Health IT Perspectives

We built a $5 trillion healthcare system around billing codes — not patient outcomes. That is not an opinion. That is history. And understanding exactly how it happened is the foundation of every meaningful health IT product being built today.

The Interview

I recently sat down with Dr. Yin Ho — physician, health IT entrepreneur, and author of Rushing Headlong: Health IT's Legacy and the Road to Responsible AI — to ask the questions every health IT founder needs answered. This is the first in a series of eight conversations covering the original failures of health IT, the AI opportunity that may finally change everything, and the hard lessons that only come from building inside this world for decades.

In this first conversation, Dr. Ho traces the single most consequential decision that locked the entire US digital health infrastructure into a billing-first architecture. Her diagnosis is both historical and urgent — because you cannot fix a system you do not understand.

What the Industry Got Wrong

The story of health IT is not a technology story. It is an incentives story. The systems that were built — the EHRs, the clearinghouses, the clinical documentation tools — were designed primarily to process payments. Not to understand patients. Not to generate knowledge. Not to support clinical decisions. To process billing transactions.

This was not an accident. It was the rational outcome of a set of economic incentives that rewarded transaction processing and penalized everything else. The HITECH Act of 2009 accelerated EHR adoption — but it accelerated adoption of a billing-first architecture that was already broken. It digitized the wrong thing.

The Physician as the Missing Link

The insight from Dr. Ho that surprised me most in this conversation was not about technology. It was about people.

The physician is the most trusted voice in healthcare. Patients make decisions based on what their physician tells them. They enroll in trials their physician recommends. They trust the treatments their physician prescribes. And yet — the physician is simultaneously the most excluded actor in the systems built around them.

The systems were not designed to support clinical judgment. They were designed to capture billable activity. The physician became a data entry operator in a billing system — spending more time documenting transactions than caring for patients. This is not a technology failure. It is a design failure. And it has consequences that ripple through every corner of health IT.

What This Means for Founders

If you are building in health IT, understanding this history is not optional. The incumbent systems you are navigating — the EHRs, the payer networks, the claims infrastructure — were all designed around billing-first logic. The data they produce reflects that logic. The gaps in that data reflect that logic. The resistance you encounter when trying to access or use that data reflects that logic.

You are not fighting a technology problem. You are navigating an incentives problem that has been decades in the making. The founders who succeed in health IT are the ones who understand this distinction — and build accordingly.

About This Series

This is Episode 1 of 8 from my conversation with Dr. Yin Ho. Over the next four weeks I will be publishing one post every few days — each one a focused insight for founders, builders, and operators working in health IT, pharma, and digital health.

Rushing Headlong is available on Amazon. It has a foreword by J.D. Kleinke — one of the founders of Truven Health Analytics — and an endorsement from Stéphane Bancel, CEO of Moderna. It is one of the most honest books written about health IT in a generation.

Every year, healthcare generates billions of data points. Almost none of it is clinically useful. This is not a storage problem or a technology problem. It is a structural problem that generative AI may finally be capable of solving.

A Founder's Frustration

I have spent months trying to build reliable real world evidence using claims data. And I kept hitting the same wall. The data was there — billions of data points. But the context was missing. The patient journey was missing. What looked like a complete dataset was actually a skeleton — transactions without story, codes without meaning, records without the human being at the center of them.

I was not alone in this frustration. In my conversation with Dr. Yin Ho, she named this problem with a precision I had not heard before — and explained why generative AI may finally offer a path through it.

The Claims Data Problem

Claims data is the most abundant data generated by the US electronic health system. It is also the most structurally limited for clinical purposes. Claims data was designed to process payments — to document what services were rendered, to whom, and at what cost. It was never designed to capture the clinical reality of what happened to a patient.

Turning claims data into clinically useful real world evidence has historically required something expensive, slow, and deeply human: trained clinical personnel doing what Dr. Ho calls detective work — reading through unstructured physician notes, piecing together a patient's longitudinal journey line by line, filling in the gaps that billing codes leave behind.

Turning clinical data into meaningful research datasets has been challenging and has required expensive, manual abstraction by clinically trained personnel.

Where the Intelligence Actually Lives

The richest clinical intelligence in the US healthcare system is not in the structured data fields of an EHR. It is in the unstructured narrative sections — the physician notes, the clinical observations, the treatment rationale, the patient reported symptoms that never make it into a billing code.

This is where the detective work happens. A physician documents not just what they did but why. They note the patient's hesitation about a treatment, the comorbidity that complicated the decision, the observation that did not fit the expected pattern. This information is irreplaceable. And it has been almost entirely inaccessible to researchers and analysts because no system could read it at scale. Until now.

What Generative AI Changes

Generative AI can be applied to augment and accelerate this abstraction process — reading unstructured clinical notes at scale, extracting the clinical context buried inside them, and building the accurate longitudinal datasets that real world evidence analysis demands.

This is not AI replacing clinical judgment. The human abstractor — the trained clinical professional doing the detective work — remains essential for quality validation and contextual interpretation. But AI can dramatically reduce the time, cost, and scale constraints that have made clinical abstraction prohibitively expensive for all but the largest pharma companies.

What This Means for Founders

If you are building in RWE, clinical intelligence, or health data — the abstraction problem is your opportunity. The companies that crack affordable, scalable, AI-assisted clinical abstraction will unlock a dataset that pharma, researchers, and payers have been trying to access for decades.

