Private Podcast

Michael's personal things to listen to

This site hosts a simple podcast RSS feed for personal listening apps. Subscribe at /feed.xml.

Audio files are served from https://audio.tapestry-labs.com.

Metabolic Response During Radiotherapy: PERCIST, Lugano, and the Mid-Treatment PET Question

Sep 14, 2026 · 00:19:28

A practical guide to the rationale, methods, limitations, and future of PERCIST and the Lugano classification, with a critical look at mid-treatment FDG-PET for response assessment and adaptive radiotherapy in stage III non-small cell lung cancer.

00:00 The Problem These Frameworks Were Built to Solve, 01:47 Why PERCIST Was Needed, 03:48 PERCIST in Practical Terms, 07:51 What Lugano Changed for Lymphoma, 10:12 Two Frameworks, Two Philosophies, 12:22 The Mid-Treatment PET in Stage Three Lung Cancer, 15:28 An Editorial View: Measure Freely, Act Carefully, 17:57 The Take-Home Framework

Reconstructing Clinical Decisions: Seven Papers for Designing Chart-Stimulated Interviews

Sep 02, 2026 · 00:49:31

A deep dive into seven papers on introspection, naturalistic expertise, framing, chart-stimulated recall, dyadic analysis, distributed decision-making, and shared decisions in advanced cancer, translated into a practical design for clinician-patient decision interviews.

00:00 Opening: What Kind of Truth Can an Interview Recover?, 02:24 Paper One: Nisbett and Wilson on Telling More Than We Can Know, 08:22 Paper Two: Klein and the Critical Decision Method, 14:35 Paper Three: Tversky and Kahneman on Framing, 19:05 Paper Four: Sinnott and Colleagues on Chart-Stimulated Recall, 25:03 Paper Five: Eisikovits and Koren on Dyadic Analysis, 29:04 Paper Six: Rapley and the Distributed Decision, 33:18 Paper Seven: Brom and Colleagues in Advanced Cancer Care, 38:08 A Practical Interview Architecture, 42:31 What the Dataset Should Capture, 44:49 Comparative Effectiveness Research: Confounding With Measurement Error, 46:31 Clinical Simulation: Reproducing the Decision Trajectory, 48:20 Closing: A Measured Reconstruction, Not a Mind Reading Device

From Treatment Prediction to Clinical Simulation: Five Papers and a Blueprint for Cancer Digital Twins

Aug 29, 2026 · 00:37:27

A step-by-step synthesis of five papers on predicting oncology decisions and patient trajectories, with a practical blueprint for high-fidelity physician and patient personas in cancer-treatment simulations.

00:00 Opening: What Exactly Are We Trying to Simulate?, 02:19 Paper One: Breast Tumor Board Design and Labels, 04:48 Paper One: Results and Lessons, 07:55 Paper Two: Longitudinal History and Joint Decisions, 09:46 Paper Two: Performance and Simulation Lessons, 12:01 Paper Three: CASCADE and the Multilabel Tumor Board, 13:55 Paper Three: Performance, Context, and Missing Process, 16:13 Paper Four: Modeling the Surgery-versus-SBRT Boundary, 17:46 Paper Four: The Intermediate Zone and Patient Choice, 20:30 Paper Five: DT-GPT as a Patient-Dynamics Model, 22:25 Paper Five: Benchmarks, Figures, and Robustness, 24:27 Paper Five: Why Forecasting Is Not Yet Counterfactual Simulation, 26:28 Cross-Paper Synthesis: Three Layers, Not One Model, 29:04 A Practical Blueprint for Physician and Patient Personas, 31:25 How to Evaluate the Simulation, 34:13 A Focused Development Plan, 36:00 Closing

Who Recommends Surgery or SABR - and Whose Preference Counts?

Aug 28, 2026 · 00:31:27

A detailed retelling of two linked Hopmans studies from 2015 and 2016 on specialty, clinical factors, uncertainty, and patient preference in stage I non-small-cell lung cancer treatment recommendations.

00:00 Opening, 01:49 The Clinical Decision in 2015, 03:39 Building the Binary Choice Experiment, 06:05 Who Participated: Table 1, 07:33 What Drove a SABR Recommendation: Table 2, 10:01 Sixteen Patients, Many Recommendations: Table 3, 12:33 Different Specialty Lenses: Tables 4 and 5, 15:13 How the 2015 Authors Interpreted the Study, 17:26 The 2016 Paper Changes the Question, 19:07 Concordance and Uncertainty: The 2016 Table, 20:59 Figure 1: Preference Through the Lens of Specialty, 22:51 Figure 2: Belief About Whether the Treatments Are Equal, 24:36 Figure 3: Certainty, Specialty, and Preference, 26:17 The 2016 Interpretation and Its Limits, 28:19 What the Two Papers Show Together, 30:10 Closing

LLM Agents Grounded in Self-Reports Enable General-Purpose Simulation of Individuals

Aug 26, 2026 · 00:29:58

A read-aloud adaptation of pages 1 through 13 of Park and colleagues' paper on using interviews and surveys to create LLM agents that simulate individuals across social-science tasks.

