Practical AI in Healthcare
Episode: S1, E12: Part 1 of 2 – Dr. S. Yin Ho Discusses Her New Book
Date: November 16, 2025
Guests: Dr. S. Yin Ho (Board Chair, Segmed; Founder, Context Matters; former executive at ScienceIO, Aetion; author)
Hosts: Dr. Steven Labkoff, Leon Rozenblit, JD, PhD
Episode Overview
This episode is the first of a two-part deep dive with Dr. S. Yin Ho, a pioneering health tech executive and author. The conversation centers on Dr. Ho's new book, "Rushing Health: Its Legacy and Road to Responsible AI", which examines the 25-year evolution of health IT, the rise and problems of the electronic health record (EHR), and the murky landscape of healthcare data. Dr. Ho, Dr. Labkoff, and Dr. Rozenblit delve into the historic roots of digital healthcare, dissect the mistaken assumptions that have shaped health data, and lay the groundwork for a critical look at generative AI's future impact.
Key Discussion Points & Insights
1. Dr. Yin Ho’s Background and Motivation for the Book
-
Healthcare IT Pioneer: Dr. Ho shares her journey from emergency physician to health IT entrepreneur and executive, spanning startups to major health tech companies.
-
Genesis of the Book: The book was inspired during her interim CEO role at Varadigm (Allscripts) just as generative AI began to dominate the industry (03:37). There, Ho noticed a persistent lack of understanding about the gap between clinical care and clinical research, leading her to write the book and document lessons so mistakes are not repeated.
Dr. Ho [03:37]: “What I found interesting at that moment was that a lot of people didn’t quite understand what gap I was referring to…The gap is actually something that has developed over the last 25 years…”
2. “The Gap”: Dividing Clinical Care and Clinical Research
-
Persistent Barriers: Despite advances, two silos remain: care delivery and research/data science. AI now promises to bridge them—but historic lack of integration persists.
-
Book Structure: Three parts:
- History and infrastructure of health IT
- Data quality and siloes
- The advent of generative AI and challenges ahead (07:31)
Dr. Ho [07:31]: “The runs through this infrastructure and how in some sense us not paying attention in the first sort of buildup of the health IT infrastructure caused us to end up with varying levels of quality around the data itself and therefore…entire industries being spent trying to, to correct a lot of the issues around the data.”
3. The “Original Sin” of Health IT
-
Design Flaws Rooted in Early Decisions: The earliest EHRs mirrored paper files—great for viewing, but not for data extraction or complex queries (10:32).
-
Emergence of Interoperability and Workflow Challenges: Early optimism gave way to unanticipated problems: lack of consistent data structuring, failure to support research, and failure to support clinical workflow complexity.
Dr. Ho [11:08]: “One of the big challenges...when clinical records came about, they were really just files...almost designed in that way where you recorded information that way. So with the assumption someone was going to go back in and just read it.”
-
Systemic Missed Opportunities: The “electronic paper system” grew without considering cross-functional needs, trapping data in forms hard to mine for research or even workflow support.
4. Clinical vs. Administrative Drivers in EHR Design
-
Trend from Physician-Centric to Billing-Centric Records: As EHRs evolved, billing and compliance—rather than clinical usability—became primary design drivers.
Dr. Ho [16:27]: “When it truly became an electronic health record was when we had to start to connect it into a payment structure...What starts to occur is you start to write in a way that’s not so much observational oriented. You write in a way that makes it easier to code.”
-
Physicians as Scribes: Doctors are forced to structure documentation mainly for billing; non-reimbursable but clinically vital observations are lost in unstructured notes.
-
Systemic Burden: Ever-tightening payment reforms and reporting requirements place the administrative onus squarely on clinicians (22:03).
Dr. Ho [22:03]: “You are now also burdening the physicians as the ones who must become the scribes to provide the documentation in order for it to get paid.”
5. Policy & Market Distortions: The Era of “Meaningful Use” and ONC
-
‘Meaningful Use’ as a Market Shaper: Post-2008, government incentives/penalties (through ONC and ARA legislation) turbocharged EHR adoption—but gave unprecedented power to a few large vendors (28:33).
-
Oligopoly of Vendors: Firms like Epic and Cerner became entrenched as government-certified vendors, giving them scale and leverage unaccountable to end users.
