, even when the digital option is more secure. The fax may not arrive. If it arrives, it may be illegible. If it is legible, it may be incomplete. If it is complete, it may sit in an inbox for days while someone attends to more pressing matters.
Meanwhile, the patient waits. The clearance expires. The labs go stale. The evaluation window closes. The tumor grows. The heart weakens. The patient is told to restart the process.
The medical literature describes patients who fall out of care pathways as “non-compliant” or “lost to follow-up,” phrases that place the burden of failure squarely on the person who is sick. A more honest description would be that the system lost them. The infrastructure did not exist to carry them through. They were asked to substitute their own exhausted bodies for missing coordination capacity, and eventually they could not.
The record of your care becomes the obstacle to your care. This is the third inversion, and it is perhaps the cruelest.
There is a certain deep cruelty in asking patients to solve structural problems through individual effort.
“Patient engagement.” “Shared decision-making.” “Health literacy.” The assumption embedded in these phrases is that if we can just educate patients sufficiently, they will navigate the system successfully. This is victim-blaming dressed in the language of empowerment. The patient with end-stage renal disease is not failing to engage. They are engaging with a system structurally designed to resist coordination. The cancer patient is not insufficiently literate. They are trying to read a map that does not exist.
So what would it actually take to resolve the Care Transfer Paradox?
Not better software alone, though better software would help. The fundamental requirement is a reorientation of how we understand health information, followed by governance mechanisms that make the reorientation stick. Currently, health information is treated as institutional property, something to be guarded, controlled, monetized. The patient exists in this framework as a subject about whom data is collected, not as an agent who owns and directs that data. The necessary shift is to reconceive health information as civic infrastructure, shared and patient-directed, with guardrails but without gatekeepers.
Three things would have to change.
First, patient-directed portability with scoped, time-bound consent. Not “release everything forever or release nothing,” but granular delegation. Allow my cardiologist to see my nephrology notes for ninety days. Let this transplant center access my dialysis records until my evaluation is complete. Consent architecture modeled on financial services, where APIs like Plaid manage permissions dynamically, rather than on the all-or-nothing release forms that dominate healthcare.
Second, enforcement mechanisms that change institutional behavior. Voluntary interoperability standards have existed for years and have been voluntarily ignored. Until there are real consequences for data hoarding, whether through reimbursement penalties, regulatory action, or competitive pressure from patients who can actually see which institutions cooperate and which do not, the incentives will continue to favor fragmentation.
Third, workflow-level routing that moves the right information to the right person at the right time. The patient should not have to know which record needs to go where. Intelligent systems, whether AI-powered or simply well-designed, should handle the logistics of care transfer the way logistics companies handle package routing. The patient should experience coordination, not perform it.
None of this is technically difficult. All of it is politically difficult. If the incentives are currently aligned against sharing, as they clearly are, then changing the incentives requires a brute-force intervention. Federal regulation. Changes in reimbursement models that pay for coordination rather than procedures. A willingness to override the institutional risk aversion that currently treats every potential data transfer as a lawsuit waiting to happen.
The patient with kidney disease, or cancer, or heart failure, or any complex chronic condition, should not have to coordinate their own records. The records should follow them automatically, to wherever their care is occurring, with their consent but without their active labor.
The Care Transfer Paradox is not an abstraction. It is a mechanism by which people are harmed every day, in every hospital, in every specialty, in every city.
We have constructed a healthcare system that imposes harm through the refusal to recognize the patient as a coherent whole rather than as a collection of fragmented records distributed across institutions that treat each other as competitors. The sicker the patient becomes, the more fragmented their care. The more fragmented their care, the more valuable their longitudinal data. The more valuable the data, the harder it is to access.
This is the paradox. This is the trap. And the only way out is to stop pretending the problem is technical.
We already have the pipes. What we lack is a system that is rewarded for letting information move.
1.\tPatzer RE, et al. Variation in dialysis facility referral for kidney transplantation among patients with end-stage renal disease in Georgia. JAMA. 2015.
2.\tAxelrod DA, et al. Rates of solid-organ wait-listing, transplantation, and survival among residents of rural and urban areas. JAMA. 2008.
3.\tData on facility-level variation in transplant referral rates from CMS Dialysis Facility Compare and related USRDS analyses.
![\"\"](\"https://cdn-images-1.medium.com/max/380/1*s7nktidVhCeLS3fslA96Ow@2x.jpeg\")
By Luis Cisneros, CEO | Serelora
On January 10, 2024, Cole Schmidtknecht walked into a Walgreens in Appleton, Wisconsin to refill his inhaler. He was 22. He’d had asthma his entire life and managed it the same way for over a decade. A preventative inhaler called Advair Diskus, taken daily. It kept him breathing. It cost him about $66 a month.
That day, the pharmacist told him his medication was no longer covered. The new price was $539.19. No one had warned him. Not his insurance. Not the pharmacy. Not his doctor. Walgreens told him there were no cheaper alternatives, no generics available.
Cole couldn’t afford $539. He had rent to pay. So he left with only a rescue inhaler, the kind you use when an attack is already happening.
Five days later, he had a severe asthma attack. His roommate drove him to the emergency room. Two minutes before they arrived, Cole stopped responding. When the staff reached him, he was unconscious, pulseless, and blue. They tried to resuscitate him. He never woke up. His parents took him off life support on January 21.
He was 22 years old. He had insurance. He had a prescription from his doctor. He died because a pharmacy benefit manager called OptumRx, a subsidiary of UnitedHealth Group, had quietly removed his medication from the formulary. Because Walgreens didn’t offer him a workaround or contact his doctor. Because no one in the system was responsible for making sure Cole Schmidtknecht could actually get the medication that kept him alive.
His father, Bil, put it simply. Cole chose rent over his medicine.
Cole’s family is suing OptumRx and Walgreens. They’re also pushing for legislation that would require insurance companies to give patients 90 days notice before changing a formulary. They want Cole’s death to mean something.
But here’s the question that haunts me. Can a family actually hold an insurance company accountable when a denial leads to death?
In the U.S., the answer is brutal. It depends what kind of plan the patient had.
If the coverage was employer-sponsored, it’s almost certainly governed by ERISA, the Employee Retirement Income Security Act of 1974. ERISA was written to protect pension plans. But courts have stretched it to cover health benefits in ways that now shield insurers from most state-law accountability.
When a denial leads to serious harm under an ERISA plan, families often find their wrongful death, negligence, and bad-faith claims “preempted.” Kicked out of state court. Forced into federal claims with limited remedies. The Supreme Court has reinforced this wall repeatedly.
The practical result is that the lawsuit becomes “pay the benefit you should have covered” rather than “pay damages for the death.” Families feel like there’s no accountability because, functionally, there isn’t.
Cole had insurance through his employer, Kriete Truck Center Green Bay. That likely means ERISA applies. OptumRx has already filed a motion to dismiss, arguing that federal law prohibits the case from being brought in state court. They expressed “deepest sympathies.” They also pointed out that Cole filled a $5 rescue inhaler that day, as if that were the same thing as the preventative medication he needed to stay alive.
The appeals process exists. Under the ACA, many plans must offer internal appeals followed by independent external review. In theory, this creates a path to overturn wrongful denials.