00:00 Introduction: General-Purpose Simulation, 02:09 Evaluation Framework and Research Uses, 04:24 Using Self-Reports to Create Agents, 06:04 Sample and Data Collection, 08:00 Surveys, Games, and Experimental Tasks, 09:37 Agent Architecture, 11:37 Accuracy Metrics and Baselines, 12:55 Predicting Attitudes: The General Social Survey, 14:35 Personality and Economic Games, 16:27 Interview Ablations, 17:46 Predicting Experimental Replications, 19:26 Accuracy Disparities Across Groups, 21:16 Discussion, 22:49 Why Might Interviews Work?, 24:13 Limitations, 26:13 Materials and Methods: Participants, 28:03 Materials and Methods: Experiments and Robustness

The Fidelity and Feedback Traps: Health Digital Twins as Evolving Causal Systems

Aug 26, 2026 · 00:21:36

A direct read-aloud of the main text of Freeman and colleagues' paper on why health digital twins should be causally valid, modular, and governed as they evolve. The abstract, references, figure caption, affiliations, and end matter are omitted.

00:00 Title and authors, 00:16 Introduction, 02:51 Introduction: The two traps, 04:57 The fidelity and feedback traps, 05:32 The fidelity trap, 06:57 The feedback trap, 08:39 Causally valid, modular, evolving systems, 09:26 Causally valid and modular: answering the fidelity trap, 11:49 Evolving and living: answering the feedback trap, 14:45 Relation to adjacent approaches, 16:34 Why this matters and how it scales, 19:31 Discussion

ResidencyRL and the Road to Counterfactual Clinical Simulation

Aug 13, 2026 · 00:21:50

A detailed review of the ResidencyRL preprint—how it builds long-horizon simulated clinical encounters, what reinforcement learning changed, and what the work does and does not contribute to counterfactual outcome simulation for comparative effectiveness, malpractice, and quality or safety investigations.

00:00 The question behind the paper, 02:00 What ResidencyRL actually builds, 05:15 The kratom case and why the design matters, 07:30 How the reinforcement signal works, 10:03 What improved, and how convincing is it?, 13:32 Decision counterfactuals versus outcome counterfactuals, 15:20 The three proposed use cases, 17:33 A research architecture that could close the gap, 19:46 Bottom line

Clone, Censor, Weight: A Practical Guide to Time-Zero Causal Questions

Aug 03, 2026 · 00:18:10

An intuitive and practical guide to the clone-censor-weight method for causal research with observational data, closely based on Gaber and colleagues' 2024 Cancer Medicine primer, with oncology examples, implementation choices, diagnostics, and limitations.

00:00 The problem CCW is built to solve, 01:59 Start with the target trial, not the model, 04:02 Step one: clone, 05:25 Step two: censor when a clone deviates, 07:08 Step three: weight the uncensored histories, 09:14 From weighted histories to causal contrasts, 11:13 Three oncology patterns, 13:17 What CCW does not solve, 15:33 A practical analysis checklist

Action Classification For Endoscopic Pituitary Adenoma Resection A Consensus Based Study

Aug 02, 2026 · 00:19:05

Action classification for endoscopic pituitary adenoma resection: a consensus-based study By Joachim Starup-Hansen, Danyal Z. Khan, Adrito Das, Joao Paulo Almeida, Sophia Bano, Ano

Dissecting Cardiac Surgery

Aug 02, 2026 · 00:21:45

Dissecting Cardiac Surgery A Video-Based Recall Protocol to Elucidate Team Cognitive Processes in the Operating Room By Roger D. Dias, Marco A. Zenati, Heather M. Conboy, Lori A. C

The Confounding Stack: Treatment Intent, Provider Preference, and Reliable EHR Trial Emulation

Jul 29, 2026 · 00:29:45

A 30-minute synthesis of three papers on discovering unmeasured confounders, constructing provider-preference instruments, and stress-testing instrumental-variable analyses, with a practical research architecture for emulating many randomized trials in EHR data.

00:00 Opening: The Confounding Stack, 03:27 Paper One: Discovering Confounders Through Treatment Intent, 08:00 The EHR Is an Observation System, 11:17 Paper Two: Provider Preference Is a Moving Target, 16:34 Paper Three: Robustness Under Heterogeneous Effects, 20:04 An Architecture for Many Trial Emulations, 24:23 A Practical First Study, 27:19 What the Three Papers Change

Doctor Simulation, EHR Agents, and Locally Calibrated Clinical Decisions

Jun 17, 2026 · 00:31:19

A roughly 30-minute deep dive into three papers on behavioral simulation, real-EHR medical agents, and treatment-decision pivots, framed around a planned project on locally calibrated LLM agents for simulating real specialist recommendations.

