Anil Makam-On Making Sense Of Diagnostic Tests, Defending Unbounded Clinician Excellence, & How Med Students Can Avoid Deskilling in the A.I. Era

“Anil Makam, MD, MAS, is a hospital medicine physician and nationally recognized health services researcher whose work sits at the intersection of hospital medicine, aging, and post-acute care. He is a passionate clinician educator and is well known for his work in evidence-based medicine (EBM), critical appraisal, and high value care. In this conversation, we cover a wide range of topics in EBM, diagnostic thinking, and A.I. in medicine.”

- Taken from Dr. Makam’s UCSF profile page (https://healthpolicy.ucsf.edu/profile/anil-makam)

Show Notes:

[0:00] Introduction & the Importance of clinical epidemiology

[3:00] A Brief Introductory Exploration of Bayesian Reasoning

• Bayesian reasoning formalizes how doctors already think intuitively

  • The formalisation allows doctors to make their reasoning explicit (a key rational norm of our profession)

• …promotes high value care, avoids diagnostic cascades, and it’s aesthetically satisfying

• Vibes will always be a part of clinical reasoning and bayesian reasoning can make it more explicit so that—one hopes—we can reason with it better

• Reference Class Problem

  • The reference class problem is a fundamental issue in probability and statistics (especially when trying to assign probabilities to single events or individuals). It arises when you need to estimate the probability of something happening to a specific case, but you only have data in the form of frequencies or relative frequencies from groups/classes of similar cases.The core question is: Which group (reference class) should you use as the basis for your probability estimate? Different choices of reference class can lead to dramatically different probabilities — sometimes wildly contradictory — even though all seem plausible at first glance.

[11:30] How Did Dr. Makam Revamp his Clinical Reasoning w/ his Epidemiology?

• As clinicians grow in experience and mastery they can evolve from an anxious shotgun approach of diagnostic testing to a hypothesis driven sniper approach of diagnostic testing

  • General population data » re-calibrates relative diagnostic weight you place on H&P and lab tests

  • Likelihood ratios > sensitivity or specificity because it more closely mirrors how clinicians intuitively think

    Sensitivity and Specificity Explained Clearly (Biostatistics)

    

    YouTube · MedCram - Medical Lectures Explained CLEARLY

    386.9K+ views · 11 years agoThe sensitivity and specificity are lying to you

    

    First10EM

    http://first10em.com › the-sensitivity-and-specificity-are...

  • For lab findings that are continuous variables we should use continuous likelihood ratios because it quantitates the intuition that the severity of abnormality might merit differing levels of diagnostic weights

    • https://pubmed.ncbi.nlm.nih.gov/28370759/

  • PPV and NPV is contingent on the tested population which may not match your particular clinical context—hence we may use likelihood ratios instead (there are assumptions baked into this inference

• Diagnostic literature makes it difficult to make sense of continuous likelihood ratios but if raw data is available, you can calculate it yourself

• Dr. Makam has graciously provided resources as a primer to interval likelihood ratios and likelihood ratios in general as well as bayesian clinical reasoning for problem representations and illness scripts

  • His worksheets are here:

    • https://drive.google.com/drive/folders/10a-XAj2FCh1UTEuoKRiXtT2QJ913f0nM?usp=sharing

  • His resources are here:

    • Basic primer on simplifying LRs: https://pubmed.ncbi.nlm.nih.gov/12213147/

      Advanced primers & examples on WHY and HOW to use interval LR for continuous tests:

      https://pubmed.ncbi.nlm.nih.gov/34298090/

      https://pubmed.ncbi.nlm.nih.gov/28370759/

      Scaffold for Bayesian clinical reasoning is problem representation & illness scripts. Here are necessary scaffolds to apply this type of thinking:

      Simple PR:

      

      Anil Makam@AnilMakam

      Clinical Reasoning: Problem Representation SGIM had outstanding primers put together by UCSF colleagues website down so here are pics

      

      

      11:04 PM · Jul 23, 2025 · 143 Views

      ---

      1 Reply · 2 Likes

      Simple illness scripts:

      

      Anil Makam@AnilMakam

      Clinical Reasoning: Illness Scripts SGIM primer in pics This is also good primer: journals.lww.com/academicmedici…

      

      

      11:04 PM · Jul 23, 2025 · 128 Views

      ---

      1 Reply

      More detailed putting it all together: https://www.nejm.org/doi/full/10.1056/NEJMra054782

[26:30] Why Dr. Makam believes that certain doctors can outperform the specificity and sensitivity findings on physical exam manoeuvres commonly documented in the scientific literature

• Dr. Makam provides a conceptual argument that clinical excellence is under-appreciated in discourse regarding the diagnostic weight of physical exam manoeuvres and POCUS interventions

  • To learn more about his paper “Striving For Diagnostic Excellence: The Median is Not the Message”

    • https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5044377

  • The Threshold Model of Clinical Decision-Making (Strong ...

    

    YouTube·Strong Medicine·Apr 7, 2021

[37:30] What medical education is NOT doing to prepare students for the rapid integration of artificial intelligence tools into healthcare

• Students need to learn “human skills”

  • Leaning into Vibes based medicine (cultivating an informed prior probability, curating context (H&P), building a well calibrated clinical sense, moving beyond schemas and algorithms, and building experience in edge cases)

  • Embodied knowledge (learning to learn, loving learning, evaluative thinking skills, etc)

  • Communication

    • Building rapport

    • Communicating risk