In practice, the process is too slow, too opaque, and too burdensome when the patient is standing at the pharmacy counter being told their medication costs eight times what it did last month. Cole didn’t have 30 days to file an exception. He had five days before he couldn’t breathe.
Here’s what I keep coming back to. The burden of proof is backwards.
Right now, the insurer or PBM can deny first. The patient has to fight uphill through appeals. Coverage continues only if the patient knows which forms to file, whom to call, and how to escalate. Most people don’t. Most people are not health policy experts. Most people, when they’re told at the pharmacy that their medication costs $539, just leave.
What should happen is the opposite. If a doctor prescribes a medication, the default should be coverage. The insurer should only be able to deny after rapid physician-to-physician review. The burden should fall on the payer to prove the medication isn’t necessary, not on the patient to prove it is.
If a doctor is recommending it, it should be illegal for the insurance company to deny it without appealing to a board of physicians. Not the other way around.
That’s not how it works. But that’s how it should work.
People pay premiums every month for one reason. So the plan shares the cost when the medication matters most.
But the economics of denial are simple. Saying no is profitable when the process is slow and causality is hard to trace. If a patient dies six months after a denial, good luck drawing a clear line from the coverage decision to the death certificate. The harm is diffuse. The accountability is untraceable. And the insurer keeps the premium.
Cole Schmidtknecht died five days after his denial. The causality is about as clear as it gets. And his family still has to fight through ERISA preemption arguments to have their case heard.
This is not conspiracy. It’s just how the incentives line up.
The denial that killed Cole is a symptom of something deeper. Healthcare data is siloed and fragmented. A cardiologist doesn’t see what the endocrinologist saw. The PCP doesn’t know what happened in the ER last month. Nobody is connecting the dots across disciplines, which means nobody is catching the gaps in care until it’s too late. The interdisciplinary bridges that complex patients need to stay healthy simply don’t exist in most systems.
And here’s the thing. If people understood their risks better, they’d make better decisions. We saw this with smoking. We saw this with alcohol. When the data is clear and it’s communicated consistently, every time someone sees a doctor, behavior changes. Not for everyone. Some people will always fall through the cracks. But many will change course when they actually understand what’s at stake.
Nobody wants to die. That’s not a policy insight. It’s a human truth. And if the system made risk visible instead of hiding it inside claims data and formulary spreadsheets, more people would act on it.
Imagine a system that already knows whether a prescription is walking into a denial before the patient gets to the pharmacy counter. It checks the formulary, flags step-therapy requirements, identifies prior authorization traps. If there’s a problem, it surfaces a covered alternative or generates the documentation the payer will demand. At the moment of prescribing, not after the patient is stranded.
When a denial happens anyway, the system launches a rescue path immediately. Find the cheapest cash option. Surface discount programs and manufacturer assistance. Identify therapeutic equivalents. Give the patient something they can actually do, not a phone tree to navigate while they’re scared and short of breath.
And when a payer says no, the system makes that decision visible. Timestamps. Names. The clinical criteria used. A clear record of who overruled the prescribing physician and why. Because right now, denials happen in the dark. Nobody is watching. Nobody is keeping score.
If you make denial operationally harder than coverage, the economics change. If you give employers visibility into how their plans actually perform, they start asking different questions of their PBMs. If you create an auditable trail from prescription to outcome, causality stops being blurry.
That’s the system we need. That’s what I built. It’s called Nodesian, and now we’re scaling it.
Cole Schmidtknecht would be 24 now. His parents think about all the things he could have been doing.
He didn’t die because medicine failed him. He died because the system made it easier for a PBM to change a formulary in silence than for a 22-year-old to get the inhaler his doctor prescribed.
The system doesn’t need a nicer speech. It needs a different architecture. One where the patient isn’t the one holding the burden of proof while they’re trying to breathe.
Luis Cisneros is CEO of Nodesian, a healthcare AI company building data infrastructure for clinical and financial care coordination.
![\"\"](\"https://cdn-images-1.medium.com/max/720/1*uUa9niw97XoMnIfvjhhoOg@2x.jpeg\")
By Luis Cisneros, CEO | Serelora
In 1995, Taiwan did something remarkable. After years of careful study, the island nation launched a universal healthcare system that borrowed liberally from models around the world while remaining stubbornly pragmatic about what would actually work for Taiwanese citizens. The architects of the National Health Insurance program studied Germany’s social insurance, Canada’s single-payer framework, Japan’s fee schedules, Switzerland’s managed competition, and even the American Medicare system. They took what worked, discarded what did not, and built something new.
The result has been nothing short of transformative. Within a single year, coverage expanded from roughly 57 percent of the population to 97 percent, eventually reaching 99.9 percent of residents by 2023.¹ Today, nearly every citizen carries a smart card linked to a unified digital database, enabling seamless billing, rapid record-sharing, and dramatically reduced administrative friction.² Patients can see any doctor or visit any hospital without prior authorization, referral requirements, or the byzantine permissioning rituals that define American healthcare. Public satisfaction with the system now exceeds 90 percent.³ Taiwan spends just 6.2 percent of GDP on healthcare compared to America’s 17.3 percent, yet achieves better outcomes on virtually every metric that matters.⁴ Life expectancy is higher at 81 years versus 77.⁵ Infant mortality is lower at 4.3 per 1,000 live births versus 5.4.⁶ Preventable deaths are rarer.
Taiwan’s success is often invoked in American healthcare debates as evidence that single-payer systems can work. But this framing misses the more profound lesson. Taiwan did not succeed because it adopted any particular ideology about healthcare financing. It succeeded because it treated healthcare as a systems engineering problem rather than a political football. The reformers asked a simple question that American policymakers seem incapable of asking themselves. What actually works? Then they built the answer.
Yet as inspiring as Taiwan’s story is, it would be a mistake to imagine that America could simply copy the Taiwanese model and achieve the same results. The United States is not Taiwan. It is not even close. And the reasons why matter enormously for understanding what kind of reform might actually be possible in this country.
The standard objections to applying Taiwan’s lessons to the United States tend to focus on scale and political structure. America is vastly larger, with over three hundred million people spread across fifty states, each with its own regulatory apparatus, political culture, and healthcare markets. The federal system creates veto points that Taiwan’s unitary government simply does not face. These are real obstacles. But they are not the deepest ones.
The more fundamental barrier is what we might call America’s volatility of interpretation around health itself. This is not simply a matter of political disagreement about policy. It goes deeper, to the very semantics of what health means and how care should be sought and delivered.
Consider the diversity of perspectives that exist within American borders. In some communities, preventive care is seen as essential wisdom. In others, it is viewed with suspicion, associated with intrusive government programs or dismissed in favor of folk remedies and family traditions. Mental health can be understood as a legitimate medical concern requiring professional treatment, or it can be framed as a spiritual problem, a moral failing, or simply a sign of weakness. Chronic disease management means one thing to a college-educated professional with employer-sponsored insurance and something quite different to a recent immigrant working working three jobs without a primary care physician.
This is the peculiar genius and curse of American pluralism. We have managed to create a nation where the very definition of “being healthy” varies not just by ZIP code but by the side of the street you happen to live on. It is as though we built a country specifically designed to resist standardization, then wondered why our healthcare system refuses to behave like a system at all.
