Practical AI in Healthcare

When Adam Blum was diagnosed with follicular lymphoma, he tried over a dozen commercial trial matchers. None returned actual matches. So the serial AI entrepreneur built CancerBot, a free precision-matching service that assesses 100% of eligibility criteria — not the five surface-level attributes most matchers use. On this episode, Blum explains the Prompt Workbench (where biomedical experts refine extraction prompts to above 90% accuracy), how conjunctive normal form makes complex eligibility logic tractable, and why "best trial" means something different for every patient. A masterclass in AI scaffolding for healthcare.

For 37 years, Charlie Harp heard the same thing from healthcare organizations: "Our data quality is fine." They were right — for billing and scheduling. But AI changed the equation. Harp, founder of Clinical Architecture, built the PIQI framework to measure patient data quality across four dimensions: availability, accuracy, conformance, and plausibility. His PIQXL Gateway scores data on a 1-100 scale before it enters your systems — not after. Early deployments reveal uncomfortable truths: lab data averages 70% quality against USCDI standards, and one facility coded every blood test to a single LOINC code. The framework is now going through HL7 balloting as an open national standard.

Every day, patients leave US hospitals with discharge instructions they can't read. Giovanni Donatelli, CEO of The Language Group, built FETCH — a patented AI system embedded in Epic that translates discharge documents in 15 minutes with human review. He did it because he was the 8-year-old interpreting for his immigrant parents at doctor's appointments. Hosts Steve Labkoff and Leon Rozenblit explore the discharge instruction gap, the tragic cases that make it personal, FETCH's three-layer translation pipeline, the case for keeping humans in the loop, and why healthcare executives think they've already solved a problem that doesn't yet have a solution.

Steve and Leon have reviewed blocks of guest episodes twice before on Practical AI in Healthcare. Both times the themes snapped into place. This time they didn't -- and the disagreement between them became the episode. Across five recent conversations, they found stories that kept spilling past the edges of their framework: AI that works but can't get paid, laws that already apply but nobody realizes it, and a scientific record under threat from AI-generated paper mills. The hosts' attempt to make sense of it all reveals where their thesis holds, where it breaks, and what needs to change.

Bob Wachter wrote the book on the EHR disaster. Now he's written one about AI.

The UCSF Chair of Medicine joins hosts Steve Labkoff and Leon Rozenblit to discuss A Giant Leap, his argument that AI doesn't need to be perfect—it needs to beat a healthcare system already failing at scale. They cover Watson's $3B collapse, why ambient scribes became AI's first clinical success story, the human-in-the-loop problem that nobody has solved, and the dangerous gap between how experts and novices use AI tools.

Key topics: productivity paradox, complementary innovations, clinical decision support design, AI literacy, and the "compare me to the alternative" thesis.

The link to the book can be found here: https://a.co/d/07JFNwIw

What if early signs of psychosis could be detected from how patients speak—not what they say, but how they organize their thoughts?

Amar Mandavia (VA Boston, Boston University) and Enrique "Kike" Gutiérrez (Polytechnic University of Madrid) join hosts Steve Labkoff and Leon Rozenblit to discuss CHiRP, an AI tool that identifies formal thought disorder from routine clinical conversations. They explain why the gold-standard manual test takes 5+ hours, how their system reduces that to minutes, and the hard ethical questions around labeling patients as "at risk."

Key topics: prodromal psychosis detection, NLP in mental health, clinical workflow integration, MIT linQ Catalyst, and the payer challenges that make prevention hard to fund.

Real-world evidence was supposed to accelerate drug development. Instead, we've created definitional chaos—over 100 data vendors, inconsistent definitions, and studies that can't be compared.

Dr. Aaron Kamauu, CEO of Navidence and co-host of Real World Wednesday, explains why one missing diagnosis code can exclude 30% of your cohort, how GLP-1 eligibility criteria vary wildly between NHS and US guidelines, and what it means to document "the seven definitions you chose NOT to use."

A conversation about the unsexy infrastructure that makes evidence trustworthy.

In this episode of Practical AI in Healthcare, we sit down with Dr. Jeff Chuang, a computational biologist at The Jackson Laboratory, to explore how AI is reshaping cancer diagnostics, starting with pediatric sarcoma. Jeff shares his journey from physics and protein folding to computational pathology, where machine learning is being applied to standard H&E pathology slides to deliver faster, cheaper, and more accurate diagnoses.

The conversation dives into how AI models trained on relatively small but carefully curated image datasets can outperform traditional diagnostic approaches, especially in rare cancers where expertise is scarce. We also explore the challenges of data sharing, IRB approvals, and real-world deployment, along with a glimpse into the future of spatial genomics and ultra-high-resolution tissue analysis. This episode is a powerful example of how practical AI can directly improve patient care today.

In this episode of Practical AI in Healthcare, we sit down with physician–informaticist Josh Geleris, MD, co-founder and Chief Product Officer of SmarterDx, to unpack one of healthcare’s most overlooked AI opportunities: revenue cycle intelligence. Drawing on his clinical training, deep technical background, and firsthand experience inside large health systems, Josh explains how AI can bridge the gap between clinical reality and billing documentation. The conversation explores how machine learning and large language models translate thousands of data points from an inpatient stay into accurate, compliant coding, helping health systems reduce revenue leakage while staying firmly within regulatory guardrails. From SQL queries to post-trained LLMs, Josh walks us through the evolution of SmarterDx’s AI stack and why human-in-the-loop design remains essential. This is a grounded, practical look at AI delivering real value where healthcare operations and clinical truth collide.

In this episode of Practical AI in Healthcare, we sit down with Dr. Alvin Liu, retinal surgeon and Professor of Artificial Intelligence and Ophthalmology at Johns Hopkins University, to explore one of the earliest and most successful real-world deployments of medical AI.

Dr. Liu walks us through the evolution of autonomous AI for diabetic retinopathy screening, from FDA approval to large-scale clinical implementation across health systems. We unpack what it really takes to move AI from validation to impact, including workflow integration, sensitivity and specificity tradeoffs, reimbursement challenges, and post-market monitoring. The conversation also looks ahead to emerging AI applications using retinal imaging to predict cardiovascular disease, dementia, and kidney disease at the population level.

This episode is a masterclass in how AI can meaningfully improve access, equity, and outcomes in healthcare when deployed thoughtfully and responsibly.