00:00 Opening, 02:43 The Project Target, 05:24 Paper One: Behavioral Turing Tests, 09:48 Paper Two: PhysicianBench, 15:11 Paper Three: ClinPivot, 20:10 The Three Papers Together, 22:38 Architecture Implications, 25:01 Evaluation Implications, 27:25 A First-Paper Shape, 29:27 Closing

Physician Expertise, Procedural Judgment, and What AI Medical Agents Need to Learn

Jun 14, 2026 · 00:32:03

A 30-minute literature synthesis on physician expertise in procedural specialties: what excellence means, what subcomponents contribute to superior outcomes, how expertise can be measured, and what this implies for future AI medical agents.

00:00 Opening, 02:57 Competence Is Integrated Performance, 05:29 How Experts Think, 09:08 Slowing Down When It Matters, 12:14 Skill You Can See, 15:22 Skill That Predicts Harm, 18:43 Experience Is Not Enough, 21:33 Expertise As A System, 23:42 A Measurement Framework, 27:47 What This Means For AI Medical Agents, 30:33 Closing

Guidelines, Innovation, and Medical Judgment in the Age of LLMs

Apr 29, 2026 · 00:28:59

A detailed discussion of two medical innovation case studies, Semmelweis and laparoscopic cholecystectomy, and what they teach us about using LLMs to recommend or judge treatment.

00:00 Opening, 03:05 Semmelweis: The Evidence, 06:10 Semmelweis: The Resistance, 08:15 Laparoscopic Cholecystectomy: Rapid Adoption, 09:33 Laparoscopic Cholecystectomy: The Evidence Gap, 11:17 Laparoscopic Cholecystectomy: Critical Appraisal, 14:00 Laparoscopic Cholecystectomy: Why Enthusiasm Won, 15:49 The Pairing, 18:53 What This Means For LLM Recommendations, 21:51 What This Means For Judging Doctors, 24:52 A Practical Framework, 27:24 Closing

Crossing the Chasm for Agentic Cancer Registry Abstraction

Apr 29, 2026 · 00:18:55

An overview of Crossing the Chasm, translated into practical go-to-market ideas for an early-stage startup building agentic AI for hospital records abstraction, starting with cancer registry abstraction.

00:00 Opening, 01:47 The Adoption Curve, 04:00 What The Chasm Looks Like In Hospitals, 06:02 The Beachhead, 08:08 The Whole Product, 09:57 Positioning, 11:46 Competition And Alternatives, 13:22 The Bowling Alley, 14:56 What This Means For Your Startup, 16:47 Questions To Carry Forward, 17:43 Closing

OPTN Registry Meeting Prep: Data Flow, Abstraction, and AI Agent Opportunities

Apr 26, 2026 · 00:18:10

A focused briefing for a meeting with an OPTN registry leader, covering OPTN background, data flow, major forms and fields, submission cadence, current abstraction work, and where AI agents may fit.

00:00 Opening, 01:31 What OPTN Is, 03:21 The Data Flow, 06:19 Forms, Fields, and Cadence, 09:05 Who Does the Work Today, 10:58 Where AI Agents Could Fit, 13:21 What To Ask In The Meeting, 15:25 The Meeting Mental Model, 17:13 Closing

Anatomy as the Intermediate Representation: Four Papers on Spatially Aware Medical Vision Models

Mar 28, 2026 · 00:40:14

A long technical podcast on four recent papers about anatomy-aware reasoning, grounding, segmentation, and 3D spatial evaluation in medical imaging, tied back to the idea that future models may need an intermediate representation of the body rather than direct pixels-to-diagnosis shortcuts.

00:00 Opening, 03:26 The Big Thesis, 05:48 Paper One: AOR Overview, 08:21 Paper One: AOR Method and Ontologies, 10:49 Paper One: AOR Results and Interpretation, 13:33 Paper Two: AnatomiX Overview, 16:34 Paper Two: AnatomiX and Shortcut Resistance, 18:40 Paper Two: AnatomiX Results and Relation to the Ideas File, 21:01 Paper Three: KG-SAM Overview, 22:28 Paper Three: KG-SAM Method in Detail, 24:30 Paper Three: KG-SAM Results and Broader Meaning, 27:06 Paper Four: SpatialMed Overview, 28:42 Paper Four: SpatialMed Construction and Why It Matters, 30:20 Paper Four: SpatialMed Results and Failure Modes, 32:25 Cross-Paper Synthesis, 34:31 What Is Still Missing, 36:28 Which Papers Best Support the Post, 37:57 Final Thoughts

Chapter Test Episode

Mar 28, 2026 · 00:01:11

A longer test episode to verify automatic chapter timing from Markdown headings.

00:00 Opening, 00:25 Middle Note, 00:48 Final Check