This diversity is not merely cultural. It is shaped by socioeconomic position, educational attainment, religious belief, historical trauma, and the lived experience of systemic discrimination. Below certain income and education thresholds, trust in medical institutions erodes, and interpretations of health become more volatile, more tied to immediate survival than to long-term wellness. People who have been failed by institutions do not easily trust those same institutions with their bodies.
The United States is arguably the only wealthy nation where the meaning of health varies so dramatically not just state to state but neighborhood to neighborhood. European countries often cited as diverse tend to exhibit regional variation, but they generally share a baseline cultural understanding of healthcare as a public good. America has no such consensus. Its melting pot character, which in many ways represents a great national strength, creates epistemic fragmentation that renders top-down healthcare reform extraordinarily difficult.
Taiwan could standardize its system rapidly because it could assume a relatively shared understanding of what healthcare was for and how citizens would engage with it. America cannot make that assumption. Any reform that ignores this reality is likely to fail, not because the policy is wrong in principle, but because it will be adopted unevenly, resisted unpredictably, and interpreted differently depending on who is receiving care.
If Taiwan illustrates the possibility of intentional healthcare design, Venezuela provides a cautionary tale about how even well-functioning systems can collapse when the political and economic foundations beneath them crumble.
Before the turn of the millennium, Venezuela’s healthcare system was widely regarded as one of the best in Latin America. The country had achieved remarkable public health victories, most notably the near-elimination of malaria under the leadership of Dr. Arnoldo Gabaldón (My father Dr. Gino Cisneros directly working in campaigns created by Dr. Gabaldón deep in the Amazonian rainforests with respects to other neo-tropical diseases), whose national spraying campaign using DDT became a model for disease eradication worldwide.⁷ By 1961, malaria had been wiped out from 68 percent of Venezuelan territory.⁸ The mortality rate from the disease dropped from 164 deaths per 100,000 people in 1936 to effectively zero by 1962.⁹
This was accomplished through a mixed system that combined private delivery with broad insurance coverage. Most Venezuelans had access to care through social security institutes, employer-based coverage, or public programs. Hospitals were well-equipped. Medical training was rigorous. The infrastructure supported both acute care and public health surveillance. Venezuela’s oil wealth during the 1970s funded substantial investments in health facilities, and the country had the highest per capita GDP in Latin America during that decade, along with the highest growth rate and lowest inequality in the region.¹⁰
Then it fell apart. The economic collapse that began in the 1980s and accelerated catastrophically after 1999 hollowed out the medical system from within. Hospital functionality dropped to just 44 percent of capacity by 2015 according to Human Rights Watch assessments.¹¹ Doctors fled the country in droves, with the Venezuelan Medical Federation reporting that 15,000 physicians had abandoned the public health system by mid-2015 alone.¹² Diseases that had been conquered decades earlier came roaring back. By 2019, malaria cases had increased by 1,200 percent compared to the year 2000 according to the World Health Organization’s World Malaria Report.¹³ Venezuela now accounts for 73 percent of all malaria deaths on the continent, despite being a country that once led the world in eradication efforts.¹⁴ Maternal mortality surged 65 percent in a single year. Infant mortality climbed 30 percent. Diphtheria, once thought eliminated, returned.¹⁵
The lesson here is worth pausing over. Venezuela did not fail because it had the wrong ideology about healthcare financing. It failed because the society around its healthcare system disintegrated. You cannot maintain medical infrastructure when hyperinflation makes your currency worthless and your best doctors are now driving Ubers in Miami.
What emerged in the vacuum was what can only be called a shadow system of healthcare, a fragmented, informal network of communal support, cross-border clinics in Colombia, black-market medications, and desperate improvisation. People found ways to access care because they had to, but these solutions bore no resemblance to a functioning medical system.
Venezuela’s collapse demonstrates several important truths. First, healthcare systems do not exist in isolation from political and economic stability. No matter how well-designed, they cannot survive if the society around them disintegrates. Second, the achievements of one generation are not automatically inherited by the next. Gains must be actively maintained, or they will be lost. Third, and perhaps most importantly for American observers, the Venezuelan case shows that it is entirely possible to have private healthcare delivery with broad public access. The two are not inherently opposed. What destroyed Venezuelan medicine was not its hybrid structure but rather the political and economic catastrophe that consumed the country’s institutions wholesale.
While Venezuela was losing its medical infrastructure, India was slowly building its own, and doing so through a mechanism that receives far too little attention in global health discussions. The Indian diaspora, one of the largest in the world at over 35.4 million people living abroad according to the Ministry of External Affairs’ 2023 report, has become a significant driver of healthcare development in their ancestral communities.¹⁶
India faces healthcare challenges that dwarf those of most wealthy nations. Nearly 65 percent of the population lives in rural areas where access to quality care remains severely limited.¹⁷ Only 46 percent of rural Indians live within five kilometers of a health facility.¹⁸ Physician shortages are endemic. A vast proportion of healthcare spending comes from out-of-pocket payments, leaving the poor especially vulnerable to catastrophic medical expenses.
The formal public health system, anchored by Primary Health Centers and Community Health Centers, struggles with chronic understaffing and inadequate resources. Data from the Union Ministry of Health and Family Welfare show that 8 percent of these centers have no doctors at all, 39 percent lack laboratory technicians, and 18 percent operate without pharmacists.¹⁹ The gap between need and capacity is enormous.
Into this breach have stepped the Non-Resident Indians, particularly those who emigrated in the 1980s and 1990s, built successful careers abroad, and now seek to invest in their home communities. Many of these emigrants are themselves physicians or healthcare professionals whose children have followed them into medicine. They are funding polyclinics, diagnostic centers, and small hospitals in second and third-tier cities and in rural villages that the government healthcare system has struggled to reach.
This phenomenon is visible in places like Dharmaj, a village in Gujarat now referred to as the “Model NRI Village” where diaspora contributions have funded solar panels, modern roads, smart classrooms, and robust healthcare facilities.²⁰ Similar patterns are emerging across the country, with NRI investment supplementing public sector efforts and filling gaps that government programs cannot address quickly enough. According to the India Brand Equity Foundation, hospitals attracted $1.5 billion in foreign direct investment in FY23 alone, accounting for 50 percent of all healthcare FDI.²¹
The government has responded with its own ambitious initiatives. The Ayushman Bharat program, launched in 2018, aims to achieve universal health coverage through a network of over 150,000 Health and Wellness Centers and a health insurance scheme covering 500 million vulnerable citizens.²² Early assessments suggest the program has already reduced out-of-pocket spending significantly for enrolled populations, with savings estimated in the hundreds of billions of rupees.²³
The Indian model is neither planned nor coordinated. It is sprawling, distributed, and chaotic. But it is growing. The lesson it offers is that healthcare infrastructure can be bootstrapped from below when the public sector lacks capacity, provided there are motivated actors with resources and emotional connections to communities in need. It also suggests that healthcare does not require cultural uniformity. India is among the most diverse nations on earth, yet its diaspora-driven expansion respects local variation, integrating traditional medicine like Ayurveda alongside modern allopathic care.
If India represents organic, bottom-up healthcare development, China exemplifies what happens when a powerful central government commits to technological transformation in medicine.
