The Multidisciplinarian

For centuries, medicine has navigated the tension between science and uncertainty. The Covid pandemic exposed this dynamic vividly, revealing both the limits and possibilities of statistical reasoning. From diagnostic errors to vaccine communication, the crisis showed that statistics is not just a technical skill but a philosophical challenge, shaping what counts as knowledge, how certainty is conveyed, and who society trusts.

Historical Blind Spot

Medicine’s struggle with uncertainty has deep roots. In antiquity, Galen’s reliance on reasoning over empirical testing set a precedent for overconfidence insulated by circular logic. If his treatments failed, it was because the patient was incurable. Enlightenment physicians, like those who bled George Washington to death, perpetuated this resistance to scrutiny. Voltaire wrote, “The art of medicine consists in amusing the patient while nature cures the disease.” The scientific revolution and the Enlightenment inverted Galen’s hierarchy, yet the importance of that reversal is often neglected, even by practitioners. Even in the 20th century, pioneers like Ernest Codman faced ostracism for advocating outcome tracking, highlighting a medical culture that prized prestige over evidence. While evidence-based practice has since gained traction, a statistical blind spot persists, rooted in training and tradition.

The Statistical Challenge

Physicians often struggle with probabilistic reasoning, as shown in a 1978 Harvard study where only 18% correctly applied Bayes’ Theorem to a diagnostic test scenario (a disease with 1/1,000 prevalence and a 5% false positive rate yields a ~2% chance of disease given a positive test). A 2013 follow-up showed marginal improvement (23% correct). Medical education, which prioritizes biochemistry over probability, is partly to blame. Abusive lawsuits, cultural pressures for decisiveness, and patient demands for certainty further discourage embracing doubt, as Daniel Kahneman’s work on overconfidence suggests.

Neil Ferguson and the Authority of Statistical Models

Epidemiologist Neil Ferguson and his team at Imperial College London produced a model in March 2020 predicting up to 500,000 UK deaths without intervention. The US figure could top 2 million. These weren’t forecasts in the strict sense but scenario models, conditional on various assumptions about disease spread and response.

Ferguson’s model was extraordinarily influential, shifting the UK and US from containment to lockdown strategies. It also drew criticism for opaque code, unverified assumptions, and the sheer weight of its political influence. His eventual resignation from the UK’s Scientific Advisory Group for Emergencies (SAGE) over a personal lockdown violation further politicized the science.

From the perspective of history of science, Ferguson’s case raises critical questions: When is a model scientific enough to guide policy? How do we weigh expert uncertainty under crisis? Ferguson’s case shows that modeling straddles a line between science and advocacy. It is, in Kuhnian terms, value-laden theory.

The Pandemic as a Pedagogical Mirror

The pandemic was a crucible for statistical reasoning. Successes included the clear communication of mRNA vaccine efficacy (95% relative risk reduction) and data-driven ICU triage using the SOFA score, though both had limitations. Failures were stark: clinicians misread PCR test results by ignoring pre-test probability, echoing the Harvard study’s findings, while policymakers fixated on case counts over deaths per capita. The “6-foot rule,” based on outdated droplet models, persisted despite disconfirming evidence, reflecting resistance to updating models, inability to apply statistical insights, and institutional inertia. Specifics of these issues are revealing.

Mostly Positive Examples:

• Risk Communication in Vaccine Trials (1)
  The early mRNA vaccine announcements in 2020 offered clear statistical framing by emphasizing a 95% relative risk reduction in symptomatic COVID-19 for vaccinated individuals compared to placebo, sidelining raw case counts for a punchy headline. While clearer than many public health campaigns, this focus omitted absolute risk reduction and uncertainties about asymptomatic spread, falling short of the full precision needed to avoid misinterpretation.
  
  
• Clinical Triage via Quantitative Models (2)
  During peak ICU shortages, hospitals adopted the SOFA score, originally a tool for assessing organ dysfunction, to guide resource allocation with a semi-objective, data-driven approach. While an improvement over ad hoc clinical judgment, SOFA faced challenges like inconsistent application and biases that disadvantaged older or chronically ill patients, limiting its ability to achieve fully equitable triage.
  
  
• Wastewater Epidemiology (3)
  Public health researchers used viral RNA in wastewater to monitor community spread, reducing the sampling biases of clinical testing. This statistical surveillance, conducted outside clinics, offered high public health relevance but faced biases and interpretive challenges that tempered its precision.

Mostly Negative Examples:

• Misinterpretation of Test Results (4)
  Early in the COVID-19 pandemic, many clinicians and media figures misunderstood diagnostic test accuracy, misreading PCR and antigen test results by overlooking pre-test probability. This caused false reassurance or unwarranted alarm, though some experts mitigated errors with Bayesian reasoning. This was precisely the type of mistake highlighted in the Harvard study decades earlier.
  
  
• Cases vs. Deaths (5)
  One of the most persistent statistical missteps during the pandemic was the policy focus on case counts, devoid of context. Case numbers ballooned or dipped not only due to viral spread but due to shifts in testing volume, availability, and policies. COVID deaths per capita rather than case count would have served as a more stable measure of public health impact. Infection fatality rates would have been better still.
  
  
• Shifting Guidelines and Aerosol Transmission (6)
  The “6-foot rule” was based on outdated models of droplet transmission. When evidence of aerosol spread emerged, guidance failed to adapt. Critics pointed out the statistical conservatism in risk modeling, its impact on mental health and the economy. Institutional inertia and politics prevented vital course corrections.

(I’ll defend these six examples in another post.)

A Philosophical Reckoning

Statistical reasoning is not just a mathematical tool – it’s a window into how science progresses, how it builds trust, and its special epistemic status. In Kuhnian terms, the pandemic exposed the fragility of our current normal science. We should expect methodological chaos and pluralism within medical knowledge-making. Science during COVID-19 was messy, iterative, and often uncertain – and that’s in some ways just how science works.

This doesn’t excuse failures in statistical reasoning. It suggests that training in medicine should not only include formal biostatistics, but also an eye toward history of science – so future clinicians understand the ways that doubt, revision, and context are intrinsic to knowledge.

A Path Forward

Medical education must evolve. First, integrate Bayesian philosophy into clinical training, using relatable case studies to teach probabilistic thinking. Second, foster epistemic humility, framing uncertainty as a strength rather than a flaw. Third, incorporate the history of science – figures like Codman and Cochrane – to contextualize medicine’s empirical evolution. These steps can equip physicians to navigate uncertainty and communicate it effectively.

Conclusion

Covid was a lesson in the fragility and potential of statistical reasoning. It revealed medicine’s statistical struggles while highlighting its capacity for progress. By training physicians to think probabilistically, embrace doubt, and learn from history, medicine can better manage uncertainty – not as a liability, but as a cornerstone of responsible science. As John Heilbron might say, medicine’s future depends not only on better data – but on better historical memory, and the nerve to rethink what counts as knowledge.

______

All who drink of this treatment recover in a short time, except those whom it does not help, all of whom die. It is obvious, therefore, that it fails only in incurable cases. – Galen