Probabilistic reasoning

Probabilistic Reasoning - The Diagnostic Bet

This approach uses probability to refine a diagnosis. It starts with an initial suspicion (Pre-Test Probability) and updates it using test results via Bayes' theorem, which adjusts the odds of disease.

• Pre-Test Probability (PTP): Baseline chance of disease; often prevalence.
• Likelihood Ratios (LR): A test's power to shift probability.
  • LR+ for positive result: $sensitivity / (1 - specificity)$
  • LR- for negative result: $(1 - sensitivity) / specificity)$
• Post-Test Probability: Updated probability. $Post-test,odds = Pre-test,odds \times LR$.

⭐ Likelihood Ratios >10 or <0.1 cause large, often conclusive, changes in post-test probability.

📌 SPIN/SNOUT

• SP-IN: A highly SPecific test, if Positive, rules IN disease.
• SN-OUT: A highly SNensitive test, if Negative, rules OUT disease.

Bayesian Toolkit - Odds & Ends

Probabilistic reasoning integrates new data to update diagnostic certainty. Key is moving from pre-test to post-test probability.

• Pre-test Probability: The likelihood of disease before a test result is known (i.e., prevalence).
• Post-test Probability: The likelihood of disease after incorporating a test result.

Bayes' Theorem mathematically connects them: $P(A|B) = [P(B|A) * P(A)] / P(B)$.

---

Diagnostic Test Metrics

• Specificity = $TN / (TN + FP)$ → Rules IN (SPIN)
• PPV = $TP / (TP + FP)$
• NPV = $TN / (TN + FN)$

⭐ As disease prevalence increases, PPV increases and NPV decreases. Specificity and sensitivity are intrinsic to the test and do not change with prevalence.

Likelihood Ratios (LR)

• LR+: $Sensitivity / (1 - Specificity)$. For a positive test, how much to ↑ post-test probability.
• LR-: $(1 - Sensitivity) / Specificity$. For a negative test, how much to ↓ post-test probability.

Clinical Application - The Fagan Nomogram

A Fagan nomogram is a graphical tool that simplifies determining a patient's post-test probability of disease after a diagnostic test result.

• Workflow: A straight line is drawn from the pre-test probability through the likelihood ratio (LR) for the specific test result, arriving at the post-test probability.
• It's a visual representation of how new information (the test result) modifies an existing belief (pre-test probability).
• The underlying calculation converts odds:
  • $Post-test,odds = Pre-test,odds \times LR$

⭐ LRs >10 or <0.1 are considered to provide strong evidence to rule in or rule out a diagnosis, respectively.

Cognitive Biases - Mindfield Navigation

• Anchoring Bias: Over-relying on initial data (the "anchor"), like a prior diagnosis, and failing to adjust for new information.
• Availability Heuristic: Judging a diagnosis as more likely if it's easily recalled (e.g., a recent, vivid case).
• Base Rate Neglect: Ignoring the true prevalence of a disease in favor of specific, but misleading, clinical features.
• Confirmation Bias: Favoring information that confirms a pre-existing belief, while dismissing contradictory evidence.

⭐ The availability heuristic can lead to over-diagnosis of rare but memorable diseases recently seen during training or in media.

High‑Yield Points - ⚡ Biggest Takeaways

• Bayes' theorem mathematically links pre-test probability and test results to calculate post-test probability.
• Sensitivity and specificity are intrinsic test properties, independent of prevalence.
• PPV and NPV, however, are critically dependent on disease prevalence.
• As prevalence increases, PPV increases while NPV decreases.
• Likelihood Ratios (LRs) modify pre-test odds to post-test odds, offering a direct measure of test utility.
• A high LR+ (e.g., >10) strongly rules in disease; a low LR- (e.g., <0.1) strongly rules it out.