China has made artificial intelligence and digital health national priorities, backed by investment projections exceeding 1.4 trillion dollars by 2030 according to McKinsey Global Institute analysis.²⁴ The most dramatic manifestation of this commitment is the emergence of AI-powered medical systems, culminating in the development of “Agent Hospital” by researchers at Tsinghua University. This virtual medical facility operates with 14 AI agents acting as physicians across 21 medical specialties, capable of diagnosing, treating, and managing up to 10,000 virtual patients per day.²⁵
The AI doctors in this system have achieved a 93.06 percent accuracy rate on questions from the United States Medical Licensing Examination, rivaling and in some cases exceeding human physician performance.²⁶ They can simulate empathy, adjust their communication for different patient populations, and continuously improve through reinforcement learning and real-time feedback. The system processes patient histories, laboratory results, and imaging with computational speed that no human physician could match.
Beyond the experimental realm, China is already deploying AI-assisted diagnostics across the country. The DeepSeek AI system, an open-source medical language model, has been embedded in the intranets of more than 260 hospitals spanning 93.5 percent of China’s provinces according to company deployment reports.²⁷ Remote robotic surgeries have been successfully conducted from distances of over 5,000 kilometers at Fudan University’s Eye & ENT Hospital, demonstrating sub-millimeter precision and reducing surgical times by nearly 30 percent.²⁸ By mid-2025, AI hospital deployments had expanded to over 100 sites across major urban centers.²⁹
There is something simultaneously thrilling and terrifying about watching a country decide that the solution to physician shortages is to simply build more physicians out of software. It is the healthcare equivalent of answering “we can’t find enough bus drivers” with “have you considered autonomous vehicles?” The audacity alone is instructive.
China’s approach addresses a fundamental constraint that many nations face, including the United States. There are simply not enough physicians to meet demand, particularly in rural and underserved areas. Urban hospitals in China are chronically overcrowded, while rural populations often lack access to specialists entirely. AI systems offer a potential bridge, augmenting human providers in high-density urban centers while extending diagnostic capabilities to regions where specialist access would otherwise be impossible.
The Chinese model carries its own risks and limitations. Equity concerns persist, with urban-rural disparities remaining substantial despite technological investment. Data from China’s National Bureau of Statistics indicates rural healthcare spending remains at roughly 40 percent of urban levels.³⁰ Questions about data privacy, algorithmic bias, and the appropriate role of human oversight are only beginning to be addressed. And centralized approaches that work in China may not translate easily to more pluralistic political systems.
But the core insight is valid. Healthcare can be leapfrogged through technology when there is sufficient capital, coordination, and political will. Nations that cannot train enough physicians in the near term can potentially compensate through technological augmentation. This matters for the United States, which faces its own physician shortages, particularly in primary care and in rural communities.
Perhaps the most underappreciated laboratory for healthcare innovation is not in wealthy nations at all but in Africa, where resource constraints have forced the development of modular, adaptive care delivery systems that challenge conventional assumptions about what healthcare infrastructure must look like.
Africa faces healthcare challenges of a magnitude that would overwhelm systems designed around Western hospital-centric models. According to WHO Africa Region data, only 1.3 healthcare workers per 1,000 people serve the continent, well below the World Health Organization’s minimum threshold of 4.5.³¹ Physical infrastructure is sparse, with low hospital-to-patient ratios and unreliable electricity in many facilities. International donors have historically focused on vertical disease programs targeting specific conditions like HIV/AIDS, malaria, and tuberculosis, creating fragmented systems that miss the opportunity for integrated primary care.
Yet within these constraints, remarkable innovations have emerged. Mobile health clinics deployed in countries like Malawi, Kenya, and Zimbabwe bring primary care directly to communities that would otherwise have no access. Research published in Public Health Action found that in Malawi, these clinics conducted over 300,000 patient visits, treating malaria, respiratory infections, and gastrointestinal conditions while providing HIV testing and counseling, operating on 99 percent of planned days and effectively reaching populations that fixed facilities cannot serve.³²
Community health workers form the backbone of care delivery across much of the continent. The African Union has committed to deploying two million community health workers across member states, recognizing their role as the critical link between communities and formal healthcare structures according to Africa CDC workshop proceedings.³³ These workers are trained to handle a broad range of services, from health promotion and disease prevention to screening and basic treatment. They integrate into existing community structures, building trust and cultural competence that outside providers would struggle to achieve.
Digital technology has been layered onto these human networks. SMS-based systems in countries like Uganda, documented by GSMA, provide real-time information on medicine inventory at clinics, ensuring that rural patients who travel long distances find their medications in stock.³⁴ Telehealth platforms reach patients who lack internet access through voice and text services. Mobile training applications like Kenya’s M-JALI platform have equipped thousands of community health workers with updated clinical protocols and decision support tools according to Health Policy Watch reporting.³⁵
The African model demonstrates something that wealthy nations often forget. Healthcare does not require massive centralized infrastructure to be effective. Modular systems designed around community networks, mobile delivery, and appropriate technology can achieve remarkable results under severe resource constraints. The key is building trust within communities, adapting to local conditions, and prioritizing access over architectural grandeur.
Richard Bohmer’s research on healthcare delivery in resource-limited settings, published in his book Designing Care, highlights these lessons.³⁶ The organizations that succeed in such environments emphasize coordination over scale, modularity over uniformity, and community integration over institutional authority. These principles have profound implications for the United States, where many communities face healthcare deserts that rival developing world conditions even as neighboring areas enjoy world-class facilities.
If Africa can deliver primary care through solar-powered mobile clinics in villages without electricity, one wonders what exactly prevents the richest nation in history from ensuring that a patient’s medical records can travel the 3.2 miles from their cardiologist’s office to their primary care physician without being faxed in 2025.
El Salvador offers a different kind of lesson, one about rebuilding healthcare systems after catastrophic disruption and about the role that community organization can play in driving reform — a dynamic that has evolved dramatically under recent leadership, where top-down security gains have intersected with persistent economic vulnerabilities to reshape well-being and care delivery.
During the 1970s and into the civil war that erupted in 1980, El Salvador ranked among the five least healthy countries in the world according to research by Ugalde and colleagues published in the British Medical Journal.³⁷ The physician-to-patient ratio was the lowest in Latin America. Eighty percent of rural peasants had no potable water. Sixty percent had no access to health services. When the civil war began, the government slashed the health budget by 50 percent and withdrew services from many regions entirely according to analysis from the University of Washington START Center.³⁸ Life expectancy, already dismal at 56 years, dropped to 50.7 years. Child mortality rates were catastrophic at 87 per 1,000, and malnutrition affected three out of four children as documented in the International Journal for Equity in Health.³⁹
The peace accords that ended the conflict in 1992 created an opening for reconstruction, but meaningful healthcare reform did not arrive until the leftist FMLN party won its first presidential election in 2009. The new government launched a comprehensive health reform built on lessons learned from both the war experience and from studying successful primary care models in Brazil, Paraguay, Cuba, and Spain, as analyzed by political scientist Christopher Clark in Latin American Politics and Society.⁴⁰
The reform sent 481 community healthcare teams to permanent placements throughout the country, prioritizing the poorest municipalities first according to Ministry of Health records.⁴¹ Fees for procedures that had previously been charged, ostensibly as “voluntary” contributions, were eliminated. Access to the public system expanded dramatically. According to World Bank assessment, births in health facilities increased by 20 percent in the poorest municipalities.⁴² Neonatal units were equipped. Infrastructure investments modernized over 51 primary care hospitals and 30 secondary and tertiary hospitals.⁴³ Maternal mortality decreased from 50.8 deaths per 100,000 live births in 2011 to 48 by 2015 according to World Bank Development Indicators.⁴⁴
What made El Salvador’s reform distinctive was the role of community organization. During the civil war, communities in rebel-controlled areas had developed their own health systems when the government withdrew services. They trained lay health workers, created mobile clinics, and established decision-making structures through general assemblies and coordinated sector meetings as documented by medical anthropologist Sandy Smith-Nonini in her book Healing the Body Politic.⁴⁵ These organizational capacities did not disappear after the war. They became the foundation for the National Health Forum, a civil society alliance present in 12 of 14 departments that advocated for comprehensive reform and participated actively in its implementation.⁴⁶
El Salvador demonstrates that healthcare systems can be rebuilt after even severe disruption, but only if there are organized communities capable of both demanding reform and participating in its execution. Trust in institutions, once destroyed, cannot be recreated through policy alone. It must be rebuilt through engagement, participation, and demonstrated responsiveness to community needs. This principle has been tested anew under President Nayib Bukele, who assumed office in 2019 and has pursued a bold, centralized agenda that has transformed the country’s security landscape while exposing tensions in economic sustainability and social equity.