Dr. Ho [31:53]: “Whatever electronic health record vendors existed at that moment were going to be the only ones who were going to be able to go forward…they only have to meet the government regulations and say, I’m a certified…EHR vendor. You are forced to have to buy from me.”
-
Resulting Hostility: The system is not user-centered but transaction-centered, leading to widespread provider frustration and high clinician exit rates.
Dr. Labkoff [34:14]: “Every time [clinicians] turn around, they’re at the forefront of the risk. And the other folks in the equation don’t get the risk…in the same way that the clinicians do.”
6. The Rise of the “Second Healthcare IT System”: RWD & RWE
-
Two Worlds, Poorly Connected: Dr. Ho introduces the idea that there are now “two systems”—the EHRs built for operations/billing, and a secondary layer built for data analysis (RWD/RWE) (35:49).
-
Genesis of Context Matters: Her company Context Matters structured global health tech assessment data, exposing how analytic needs are distinct from operational records (36:21–41:29).
-
International Perspective: Many countries employ independent agencies to evaluate real-world evidence (RWE) and make population-specific decisions about drug and device coverage—unlike the US.
7. The Real-World Data (RWD) Conundrum
-
Data Not Collected for Research: Clinical notes and observations are inconsistently structured; retrospective research is hampered by ambiguity (42:33).
- Patient privacy adds another layer of complexity, requiring de-identification and reducing analytic fidelity.
- Example: A “six-minute walk test” might be described in prose, never tagged as a standardized data element—making aggregation for research very hard.
Dr. Ho [43:36]: “Since your data in the clinical care electronic health record isn’t structured, it does make it very difficult for you to understand if what one patient experienced is the same thing as what another patient may have experienced…It is almost like you’re reversing the concept of a clinical trial.”
8. Preview of Part 2: Responsible AI in Healthcare
- AI’s Potential (and Limits) in Health Data: The group foreshadows how generative AI and LLMs might automate extraction and structuring of clinical prose—for example, reliably identifying six-minute walk tests in large note sets (46:39).
- Foundational Problems Remain: Unless legacy data/workflow challenges are addressed, even the most advanced AI may hit systemic speedbumps.
Notable Quotes & Memorable Moments
-
On Data Quality:
“Entire industries being spent trying to, to correct a lot of the issues around the data.” — Dr. Ho [07:45]
-
On Shifting EHR Purpose:
“You start to write in a way that’s not so much observational oriented. You write in a way that makes it easier to code.” — Dr. Ho [17:44]
-
On Administrative Burden:
“You’ve not only just weakened sort of the physician’s role in taking care of patients, you’ve now given the physician even more burden of basically making sure that that engine continues to go forward.” — Dr. Ho [22:34]
-
On Market Distortion:
“It was transformative, but not necessarily in a great way from my perspective.” — Dr. Labkoff [28:26]
-
On RWD Challenges:
“Clinical [EHR] data is variable, messy, inconsistent, random syntax, collected for a different reason.” — Dr. Ho [43:53]
-
On the Hope for AI:
“An AI model might actually be able to recognize that that actually means a six-minute walk test…and do it instead of a human…” — Dr. Labkoff [47:13]
Timestamps for Key Segments
- [01:39] – Dr. S. Yin Ho’s background and motivations for writing the book
- [03:37] – Defining the “gap” between clinical care and research
- [09:58] – The “original sin” of early health IT design decisions
- [16:27] – How billing and payment priorities changed the EHR’s purpose
- [22:03] – How clinicians became the administrators by proxy
- [28:33] – Meaningful Use and ONC: how market incentives distorted the health IT landscape
- [35:49] – The two “healthcare IT systems” and comparative effectiveness
- [36:21] – The Context Matters story and international health tech assessment
- [42:33] – Structural and privacy challenges in using RWD/RWE
- [46:39] – Preview of part two: Generative AI’s role in the future of health IT
Final Thoughts
This episode is a candid, unsparing look at how health IT’s legacy of EHR-centric design, administrative imperatives, and market distortion has created deep problems for data quality and clinician experience. Dr. Ho’s personal and professional insights reveal the magnitude of the challenges facing AI in healthcare—and set the stage for a further discussion of what responsible AI could (and cannot) fix in part two.
Next Week: Dr. Ho returns to explore “the good, the bad, and the ugly” of generative AI’s collision with entrenched healthcare systems and what practical steps might lead to “responsible AI” in medicine.
