Bayes' Theorem Example: Drug Testing 🌿

Bayes' Theorem Application: Drug Testing
📌 Bayes' theorem is crucial for computing inverse probabilities or solving inverse problems, like estimating a hard-to-measure value from an easy-to-measure one, such as disease screening or drug testing.
🧪 The initial example assumed a drug test is 90% accurate (true positive rate = 0.9, true negative rate = 0.9) and 10% of the population uses the drug ($P(User) = 0.1$).
📉 In this first scenario, the probability of actually being a user given a positive test was only 50% ($1/2$), highlighting issues caused by rare events and test accuracy.
⚠️ If a screening test yields low certainty (e.g., 50%), follow-up testing on the positive group is necessary, especially for high-stakes roles like pilots or surgeons.

Advanced Test Accuracy: Sensitivity and Specificity
🔬 A more complex example involved a test that is 90% sensitive (true positive rate for users) but has a 20% false positive rate for non-users ($P(\text{Positive}|\text{Non-User}) = 0.2$).
📉 With the 10% usage rate, this worse test resulted in the probability of being a user given a positive test dropping further to only 33% ($1/3$).
🛑 A test that is 100% sensitive (always testing positive) is useless for inference because it ignores specificity (the false positive rate among non-users).
📊 High accuracy can be misleading in Machine Learning (ML) when training on rare events; high overall accuracy can be achieved simply by never predicting the rare event, leading to poor detection of what you actually want to find.

Key Points & Insights
➡️ For events that are relatively rare (low prior probability), highly sensitive AND highly specific tests are required to avoid completely botching the inference.
➡️ Test accuracy must be considered alongside prevalence (how rare the event is in the population) and the trade-off between sensitivity and specificity.
➡️ A low probability of a positive outcome given a rare event means you will get many false positives, undermining the predictive power of the test score.
➡️ Consider using sequential testing strategies to update probability estimates iteratively based on successive test results.

📸 Video summarized with SummaryTube.com on Mar 04, 2026, 09:28 UTC