Bukele’s administration has prioritized aggressive anti-gang measures, declaring a state of emergency in 2022 that facilitated mass arrests and dismantled major criminal networks like MS-13 and Barrio 18. Homicides plummeted from 2,398 in 2019 to just 114 in 2024, yielding the lowest murder rate in El Salvador’s history and fostering a profound sense of safety that has rippled into everyday well-being.⁷⁵ Communities once paralyzed by extortion and violence now report calmer streets and renewed economic activity in poor neighborhoods, where residents can access markets, schools, and clinics without fear — a direct boon to preventive care and community health initiatives rooted in the post-war era.⁷⁶ Yet this security dividend has come at a steep human cost: over 85,000 arrests, often without due process, have swelled the prison population to more than 107,000 — nearly 2 percent of Salvadorans — with reports of torture, incommunicado detention, and at least 430 deaths in custody by mid-2025, eroding civil liberties and creating a climate of fear that stifles community advocacy.⁷⁷
Economically, Bukele’s policies have yielded mixed results that underscore the fragility of health gains amid broader well-being challenges. GDP growth slowed from 2.5 percent in 2019 to 2.6 percent in 2024, hampered by fiscal adjustments, natural disasters, and a public debt burden reaching 88.9 percent of GDP.⁷⁸ Poverty rose from 26.8 percent in 2019 to 30.3 percent in 2023, with incomes for the poorest households falling 10 percent, particularly in rural areas, exacerbated by food inflation and declining remittances.⁷⁹ These strains have pressured healthcare affordability, even as the administration modernizes the Integrated National Health System through technological leapfrogs. In November 2025, Bukele launched DoctorSV, an AI-powered app enabling free video consultations, electronic prescriptions, and home delivery of medications and lab tests (including X-rays and ultrasounds) via partnerships with Google Cloud and over 350 pharmacies ; initially targeting ages 18–30 before nationwide rollout.⁸⁰ Complementing this, a $120 million World Bank-funded Improving Health Care in El Salvador Project (PROMAS), approved in March 2025, spans five years to bolster primary care infrastructure, digital transformation, supply chains, and training, with a focus on vulnerable groups like women, children, rural residents, and Indigenous Peoples.⁸¹ A February 2025 budget reform further increased health spending, though a proposed $91 million cut in the overall 2025 allocation highlights ongoing fiscal trade-offs.⁸²
These developments illustrate how Bukele’s top-down security focus has unlocked community potential for health engagement — safer environments enable the participation that fueled earlier reforms — but economic stagnation risks unraveling these advances, as poverty deepens barriers to care and strains public resources. The Salvadoran Medical Association has cautioned that digital tools like DoctorSV, while innovative, cannot substitute for addressing staff shortages, rural inequities, and supply gaps without robust community integration and universal access, including for those without devices or internet.⁸³ For American reformers, El Salvador under Bukele reinforces that well-being hinges on balancing security with economic vitality; without the latter, even the most responsive health systems falter, demanding ongoing community vigilance to hold leaders accountable.
Australia presents yet another variation, a hybrid system that combines universal public insurance with a substantial private sector and that achieves generally good outcomes while generating its own set of contradictions worth examining closely, particularly for Americans who imagine that simply adding a public option would solve everything.
The Australian system rests on Medicare, a universal public insurance program established in 1984 after years of political struggle as documented in Parliament of Australia background briefs.⁴⁷ Medicare provides free public hospital care and subsidizes physician services, pharmaceuticals, and diagnostic imaging. All Australian citizens, permanent residents, and those from countries with reciprocal agreements are eligible. The system is funded through general taxation supplemented by a Medicare levy of 2 percent of taxable income according to Australian Taxation Office guidelines.⁴⁸
Layered on top of Medicare is a robust private insurance sector. Data from the Australian Bureau of Statistics show that approximately 45 to 55 percent of Australians carry private coverage, which provides access to private hospitals, choice of physician, shorter wait times for elective procedures, and coverage for services that Medicare does not fully cover, including dental care and physiotherapy.⁴⁹ The government encourages private insurance through tax rebates for purchasers and tax penalties for higher earners who do not carry coverage.
By design, the public and private systems are meant to work together, with private insurance taking pressure off the busier public system. The structure echoes American employer-sponsored insurance in some ways, but with crucial differences. Government sets prices for drugs, treatments, and other expenses. There are caps on cost-sharing and safety nets for high out-of-pocket costs as outlined by the Commonwealth Fund’s international health system profiles.⁵⁰ The rebate system means the government subsidizes private insurance directly, spending over $6 billion annually according to Canadian physician and healthcare analyst Dr. Robert Bell, creating incentives that do not exist in the American market.⁵¹
Australians can choose whether to be treated as public or private patients, and many exercise both options depending on circumstances. The hybrid structure has produced generally good results. Australians live longer than Americans at 83 years versus 77, are healthier, see their doctors more frequently, and die of preventable diseases less often according to PBS NewsHour’s comparative reporting.⁵² They achieve these outcomes while spending roughly half what the United States spends per person, with more hospital beds per capita at 3.8 per 1,000 versus America’s 2.7, more physicians at 3.5 per 1,000 versus 2.6, and more nurses at 11.5 per 1,000 versus 9.5 based on OECD Health Statistics.⁵³
Yet, the Australian model also reveals the tensions inherent in hybrid systems that Americans would do well to study before assuming hybridity is a magic solution. Private hospitals focus on elective procedures and serve wealthier populations in urban areas, leaving rural and remote communities dependent on an underfunded public system. The Australian Institute of Health and Welfare has documented that rural wait times can run 50 percent longer than urban areas for certain procedures.⁵⁴ Indigenous Australians, who have suffered decades of discrimination, experience health outcomes far worse than the general population, with life expectancies 8 to 11 years shorter on average according to research from the Lowitja Institute.⁵⁵
The fundamental policy incoherence is instructive. Stephen Duckett, Director of the Health Program at Grattan Institute and Honorary Professor at the University of Melbourne, has noted in Healthcare Policy that there has been “no coherence or consistency in the rhetoric about the roles of the two systems,” with policymakers variously suggesting that private care substitutes for public care, complements it, or that Medicare is alternately a universal scheme or a residual one.⁵⁶ Duckett concludes that this ambiguity creates what he calls “an incoherent mess” in which neither system functions optimally and the billions spent subsidizing private insurance could arguably be better invested in strengthening the public system or addressing equity gaps.⁵⁷
The Australian experience suggests that hybrid systems can work but require careful management to prevent the private sector from undermining the public one. Without coherent policy guidance about the relationship between public and private care, the result is a system that achieves decent outcomes overall while perpetuating disparities and leaving its fundamental logic undefined. America, with its gift for taking any policy incoherence and amplifying it tenfold through fifty state legislatures and several hundred insurance companies, might want to take notes.
