Learning with Lim Episode #2

Daniel and I record another conversation discussing our latest learning pearls around clinical medicine and the philosophies behind it

Learning Point #1 (Joshua) (1:00)

• The case-based reasoning model fleshes out my approach to making sense of evidentiary disputes in clinical medicine

  • Different evidentiary commitments can be due to differing weightings of kinds of medical evidence (Bluym & Tonelli 2025)

    • This may constitute what it means for physicians to have differing evidentiary commitments

      • Patient values vs population level data vs pathophysiologic rationale vs clinical experience

• #2 The argument framework for discourse in medicine

  • Measurement vs Argument Framework

    • https://www.nature.com/articles/s41591-024-02930-x

  • The measurement framework

    • “The dominant framework for medical evidence is the measurement framework (2) , used in orthodox EBM. This framework imparts ‘gold-standard’ status to the randomised clinical trial (RCT), and underlies meta-analysis, evidence ranking schemes such as GRADE (Grading of Recommendations Assessment, Development and Evaluation) and the US Food and Drug Administration (FDA)’s traditional approval pathway for drugs culminating in phase 3 clinical trials. In the measurement framework, therapeutic evidence consists of measurements of an intervention’s effects. Consequently, only comparative studies such as RCTs that directly measure effects, by comparing differently treated clinical populations, provide therapeutic evidence. The measurement framework draws mainly on clinical epidemiology for evaluating and synthesizing evidence. Clinical epidemiology is a science of measuring outcomes in clinical populations, and this field developed the standards for medical RCTs and meta-analyses.”

      • “The measurement framework works best in a scenario in which the clinical question is simple, such as “What is the effect size of statins for primary heart attack prevention in this population?”

Learning Point #2 (Daniel) (16:30)

PE patient, treated with eliquis. During rounds, we discussed extended treatment. Really confused…

1. Big picture - treatment of VTE

1. phases of treatment - initial phase and extended phase

2. balance of risks and patient characteristics

2. Risk factors were characterized using the language as provoked and unprovoked (first appeared in 2012)

1. Provoked = identifiable and reservable risk ie major surgery, trauma, immobility.

2. Unprovoked = risk cannot be identified

1. Based on observational studies showing that patients with surgical trigger had the lowest risk of recurrence. Patients with non-surgical risk factors had an intermediate risk and patients with non-identifiable risk factors had a high risk of recurrence (citations 1 and 2)

2. Guidelines admit to the ambiguity and confusion inherent in this framework

   1. long haul travel vs oral contraceptives - long haul travel is often defined as unprovoked; VS oral contraceptives are provoked

   2. Furthermore framework turns a complex continuum into a binary decision.

   3. Does not acknowledge that most patients present with risk factors in situations of clinical uncertainty, in which the management decision is not clear-cut

3. 3. Risk factors are now categorized using the descriptors, transient/persistent, if transient, characterize as major/minor, and if persistent, characterise as malignant vs non malignant

   1. New framework adds to provoked/unprovoked by taking into consideration temporal patterns and strength of association.

      1. Temporal pattern - Transient and persistent

      2. If transient, consider the strength of association to index events. Risk of index event has an inverse relationship with recurrence risk

         1. Major = more than a 10-fold increase in risk of index event

            1. Major surg, confined to bed for more than 3 days, c-section, hospitalized trauma

            2. Healthy patients with major transient risk factor are in the low risk group and don’t require extended treatment

         2. Minor = <10-fold increase in risk of index event

            1. Transient and minor = intermediate risk

               1. Long haul flight, hospitalization of less than 3 days, minor surgery, oral contraceptives

               2. Transient and minor = most tricky, consider patient characteristics, type of VTE, values

                  1. 31 year old presenting with dvt after ankle sprain 2 days prior and on OC vs 31 year old, obese, and PE

         3. If transient but with underlying risk factor - patients might benefit - 2025 HIPRO trial

      3. If persistent, characterize as malignant vs. non-malignant

         1. For malignant long term anticoagulant recommended

         2. For non malignant - obesity, active inflammatory bowel disease, recurrent long-term travel, Nephrotic syndrome: continued hormonal therapy - low threshold for extended anticoagulants

      4. If risk factor cannot be identified - unprovoked - extended anticoagulant

      5. Type of risk factor is one thing to consider amongst many other variables ie patient demographic, type of vte, past hx of vte, bleeding risk, pref/values -All of these factors can determine whether a patient might be benefited or harmed by extended anticoagulants.

      6. Most patients will not fall into the clear-cut category of having a transient and major risk factor or pro - 2023 review

         1. Important to have a discussion with patient and be transparent about limitation

1. Incidence of recurrent venous thromboembolism in relation to clinical and thrombophilic risk factors: prospective cohort study
   Baglin, Trevor et al.
   The Lancet, Volume 362, Issue 9383, 523 - 52

2. Paolo Prandoni, Franco Noventa, Angelo Ghirarduzzi, Vittorio Pengo, Enrico Bernardi, Raffaele Pesavento, Matteo Iotti, Daniela Tormene, Paolo Simioni, Antonio Pagnan. The risk of recurrent venous thromboembolism after discontinuing anticoagulation in patients with acute proximal deep vein thrombosis or pulmonary embolism. A prospective cohort study in 1,626 patients. Haematologica 2007;92(2):199-205; https://doi.org/10.3324/haematol.1051

3. HiPRO trial - Piazza G, Bikdeli B, Pandey AK, et al; HI-PRO Trial Investigators. Apixaban for Extended Treatment of Provoked Venous Thromboembolism. N Engl J Med. 2024;390

4. Chest Guidelines - Stevens SM, Woller SC, Kreuziger LB, et al. Antithrombotic Therapy for VTE Disease: Second Update of the CHEST Guideline and Expert Panel Report. Chest. 2021;160(6):e545-e608. doi:10.1016/j.chest.2021.07.055

Provoked vs minimally provoked vs unprovoked - does it matter? Becattini. et al.Becattini C, Cimini LA. Provoked vs minimally provoked vs unprovoked VTE: does it matter? Hematology Am Soc Hematol Educ Program. 2023 Dec 8;2023(1):600-605. doi: 10.1182/hematology.2023000492. PMID: 38066936; PMCID: PMC10727063.

