By George Giles
March 31, 2025
Readers of LRC have long been treated to dissertations from the Midwestern Doctor that run counter to the medical 'Conventional Wisdom'. I myself suffered through many years of seeing poor quality medical research. I was hired by an elite private medical center in 1999 specifically to fix the Clinical Trials software purchased from another elite private medical center but were unable to deploy because it did not work. It was my introduction to research in the medical world. Disclaimer: I do not now and never had have a phd, but I am good at applied mathematics and do know how to make computer systems work. I have spent a fair portion of my professional life in research. This was the background I brought to medicine; math, computers and total lack of healthcare knowledge beyond finding my doctor's office. What I learned in the first fifteen years was eye opening and not in a good way.
Jakob Bernoulli was an 18th century Swiss mathematician that first recognized that statistical correlation is not causation. It was named after him Bernoulli's Fallacy. There is a superb book on this topic on Amazon; well worth the $30 cost. Here is a great example of the correlation issue: all criminals breathe oxygen thus anyone that breathes oxygen is a criminal. Clearly bad logic. The correlation is criminals and breathing oxygen; the illogic is equating breathing oxygen with criminality. What Bernoulli was really saying is look at the data and draw conclusions from it. This is also known as Bayesian statistics. Not doing this is the province of speculation a fancy word for guessing.
I soon learned that modern medicine has 'biostatistics'. After a little research I realized why: mathematical statistics is a rigorous discipline based upon proof. The diamond hard concept that all of math, physics and engineering are based upon. Biostatistics are not rigorous mathematics in fact just the opposite. It is solutions in search of problems. It is literally the reification of Bernoulli's Fallacy that Naomi Wolf, a Rhodes Scholar at New College Oxford, is a superb author but may not have a mathematical bone in her body nonetheless her book The Pfizer Papers is literally all about how correlation isn't sound science by any stretch of the imagination. In the case of COVID-19 bad science equates with bad outcomes. Outcomes are the only criteria any medical treatment or drug must be measured against using Bayesian Statistics not what the Congress, a Big Pharma CEO, their Board's of Directors and marketing department dreamed was true.
Medical biostatistics is literally the worst offender when it comes to Bernoulli's Fallacy. It's riddled with p-hacking aka data dredging, publication bias, and overreliance on frequentist methods, which often treat statistical significance as proof of causation rather than just as an observation.
The Problem: Correlation ≠ Causation
- Most medical research relies on observational studies, which inherently suffer from confounding variables.
- Even randomized controlled trials (RCTs) are often misinterpreted because they don't prove causation; they just reduce bias in correlation measurements.
- P-values are the worst culprits. If p < 0.05, researchers treat results as "significant," ignoring Bayesian probability and effect sizes.
Example: The Hormone Replacement Therapy (HRT) Disaster
- In the 1980s and 1990s, studies showed that women on HRT had lower rates of heart disease.
- Doctors assumed HRT prevented heart disease and prescribed it widely.
- Later RCTs showed no protective effect-the correlation was due to wealthier, healthier women choosing HRT (confounding factor).
Modern Biostatistics Still Falls for This
- COVID-19 Studies: Many studies used relative risk reduction (RRR) instead of absolute risk reduction (ARR), misleading the public about vaccine efficacy.
- Nutrition Studies: Coffee is bad for you. Wait, no, coffee is good for you. Nutritional epidemiology is a dumpster fire of correlation-based claims.
- Genetic Research: GWAS (Genome-Wide Association Studies) link genes to diseases, but they rarely prove a causal mechanism.
The Fix? Bayesian Thinking
- Instead of p-values, use Bayesian credible intervals-but frequentist stats dominate because they are easier to publish and explain.
- Mendelian randomization tries to fix correlation issues, but it's not foolproof.
- More RCTs, fewer observational studies-but funding biases favor quick, cheap correlations.
This is my opinion, buyt based on my experience modern medical research is not based upon Bayesian probability and statistics thus any results from biostatistic's might be taken with a grain of salt, possibly even the whole shaker's worth. When you understand Bernoulli's Fallacy and the difference between statistical significance and the errors inherent in using correlation as causation we come upon the gross problem with so much of medical research.
Researchers have a disquieting tendency to find what they were looking for. The window they see through when looking is the correlation/causation dichotomy. President Trump is a lot of things but I would bet all my bitcoin that he is not a mathematician. President Biden might not have even been conscious during 2021 and it is for this reason that bad science and ruthless marketing led to billions of taxpayer dollars going straight into Big Pharma's pockets while potentially creating millions of cripples around the world for a non-solution to a non-problem. That is where the data leads.