What lessons can the United States draw from this global tour of healthcare systems? Not, I would argue, the lesson that we should simply adopt any particular model wholesale. Taiwan’s size and cohesion enabled rapid implementation. Venezuela’s collapse warns against political instability more than it advocates any particular healthcare structure. India’s diaspora-driven expansion reflects conditions that cannot be replicated through policy. China’s technological leapfrog requires centralized coordination that American federalism makes difficult. Africa’s modular innovation responds to resource constraints that wealthy nations do not face. El Salvador’s reconstruction built on organizational capacities forged through civil war. Australia’s hybrid system produces tensions that American politics would likely intensify.
The deeper lesson is about information. The American healthcare system does not primarily suffer from wrong ideology or insufficient resources. It suffers from a breakdown in information flow that prevents coordination, creates redundancy, and leaves patients falling through cracks that should not exist.
Consider how the American system evolved. Hospitals were originally designed for acute care, treating infectious diseases and trauma. They centralized patients, and in doing so, they centralized patient information. Legal and regulatory frameworks developed to protect that information, first from abuse and later from liability. HIPAA and related privacy protections were built to safeguard individuals, but in practice they have become institutional shields that lock data within organizational silos.
This made sense when healthcare was primarily acute. Patients came in, were treated, and left. Their information stayed with the institution that treated them. But healthcare has fundamentally changed. Chronic disease now dominates the landscape. According to CDC data from the National Center for Chronic Disease Prevention and Health Promotion, over 60 percent of American adults live with conditions like diabetes, hypertension, heart disease, or other ailments that require ongoing management, not one-time interventions.⁵⁸ These patients see multiple providers across multiple settings over extended periods. Their care requires coordination, which requires information to flow freely among everyone involved in their treatment.
But the information does not flow. It sits trapped in hospital EHRs, insurance claims databases, specialist portals, and paper files in primary care offices. Patients who should have their complete medical history available to every provider they see instead find themselves repeating the same information, undergoing redundant tests, and experiencing delays while their records are faxed, mailed, or manually reentered into new systems. Research by McKinsey & Company found that up to 27 percent of hospital errors stem from data silos, costing healthcare organizations an estimated $20 million annually per organization in rework and delays.⁵⁹ A Health Affairs survey found that more than 60 percent of healthcare executives identify data fragmentation as the primary barrier to effective analytics and care coordination.⁶⁰
The sicker a patient becomes, the more providers they see, and the more their information fragments across disconnected systems. This creates what might be called the care transfer paradox. The patients who need coordination most are the least likely to receive it because their complexity scatters their data across more institutional boundaries.
We have, in effect, built the world’s most expensive healthcare system and then ensured that the left hand cannot know what the right hand is doing. It is the organizational equivalent of hiring a symphony orchestra, giving each musician noise-canceling headphones, and wondering why the concert sounds like chaos.
Taiwan solved this problem by creating a unified digital infrastructure from the beginning. Every citizen carries a smart card linked to a centralized database. Information flows seamlessly because the system was designed to enable flow. The MediCloud system, launched in 2015 by the National Health Insurance Administration, integrates medical information across all contracted providers, allowing access to 12 types of patient records including imaging, treatment history, and vaccination records.⁶¹ The United States built its healthcare system before digital technology existed, layered regulations designed to protect paper records onto digital systems, and created incentives for institutions to hoard rather than share information.
The solution is not to copy Taiwan’s single-payer structure. It is to build the informational rails that would enable coordination regardless of payment mechanism. This means national standards for patient identity, interoperable record systems, and consent architectures that allow patients to direct their information to those who need it. It means unbundling the legitimate privacy protections of HIPAA from the institutional risk-avoidance behavior that currently prevents information sharing. It means treating health data not as property to be guarded but as connective tissue that enables care — a principle that is already taking root in domestic innovations emerging from the very fractures of our fragmented system, where ambulatory and preventive care models are pioneering the fluid data exchanges essential for true coordination.
The insurance-driven fragmentation of American healthcare has long exacerbated disparities between hospital-based acute care and the outpatient, preventive, and procedural services that dominate everyday medical needs, accounting for over 90 percent of encounters, according to analyses from the Agency for Healthcare Research and Quality.⁶² This disconnect has stifled innovation in non-hospital settings, where administrative burdens from prior authorizations, claim denials, and siloed billing deter providers from expanding services. Yet, in response, a wave of large-scale Direct Primary Care (DPC) practices has emerged, leveraging value-based contracting to function as commercially privatized Accountable Care Organizations (ACOs). These entities bypass traditional insurers by contracting directly with employers, aligning incentives around outcomes rather than volume and delivering comprehensive care at predictable costs.⁶³ As noted in a 2025 guide from the Physicians Advocacy Institute, DPC arrangements with large employers often cover total care costs for employee populations, fostering preventive focus and reducing reliance on fee-for-service fragmentation.⁶⁴
What makes these large DPCs particularly noteworthy is their evolution into multi-specialty polyclinics — overlooked relics of early 20th-century healthcare design, now revitalized as scalable, physician-owned hubs for integrated care. Unlike traditional single-provider DPCs, these polyclinics house primary care alongside specialties like dermatology, orthopedics, and mental health, enabling on-site minor procedures, elective surgeries, and interventions coded under HCPCS Level II (e.g., J-codes for injections and non-invasive therapies) without hospital involvement.⁶⁵ They sidestep longstanding corporate practice of medicine (CPOM) prohibitions, which in states like California and Texas bar physicians from owning hospitals but permit clinic ownership, allowing doctor-led entities to capture a substantial share of care delivery.⁶⁶ A 2025 Milbank Memorial Fund report highlights how management services organizations (MSOs) further enable this by handling non-clinical operations, preserving physician control while scaling to serve thousands.⁶⁷ Well-structured polyclinics thus provide more comprehensive outpatient care than many standalone facilities or even hospitals for routine needs, underscoring that most care — chronic management, diagnostics, and low-acuity procedures — occurs ambulatorily, not inpatient.⁶⁸
Exemplars abound. Plum Health DPC in Michigan, a membership-based model charging $49 monthly, has expanded beyond primary care to include in-office minor surgical procedures like suturing and biopsies, alongside discounted labs and imaging, serving urban underserved populations without insurance billing.⁶⁹ Larger networks, such as Everside Health with over 385 centers across 34 states, partner directly with employers for value-based primary and specialty care, incorporating on-site orthopedics and behavioral health to handle elective procedures like joint injections and skin lesion removals.⁷⁰ Similarly, platforms like Hint Health facilitate employer-direct DPC contracts that integrate surgical networks for ambulatory electives, such as cataract or hernia repairs, performed in physician-owned facilities outside hospital billing streams.⁷¹ These models not only reduce costs — often by 20–40 percent through avoided administrative overhead — but also demonstrate polyclinics’ potential to consolidate care under one roof, minimizing the handoffs that perpetuate information silos.
