Bayes' Theorem

Related Subjects: |Basic Statistics |Sampling in Medical Statistics |Reading a Medical paper |Different Forms of Medical Trials and Studies |Hierarchy of Evidence-Based Trials |Bayes' Theorem

🧠 Bayes' Theorem is one of the most useful but underappreciated tools in medicine. It allows clinicians to combine prior probability (what we think before the test) with new evidence (test results) to reach a more accurate post-test probability. Without it, we risk overestimating what a “positive” test really means, especially in low-prevalence settings.

📐 Bayes' Theorem Formula

P(A |B) = [ P(B |A) × P(A) ] / P(B)

• P(A |B): Posterior probability — the probability of disease after the test result.
• P(B |A): Likelihood — probability the test is positive if disease is present (sensitivity).
• P(A): Prior probability — “pre-test probability,” based on prevalence, risk factors, or clinical gestalt.
• P(B): Marginal probability that the test is positive (true positives + false positives).

🏥 Applications in Medicine

• 🔬 Diagnostic Testing: Bayes explains why positive predictive value (PPV) and negative predictive value (NPV) vary with disease prevalence. A highly sensitive and specific test can still mislead in low-prevalence groups.
• 📊 Screening Programs: Why mammography or PSA screening leads to false positives when applied to the general population — because prior probability (prevalence) is low.
• 🧬 Risk Assessment: In genetic counselling, BRCA mutation carriers start with a high prior probability → a “negative” test still leaves residual risk.
• 💊 Treatment Decisions: In cardiology, deciding whether chest pain is angina: pre-test probability (age, sex, risk factors, typicality of pain) informs whether to order a stress test. A “positive” stress test means little in a young healthy patient but is much more convincing in a 60-year-old smoker with diabetes and exertional pain.

🧮 Worked Example

💡 A disease has prevalence 1% (prior probability). A test has 90% sensitivity and 95% specificity. What is the probability the patient has the disease if the test is positive?

• P(Disease) = 0.01
• Sensitivity = 0.90
• Specificity = 0.95
• P(Test+) = (Sensitivity × P(Disease)) + (1 – Specificity) × (1 – P(Disease)) = (0.90 × 0.01) + (0.05 × 0.99) = 0.009 + 0.0495 = 0.0585
• P(Disease | Test+) = (0.90 × 0.01) ÷ 0.0585 = 0.009 ÷ 0.0585 ≈ 15.4%

✅ Interpretation: Despite a “good” test, the post-test probability is only 15%. This shows why screening asymptomatic populations often leads to more harm than benefit — most “positives” are false alarms.

⚖️ Clinical Pearls for Doctors

• Always ask: What was my pre-test probability? A troponin in a young healthy person means something different than in an elderly smoker with chest pain.
• Likelihood Ratios (LR+ and LR–) are often more intuitive in practice. – LR+ = Sensitivity / (1 – Specificity) → how much a positive test raises odds. – LR– = (1 – Sensitivity) / Specificity → how much a negative test lowers odds.
• A test rarely makes the diagnosis alone. It nudges your probability higher or lower.
• Bayesian thinking prevents overdiagnosis: remember prostate-specific antigen (PSA) testing — high rates of false positives in men without symptoms.
• Medicine is probabilistic, not deterministic: a “rule-out” test lowers probability but never takes it to zero.

⚠️ Limitations

• Pre-test probability: Often subjective, but can be guided by epidemiology or validated tools (e.g. Wells score, Centor criteria).
• Real-world accuracy: Sensitivity/specificity may vary across settings and populations.
• Independence assumption: Bayes assumes independence, but comorbidities and overlapping risks complicate reality.

📌 Summary

Bayes’ Theorem is not just maths — it is the foundation of diagnostic reasoning. It teaches us:

• 🔎 No test is perfect — context matters.
• 📊 A test’s value depends on prevalence and pre-test probability.
• 🩺 Doctors must combine test results with clinical judgment.