Learning Point #3 (Joshua) (28:00)

Isolated Leukocytosis patient case

https://pmc.ncbi.nlm.nih.gov/articles/PMC6352401/

• May necessitate a conversation about the “normal range” philosophy of medicine paper (The normal range: It is not normal and it is not a range)

  • Key points that are relevant

    • The “normal range” is actually a reference range → it’s a comparator interval to a population distribution rather than an arbiter of whether or not someone has a disease

      • The reference range makes a series of assumptions

        • Assumes a Gaussian distribution (it could be skewed like fasting triglyceride!)

          • The word normal came because it was ubiquitous in life not because it identified “normal” or “healthy” people

        • No physiological theory assumes that central 95% is physiologically normal. More likely borrowing and calcification of methodological assumptions from Fisher’s development of hypothesis-testing techniques

      • EXAMPLES

        • Hyponatremia in the elderly (low levels may be physiologically expected)

        • Urea in people with cirrhosis (in the upper range may be abnormal)

• The bottom line here is that one of the reasons why history is king in internal medicine is that reference ranges are not meant to be arbiters on the presence of disease. The reference values are not diagnostic tests. They are values that fit within a particular argument for what’s happening in a particular clinical context. The story forms the bulk of the argument.

  • People take this as gospel in internal medicine and it’s mostly true…but the reasoning is a bit more nuanced

    • The range is a comparator to population values with its own interpretive assumptions that are disconnected from our aim as clinicians to diagnose disease

      • Classic example of incorrectly re-purposing a tool for a different aim

      • If you think a particular threshold of test in a reference range means something for the diagnosis of a particular disorder/disease then you need to test it with a diagnostic test

        • Classic sensitivity and specificity diagnostic testing needs to be done

Learning Point #4 (Daniel) (42:00)

1. When you are busy, using and over relying on clinical prediction models can be very tempting

2. Many new clinical prediction models have not assessed whether they make accurate predictions

   1. Assessment = comparing predictions vs outcomes

      1. External validation is important

         1. Difference between - Internal validation vs external validation

      2. Bouwmester et. Al - systemic review 3/71 were externally validated

      3. External validation important because demonstrate robustness of the model - clinicians care about whether the model with work HERE

      4. Example of model with evidence demonstrating that it performs well in various populations - which is why its recommended by professional society

         1. CHADSVASC

            1. 2012 swedish study - friberg et al 2012

            2. 2015 taiwanese study - chao et al 2015

            3. 2014 european study - lip et at 2014

3. Even externally validate clinical prediction models have limitations.

   1. Example - look back JACC study - 2025 mueller et all

      1. The authors reviewed 465 patients who got their first M.I, tested how the ASCVD Risk Estimator Plus will perform if the patient presented two days before their first MI.

         1. 33% were categorized as low risk

         2. 10% were categorized as high risk

   2. Why might validated models perform poorly?

      1. Differences between population

         1. different characteristics, type of center ie sec or tert

         2. Even when averages are similar, different dispersions can affect performance

            1. Riley et al - 3 - 2016 paper demonstrating that even when averages might be similar, less dispersed populations might have lower discrimination performance discrimination performance

            2. in other words the more alike the patients are within a population, the more difficult it is to separate those from high and low risk

      2. Changes in medicine over may cause prediction drifts

         1. medical advancement, patient populations, standards of care, life expectancy, baseline risks

            1. ex

               1. 1. in-hospital mortality after myocardial infarction now compared to the 1980s will be significantly different

               2. 2. another example is the Framingham Risk Score the population in the 1970s and 80s, or different compared to the population now. paper from 2012 and 2013 showing that the Firmingham risk score overestimates coronary heart disease in the modern population

3. Diseases are endlessly complex and our current paradigm might be inadequate to predict risk adequately

   1. james lind example - vitamin c deficiency - plug psych at margins awis aftab

      1. cure to scurvy lost by 1900s - what happened?

         1. lemon to lime switch + copper storage observation

         2. fresh meet vs preserved meat observation

      2. example of how current paradigms and concepts determines the conclusions you make from your observations

   2. CAD - our currently paradigms don’t sufficient explain what makes a plaque stable vs unstable

Bottom line - Have a healthy skepticism when using any clinical prediction model and before using, identify:

1. What is it claiming to do?

2. In what context has it been externally validated?

3. How is your population different from the population in which the model has been validated? How might these differences affect the predictions?

4. How will you use the model? What will you do if your clinical assessment disagrees with model predictions?

Van Calster B, Steyerberg EW, Wynants L, van Smeden M. There is no such thing as a validated prediction model. BMC Med. 2023 Feb 24;21(1):70. doi: 10.1186/s12916-023-02779-w. PMID: 36829188; PMCID: PMC9951847.

Prediction explainer: what should clinicians consider when considering a new clinical prediction model?

Sensible Medicine article by Nicole White and Adrian Barnett

Mueller, A, Leipsic, J, Tomey, M. et al. Limitations of Risk- and Symptom-Based Screening in Predicting First Myocardial Infarction. JACC Adv. 2025 Dec, 4 (12_Part_2) .

https://doi.org/10.1016/j.jacadv.2025.102361

Look back study evaluating ASCVD clinical prediction rule or risk assessment.