This resurgence aligns seamlessly with the informational reforms needed for chronic care coordination. Large DPCs and polyclinics are at the vanguard of advocating for EHR overhauls, demanding interoperable platforms that enable real-time data sharing across specialties and with external providers. The shift to value-based care, as detailed in a 2024 Medical Economics analysis, necessitates EHR enhancements for population health analytics, automated referrals, and seamless billing flows… capabilities these models are piloting through direct employer integrations.⁷² At the 2025 DPC Summit, discussions centered on EHR transitions to reduce clinician workload by 30 percent via AI-driven documentation and cross-practice data exchange, addressing the “battle of the EHRs” that plagues fragmented systems.⁷³ Startups like Nodesian are collaborating with these institutions to build consent-based information architectures that treat data as fluid connective tissue, facilitating secure flows between polyclinic silos and broader networks without HIPAA’s institutional barriers.⁷⁴ In essence, these innovations are not mere workarounds but proofs of concept for an ambulatory-first ecosystem where information enables, rather than impedes, care, paving the way for the broader adaptive mesh that can weave together America’s diverse healthcare fabric without unraveling its pluralistic threads.
The United States cannot reset its healthcare system the way Taiwan did. The political fragmentation, institutional entrenchment, and epistemic volatility that characterize American society make comprehensive reform extraordinarily difficult. But difficulty is not impossibility.
What becomes possible when we understand the American problem as one of information architecture rather than ideological commitment? We can pursue modular reforms that create shared rails even without shared structures. We can standardize the data layer while allowing diverse approaches to payment, delivery, and benefit design. We can build infrastructure that adapts to the volatility of interpretation that defines American healthcare rather than demanding a uniformity that will never come.
This is the lesson that Taiwan, Venezuela, India, China, Africa, El Salvador, and Australia all offer in different ways. Healthcare systems reflect national character. They succeed when they are designed with clear-eyed understanding of the conditions they must address. They fail when they assume uniformity that does not exist or ignore the foundations upon which they rest.
The United States has built one of the most resource-rich healthcare arsenals in the world and then paralyzed it by fragmenting the information that would allow it to function. We argue endlessly about who should pay while the machinery stays broken regardless of payment source. We debate ideology while patients wait for faxes.
Taiwan did not fund healthcare. It engineered the rails. The United States can do the same, not through a single grand reform but through cumulative changes that create the informational infrastructure coordination requires. Standardize patient identity. Mandate interoperability. Build consent systems that enable sharing without sacrificing privacy. Remove permissioning layers that exist to protect institutions rather than patients.
The future of American healthcare is not a single system. It is an adaptive mesh that accommodates our diversity while enabling the coordination that chronic care demands. Building it requires acknowledging what global experience teaches. The barrier is not resources. It is not ideology. It is the flow of information, and that is something we can fix.
…Or we can just continue faxing medical records between buildings that share a parking lot.
1\tCheng, T.M. (2015). “Reflections on the 20th Anniversary of Taiwan’s Single-Payer National Health Insurance System.” Health Affairs, 34(3), 502–510.
2\tLi, Y.C., et al. (2015). “Building a national electronic medical record exchange system — experiences in Taiwan.” Computer Methods and Programs in Biomedicine, 121(1), 14–20.
3\tNational Health Insurance Administration, Taiwan. (2024). “Public Satisfaction Survey Results 2020–2024.” Ministry of Health and Welfare.
4\tOECD Health Statistics. (2024). “Health Expenditure and Financing.” Organisation for Economic Co-operation and Development.
5\tWorld Bank. (2024). “Life Expectancy at Birth, Total (Years).” World Development Indicators.
6\tWorld Bank. (2024). “Mortality Rate, Infant (Per 1,000 Live Births).” World Development Indicators.
7\tLitsios, S. (1998). “Arnoldo Gabaldón’s Independent Path for Malaria Control and Public Health in the Tropics.” Parassitologia, 40, 231–238.
8\tGabaldón, A., & Berti, A.L. (1954). “The First Large Area in the Tropical Zone to Report Malaria Eradication: North-Central Venezuela.” American Journal of Tropical Medicine and Hygiene, 3, 793–807.
9\tGabaldón, A. (1983). “Malaria Eradication in Venezuela: Doctrine, Practice, and Achievements After Twenty Years.” American Journal of Tropical Medicine and Hygiene, 32, 203–211.
10\tRodriguez, J.A. (2017). “The Roots of Venezuela’s Failing State.” Origins: Current Events in Historical Perspective, Ohio State University.
11\tHuman Rights Watch. (2015). “Venezuela: Health Care in Crisis.” HRW Report, August 2015.
12\tVenezuelan Medical Federation. (2015). “Public Health Workforce Assessment 2015.” Press Release, May 2015.
13\tWorld Health Organization. (2020). “World Malaria Report 2020.” WHO, Geneva.
14\tGrillet, M.E., et al. (2021). “Malaria in Venezuela: Gabaldón’s Legacy Scattered to the Winds.” The Lancet Global Health, 9(5), e584-e585.
15\tMinistry of Health, Venezuela. (2017). “Epidemiological Bulletin 2016.” Subsequently removed from publication; data reported by Associated Press, May 2017.
16\tMinistry of External Affairs, India. (2023). “Population of Overseas Indians.” Annual Report to Parliament.
17\tNational Family Health Survey (NFHS-5). (2021). Ministry of Health and Family Welfare, Government of India.
18\tRural Health Statistics 2021–2022. Ministry of Health and Family Welfare, Government of India.
19\tKaran, A., et al. (2019). “Health Care in Rural India: A Lack Between Need and Feed.” Indian Journal of Community Medicine, 44(1), 23–26.
20\tANI News. (2024). “Gujarat’s Dharmaj Village Redefines Rural Prosperity with Global NRI Support.” December 10, 2024.
21\tIndia Brand Equity Foundation. (2024). “Foreign Investors in Love with Indian Hospitals.” IBEF Healthcare Report.
22\tAyushman Bharat Mission. (2024). “Health and Wellness Centers Progress Report.” National Health Authority, Government of India.
23\tNational Health Authority. (2025). “Ayushman Bharat Impact Assessment 2024–2025.” Ministry of Health and Family Welfare.
24\tMcKinsey Global Institute. (2024). “China’s AI Investment Outlook Through 2030.” McKinsey & Company.
25\tLiu, Y., et al. (2024). “Agent Hospital: A Virtual Hospital Powered by Large Language Models.” Tsinghua University Institute for AI Industry Research.
26\tGlobal Times. (2024). “China’s First AI Hospital Town Debuts.” May 2024.
27\tDeepSeek AI. (2025). “Deployment Report: Medical LLM Implementation Across Chinese Healthcare.” Company Technical Report.
28\tFudan University Eye & ENT Hospital. (2024). “First Remote Non-Invasive AI-Powered Surgery Completion Report.” Hospital Press Release.
