Medical Decision Making

Diagnostic Tests - Numbers Game

• Basis: 2x2 table (Disease vs. Test: TP, FP, FN, TN).
• Sensitivity (Sn): $TP / (TP+FN)$. Detects disease.
  • 📌 SNOUT: Sensitive test, Negative result, rules OUT.
• Specificity (Sp): $TN / (TN+FP)$. Confirms absence.
  • 📌 SPIN: Specific test, Positive result, rules IN.
• PPV (Positive Predictive Value): $TP / (TP+FP)$. Prob. disease if test +ve. Varies with prevalence.
• NPV (Negative Predictive Value): $TN / (TN+FN)$. Prob. no disease if test -ve. Varies with prevalence.
• Likelihood Ratios (LR): Link pre- to post-test probability.
  • LR+: $Sn / (1-Sp)$. >10 strong for disease.
  • LR-: $(1-Sn) / Sp$. <0.1 strong against disease.

⭐ Prevalence ↑ → PPV ↑, NPV ↓.

Decision Thresholds - Choice Points

• Critical probability points guiding decisions: withhold action, test, or treat.
• Test Threshold ($T_{test}$):
  • If P(Disease) < $T_{test}$: No further testing or treatment. Disease considered ruled out.
• Treatment Threshold ($T_{treat}$):
  • If P(Disease) > $T_{treat}$: Initiate treatment empirically. Disease considered ruled in.
• Test-Treat Zone:
  • If $T_{test}$ ≤ P(Disease) ≤ $T_{treat}$: Optimal to perform diagnostic testing.
• Influencing factors:
  • Disease: Prevalence, severity.
  • Test: Accuracy (Sn, Sp), risks, cost.
  • Treatment: Efficacy, risks, benefits, cost.

⭐ The treatment threshold ($T_{treat}$) is lowered by ↑disease severity, ↑treatment efficacy, or ↓treatment risk.

Clinical Scores & EBM - Evidence Edge

• Clinical Scores: Objectify risk/diagnosis (e.g., Wells' for DVT/PE, CURB-65 for pneumonia severity).
  • Metrics: Sensitivity $TP/(TP+FN)$, Specificity $TN/(TN+FP)$.
  • PPV $TP/(TP+FP)$, NPV $TN/(TN+FN)$.
  • Likelihood Ratios: $LR+ = Sens/(1-Spec)$; $LR- = (1-Sens)/Spec$.
• EBM: Integrates best evidence, clinical expertise, patient values.
  • PICO: Patient, Intervention, Comparison, Outcome.
  • Hierarchy (Top to Bottom): Meta-analysis/Systematic Review → RCT → Cohort → Case-Control → Case Series → Expert Opinion.
  • Key EBM stat: Number Needed to Treat (NNT) = $1/ARR$.
  • 📌 FINER criteria (Research Q): Feasible, Interesting, Novel, Ethical, Relevant.

⭐ Odds Ratio (OR) is used in case-control studies; Relative Risk (RR) in cohort studies. For rare diseases ($prevalence < 10%$), $OR \approx RR$

Cognitive Biases & SDM - Mind Matters

• Cognitive Biases: Systematic thinking errors impairing clinical judgment.
  • Anchoring: Over-relying on initial info.
  • Availability: Overestimating based on easily recalled examples.
  • Confirmation: Seeking info confirming existing beliefs.
  • Premature Closure: Accepting diagnosis early, missing alternatives.
  • Framing Effect: Decisions influenced by info presentation.
  • Mitigation: Metacognition, checklists, second opinions.
• Shared Decision Making (SDM): Clinicians and patients collaboratively make healthcare decisions.
  • Involves discussing options, evidence, benefits/risks.
  • Aligns care with patient values and preferences.
  • Improves satisfaction and adherence. 📌 Remember "BRAN": Benefits, Risks, Alternatives, Nothing.

⭐ SDM is not just an option but an ethical imperative in patient-centered care.

High‑Yield Points - ⚡ Biggest Takeaways

• Bayes' Theorem is crucial for post-test probability calculation using likelihood ratios.
• Understand Sensitivity (Sn), Specificity (Sp), PPV, and NPV for test interpretation.
• ROC curves help compare diagnostic tests; Area Under Curve (AUC) quantifies overall accuracy.
• Decision analysis often employs decision trees to model choices and potential outcomes.
• The threshold model (test/treat thresholds) guides diagnostic and therapeutic actions.
• Recognize common cognitive biases (e.g., anchoring, availability) that can affect clinical judgment.
• Apply Evidence-Based Medicine (EBM) principles, respecting the hierarchy of evidence for decisions.