The technology is finally ready. The question is whether you can build the trust, the physician relationships, and the quality validation processes that make AI-abstracted clinical data credible enough to act on.

Seventeen years. That is how long it takes — on average — for a clinical discovery to reach a patient. A researcher identifies something that could save lives. And the average patient waits seventeen years to benefit from it. Understanding why this happens — and where it can be shortened — is the most important strategic question in health IT today.

Not One Wall — Many Small Ones

Most people assume the seventeen year gap is about clinical trials. The phases, the approvals, the regulatory pathway. But in my conversation with Dr. Yin Ho, she reframed it completely.

It is about a feedback loop. The ability for us to get information about what is happening is not very fast.

The seventeen year gap is not one bottleneck. It is dozens of small processes — each taking a few years — stacked on top of each other in a linear pathway. Discovery. Development. Phases I, II, III. Regulatory approval. Market launch. Post-market observation. Adoption. Every step moves slowly because the data feeding each step is slow, incomplete, or locked away where nobody can access it.

The Microprocess Opportunity

The path to closing the gap is not a single breakthrough. It is additive — shrinking multiple microprocesses simultaneously. Identifying patients faster. Documenting care more completely. Feeding information back about what is actually working in the real world — without waiting years to run another specific trial.

Dr. Ho frames it this way: whatever gets observed in clinical practice, how fast does that information come back to the researchers asking the question? And how well is it documented such that analysis becomes possible without designing a new study from scratch? Those two questions — speed of feedback and quality of documentation — are where the seventeen years hide.

The Data Layer Insight

But the insight from this conversation that stayed with me longest was not about feedback loops or microprocesses. It was about architecture.

For decades, clinical care and clinical research have been treated as two separate worlds. Different applications. Different workflows. Different teams. The EHR on one side. The clinical trial data capture system on the other. And we have been thinking about them at the application level — as if the interface is the problem.

There is a data layer that exists under both that actually could be used for more than one purpose.

The same data that documents a physician caring for a patient can fuel a clinical trial. The same patient journey that informs a treatment decision can generate real world evidence. The infrastructure already exists. We have simply been thinking at the wrong level — at the application level rather than the data level.

What This Means for Founders

If you are building at the intersection of clinical care and clinical research, the data layer is your strategic foundation. The founders who win in this space will not be the ones who build better EHR interfaces or better trial management tools. They will be the ones who build intelligence on top of the shared data layer that already exists — making it possible for the same data to serve multiple purposes simultaneously.

That is how you close the seventeen year gap. Not with a single breakthrough. With a fundamentally different way of thinking about data.

Less than 5% of eligible patients ever enroll in a clinical trial. Not because they are unwilling. Because they cannot find them. And the system was never designed to help them look.

The Question I Brought to Dr. Ho

I went into this conversation with a specific reframing I wanted Dr. Yin Ho to react to. Most companies in clinical trial recruitment solve this problem from the pharma side — finding patients for trials. What if the more powerful and more ethical solution is the reverse — finding trials for patients?

Her response was immediate and unequivocal.

I think you might end up with a better result if you shift the paradigm.

Why the Current Model Has Failed Patients

Patients have technically had access to lists of clinical trials for years. ClinicalTrials.gov has existed since 2000. And yet the enrollment rate has barely moved. Why?

Dr. Ho identifies three structural failures. First — patients often do not know trials exist at all. The information has not reached them. Second — they are almost entirely dependent on their physician to tell them, and most physicians do not have easy access to trial information either. Third — even when patients find a trial, understanding whether they qualify — the inclusion and exclusion criteria — is a complex, technical task that most patients cannot navigate alone.

The entire clinical trial information infrastructure was built in a B2B direction — to serve pharma, biotech, and CROs. The information flows from sponsors to sites to investigators. It almost never flows directly to patients or their community physicians. And so the patient — the most essential participant in the entire enterprise — has been an afterthought in a system designed around finding them rather than serving them.

The Community Physician — The Insight I Did Not See Coming

I expected Dr. Ho to validate the patient-first paradigm. What I did not expect was the insight she added unprompted — and it is the most important thing she said in this entire clip.

Most patients are not seen in academic centers. Most patients are seen in their community physicians around wherever they live. The information asymmetry has to be solved in two places — one for your patients and two for your physicians.

The information asymmetry in clinical trials is not just a patient problem. It is equally a community physician problem. The specialist at a major academic medical center knows about the trials happening in their institution. The primary care physician in a community practice — who sees the patient every six months and has the relationship that makes a trial recommendation trusted and actionable — has almost no access to that information.

We have been trying to solve the last mile of clinical trial recruitment by going directly to patients. But the more powerful last mile runs through the community physician who already has the patient's trust.

What This Means for Founders

If you are building in clinical trial recruitment or patient engagement, the paradigm shift Dr. Ho validates is your strategic direction. But the harder and more important insight is the community physician angle. The academic medical center market is crowded and well-served. The community physician market — where most patients actually receive care — is almost entirely untouched.

The founder who builds the intelligence layer that reaches community physicians — giving them real-time awareness of trials their patients may qualify for — will unlock an enrollment pipeline that the industry has never been able to access. That is the white space. That is the opportunity.