29\tMedTech World. (2025). “China’s AI Hospital Expansion: Mid-Year Assessment.” Industry Report.
30\tNational Bureau of Statistics, China. (2024). “Urban-Rural Healthcare Expenditure Gap Analysis.” Statistical Yearbook.
31\tWHO Africa Region. (2024). “Health Workforce Status Report.” World Health Organization Regional Office for Africa.
32\tBemah, P., et al. (2015). “Bringing Care to the Community: Expanding Access to Health Care in Rural Malawi Through Mobile Health Clinics.” Public Health Action, 5(2), 110–116.
33\tAfrica CDC. (2024). “Africa’s Pursuit of Cost-Effective Community Health Worker Services.” Workshop Report, Accra, Ghana, August 2024.
34\tGSMA. (2024). “Mobile Technology for Health: Case Studies from Sub-Saharan Africa.” Mobile for Development Report.
35\tHealth Policy Watch. (2024). “Powering Africa’s Health Future: Innovation and Infrastructure in Primary Care.” November 2024.
36\tBohmer, R.M.J. (2009). Designing Care: Aligning the Nature and Management of Health Care. Harvard Business Press.
37\tUgalde, A., et al. (2000). “The Health Costs of War: Can They Be Measured? Lessons from El Salvador.” British Medical Journal, 321, 1369–1371.
38\tUniversity of Washington START Center. (2021). “Territorial Community Teams in El Salvador.” Policy Analysis Report.
39\tHeidari, S., et al. (2020). “The Role of Social Movements in Strengthening Health Systems: The Experience of the National Health Forum in El Salvador.” International Journal for Equity in Health, 19, Article 49.
40\tClark, C. (2015). “The New Left and Health Care Reform in El Salvador.” Latin American Politics and Society, 57(4), 117–138.
41\tMinisterio de Salud de El Salvador. (2014). “Reforma de Salud: Informe de Avances 2009–2014.” Government Report.
42\tWorld Bank. (2019). “Renovating the Public Health Care System in El Salvador.” Results Brief, April 25, 2019.
43\tWorld Bank. (2019). “El Salvador Health System Strengthening Project: Implementation Completion Report.”
44\tWorld Bank Development Indicators. (2016). “Maternal Mortality Ratio — El Salvador.” Data Catalog.
45\tSmith-Nonini, S. (2010). Healing the Body Politic: El Salvador’s Popular Struggle for Health Rights from Civil War to Neoliberal Peace. Rutgers University Press.
46\tNational Health Forum El Salvador. (2020). “Community Participation in Health System Governance.” Organizational Report, available at fnssv.com.
47\tBiggs, A. (2016). “Medicare — Background Brief.” Parliament of Australia, Parliamentary Library.
48\tAustralian Taxation Office. (2024). “Medicare Levy and Medicare Levy Surcharge.” Tax Guidelines.
49\tAustralian Bureau of Statistics. (2024). “Private Health Insurance Coverage.” National Health Survey.
50\tCommonwealth Fund. (2024). “International Health Care System Profiles: Australia.” Country Profile.
51\tBell, R. (2019). “Does Hybrid Health Care Improve Public Health Services? Lessons Learned from Australia.” drbobbell.com, Healthcare Policy Analysis.
52\tPBS NewsHour. (2020). “What the U.S. Can Learn from Australia’s Hybrid Health Care System.” Broadcast Report, January 2020.
53\tOECD Health Statistics. (2024). “Health Care Resources: Hospital Beds and Health Workers.” Organisation for Economic Co-operation and Development.
54\tAustralian Institute of Health and Welfare. (2024). “Rural and Remote Health Access Report.” AIHW, Canberra.
55\tLowitja Institute. (2024). “Close the Gap Campaign: Indigenous Health Outcomes Report.” Melbourne.
56\tDuckett, S., & Nemet, K. (2019). “Commentary: The Consequences of Private Involvement in Healthcare — The Australian Experience.” Healthcare Policy, 15(SP), 7–12.
57\tDuckett, S. (2019). “Public-Private Mix in Australian Healthcare.” Grattan Institute Analysis.
58\tCDC. (2024). “Chronic Diseases in America.” National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention.
59\tSahni, N.R., et al. (2021). “Administrative Simplification: How to Save a Quarter-Trillion Dollars in U.S. Healthcare.” McKinsey & Company, October 2021.
60\tHealth Affairs. (2023). “Healthcare Executive Survey on Data Integration Barriers.” Health Affairs Research Brief.
61\tNational Health Insurance Administration, Taiwan. (2024). “MediCloud System: Technical Overview and Utilization Statistics.” Ministry of Health and Welfare.
62\tAgency for Healthcare Research and Quality. (2024). “Healthcare Cost and Utilization Project: National Inpatient Sample Overview.” AHRQ Statistical Brief.
63\tFTI Consulting. (2022). “Direct-to-Employer Contracting Has Arrived.” Insights Article.
64\tPhysicians Advocacy Institute. (2025). “Guide to Value-Based Contracting.” PAI Report.
65\tPlum Health DPC. (2024). “Direct Primary Care Services Overview.” Company Website.
66\tRooke-Ley, H., Reddy, M., Mehta, N., & Sewak, Y. (2025). “The Corporate Backdoor to Medicine: How MSOs Are Reshaping Physician Practices.” Milbank Memorial Fund Policy Brief.
67\tRooke-Ley, H., Reddy, M., Mehta, N., & Sewak, Y. (2025). “The Corporate Backdoor to Medicine: How MSOs Are Reshaping Physician Practices.” Milbank Memorial Fund Policy Brief.
68\tAmerican Medical Association. (2024). “Ambulatory Care Trends in U.S. Healthcare Delivery.” AMA Research.
69\tDetroit Free Press. (2019). “These Detroit Doctors Help People with No Insurance, High Deductibles.” August 6, 2019.
70\tMarathon Health. (2023). “Everside Health Acquires Direct Primary Care Provider R-Health.” Press Release.
71\tHint Health. (2024). “Hint Connect: A Curated Network for DPC Practices.” Company Website.
72\tMedical Economics. (2024). “Shift to Value-Based Care Brings New Desires for EHR Capabilities.” February 6, 2024.
73\tMy DPC Story Podcast. (2025). “Battle of the EHRs at the 2025 DPC Summit.” YouTube Episode, August 3, 2025.
74\tNodesian. (2025). “Advancing Interoperable Flows in DPC Networks.” Company White Paper.
75\tForeign Affairs. (2025). “Does the Bukele Model Have a Future?” Foreign Affairs.
76\tIbid.
77\tHuman Rights Watch. (2025). “World Report 2025: El Salvador.” Human Rights Watch.
78\tPeoples Dispatch. (2025). “One more year of Bukele: tough on crime, struggling with poverty.” June 5, 2025.
79\tIbid.
80\tUPI. (2025). “El Salvador launches unprecedented primary health care access system.” November 18, 2025.
81\tWorld Bank. (2025). “World Bank to Strengthen Health Care in El Salvador.” Press Release, March 19, 2025.
82\tP4H Network. (2025). “Budget reform increases resources for health in El Salvador.” May 5, 2025.
83\tUPI. (2025). “El Salvador launches unprecedented primary health care access system.” November 18, 2025.