Why AI Could Make Health IT's Biggest Problems Worse — and What to Do About It

Dr. Yin Ho, former interim CEO of Veradigm and author of Rushing Headlong: Health IT's Legacy and the Road to Responsible AI, explains why 30 years of flawed data infrastructure pose a serious risk when artificial intelligence is applied without understanding the history behind it.

The Book That's Changing the Conversation Around Health AI

When Dr. Yin Ho stepped back from executive leadership, she didn't slow down — she started writing. The result is Rushing Headlong: Health IT's Legacy and the Road to Responsible AI, a landmark book that combines personal memoir, policy history, and a clear-eyed warning about the risks of deploying AI on top of broken data systems.

"I felt really compelled to write this book because I knew that AI was moving very fast," Ho explains. "Most people didn't know the history of how we got there."

The book is available in hardback, paperback, and ebook at all major booksellers.

What Is Rushing Headlong Actually About?

The book takes readers on a detailed journey through the origins of health information technology in the United States, stretching back to the 1990s and beyond. It traces how electronic health records (EHRs) evolved, why adoption stalled, and how a pivotal government policy decision in 2009 changed everything — but not necessarily for the better.

Most importantly, it connects that history to today's urgent question: can we trust AI systems that are trained and deployed on health data shaped by decades of compromised design decisions?

Dr. Ho's answer is nuanced: AI in healthcare has enormous potential, but only if practitioners, policymakers, and technologists understand the quality — and the limits — of the data they're working with.

The HITECH Act: A Policy Success That Created a Data Crisis

One of the most important sections of the book examines the Health Information Technology for Economic and Clinical Health (HITECH) Act, tucked inside the American Recovery and Reinvestment Act of 2009. Following the global financial crash of 2008, the U.S. government used economic stimulus legislation to dramatically accelerate EHR adoption.

The mechanism was simple but powerful: financial incentives for adoption, combined with penalties of up to 5% of Medicare and Medicaid reimbursements for non-adoption. The results were immediate — EHR adoption in the U.S. jumped from around 20% to approximately 80–90% in a short period.

"Everyone would say, that's a great policy success," Ho says. "But the unintended consequence was that it divorced the physician from the buying decision."

What That Means in Practice

When adoption became mandatory rather than market-driven, vendors no longer needed to design systems that physicians actually wanted to use. Instead, they had to satisfy the payers — primarily the government and insurance systems — who were footing the bill.

The result? EHRs became billing machines.

"All of the activity that happens in an electronic health record, even if you're capturing the notes of an observation, it really is a bunch of drop-down menus, a bunch of checkboxes," Ho explains. "A large part of it is that you're asking a physician to pre-write or pre-code a bill."

Clinical truth — the nuanced, contextual, narrative information that makes health data genuinely useful for research or AI training — got buried under layers of billing codes.

Real World Data: Converting Receipts Into Research

The consequences of billing-first EHR design ripple through to clinical research today. The entire industry of "real world data" exists, in part, because raw EHR output resembles a set of financial receipts more than a clinical record.

"I'm trying to figure out what happened to you by looking at all your receipts," Dr. Ho says. "That's really the position we're in."

Converting that claims data into something usable for research costs significant additional time and money — resources that would be unnecessary if the underlying systems had been designed with research and clinical clarity in mind.

This has direct implications for AI. If a model is trained on data that was designed for billing rather than clinical accuracy, it will optimise for billing patterns — not health outcomes.

"You're not going to end up finding more clinical truth," Ho warns. "You're only just going to accelerate the billing game."

What Physicians Can Do About It

Dr. Ho has spoken at medical schools, physician groups, and events including Bloomberg — and she's noticed a troubling pattern: resignation.

"I've come across this sort of resignation that they've been beaten down by the system," she says.

As more physicians become employees of large health systems rather than independent practitioners, they lose direct influence over the technology they use. But Ho argues that employee status also brings collective power.

Her advice to physicians:

• Demand a seat at the design table. No new system — AI-powered or otherwise — should be designed without direct physician input.
• Don't rely on the chief medical officer alone to represent your clinical perspective.
• Use collective bargaining power. Employment creates the opportunity to organise and push back — not to eliminate technology, but to shape it.

Interestingly, Ho discovered a small but notable group of independent physicians who have returned to paper records entirely — the ultimate opt-out. Her interpretation: it reflects a rational response to systems that weren't designed for clinical work. "When they write something down on a piece of paper and slide it into that manila folder... that piece of paper will still be there and it won't be corrupted."

What Patients Can Do Right Now

Patients are not passive recipients in this system. Dr. Ho has a practical, immediate recommendation:

Start collecting your own health data.

"There is no one place that has all of your data," she explains. Health records are scattered across hospitals, specialist practices, telehealth providers, and insurers. Assembling a complete picture is difficult — but increasingly important.

Ho suggests:

• Request records from every provider you've seen
• Include wearable and device data (Apple Watch, Fitbit, etc.)
• Store everything — even in a physical folder — while better digital tools emerge

New personal health record tools are in development that may eventually help aggregate this data. But in the meantime, individuals who take ownership of their records may be the only ones capable of holding a complete picture of their own health history.

"You may actually be the only place where all of the data about you can be centralised," Ho says.

The Fragmented System: Why Data Still Doesn't Talk

A recurring theme in Rushing Headlong is fragmentation — the persistent failure of health IT systems to interoperate. Records created in one system often cannot be accessed by another. Research and clinical care remain largely siloed. And patients who move between providers, states, or insurance plans frequently find that their records don't follow them.

This fragmentation isn't accidental. It reflects decades of commercial incentives, policy decisions, proprietary systems, and missed opportunities to build shared infrastructure. Understanding why the system is fragmented is essential context for anyone trying to use AI to work within — or around — it.

Why This Book Matters for AI in Healthcare

The timing of Rushing Headlong is deliberate. As generative AI capabilities expand rapidly, and as healthcare organisations race to integrate AI tools into clinical workflows, the book offers a necessary counterweight: a detailed understanding of what the data actually represents.

"Going forward, we should be at the design table," Ho says — and that applies to AI systems as much as EHRs.

The book is aimed at a general audience, not just healthcare insiders. Ho rewrote it multiple times to ensure that readers without a clinical or technical background could understand the implications. "This actually affects everyone," she says simply.

About Dr. Yin Ho

Dr. Yin Ho is a healthcare executive and author with deep experience across clinical informatics, real world data, and health technology leadership. She served as interim CEO of Veradigm, where she led the acquisition of a generative AI company focused on building small language models to improve health data quality.

She is currently on a book tour for Rushing Headlong and will be appearing at ISPOR (booth 337) in May, where she will be signing copies of the book.

Rushing Headlong: Health IT's Legacy and the Road to Responsible AI is available now in hardback, paperback, and ebook at all major booksellers, including Amazon.

This interview was conducted by Pippa Wilson of Barrington James.

Related Topics

• Health information technology (Health IT)
• Electronic health records (EHR)
• Responsible AI in healthcare
• HITECH Act and EHR adoption
• Real world data and clinical research
• Patient data rights
• AI governance in medicine
• Healthcare policy and physician burnout
• Clinical AI risk management
• Barrington James ISPOR 2025