课件：流行病学4诊断试验.ppt

Interpreting the information Incidence without treatment (InoT) Incidence with treatment (IwthT) RRR=(Inot-IwthT)/InoT ARR=InoT-IwthT NNT=1/ARR 9.6% 2.8% 71% 6.8% 14.7 96% 28% 71% 68% 1.47 0.096% 0.028% 71% 0.068% 1470 Interpreting the information (2) RRR if you want to sell an intervention AR and NNT if you have to decide on purchasing an intervention Steps in building 2×2 tables 1. Put a N (100;10,000 or 100,100) in bottom right cell 2. Use prevalence to calculate D+ total, and by subtraction D- total 3. Use sensitivity to fill D+T+ cell 4. Use specificity to fill D-T- cell 5. Compute other cells totals 6. Calculate PPV and NPV 7. Interpret ‘Screen for’ gonorrhoea in three different clinics with different prevalence Preval(%) PPV NPV Pos LR Neq LR 0.5 4.60% 100.00% 9.5 0.1 5 33.30% 99.70% 9.5 0.1 50 90.50% 94.70% 9.5 0.1 Systematic screening vs. diagnosis Do you know many GPs who would systematically ‘screen’for gonorrhoea? In clinical, ‘diagnostic’ decision making (an exercise in confirmation and exclusion of hypotheses,based on suggestive symptoms), Se, Spec and LR are equally (if not more) useful as/than PV (the prevalence of G among patients with urethral Discharge probably is quite high!) Alternative to 2×2 tables Bayes’ theorem PV = P (D+) × P (T+/D+) P (D+) × P (T+/D+) + P (D-) × P (T+/D-) = Pr × Ss Pr × Ss + (1– Pr ) × (1– Sp ) Decision trees * 0.72 cured = 0.2412 * 0.1 die = 0.0094 * 0.9 survive = 0.0844 * 0.28 not cured * 0.67 complete treatment * 0.05 heal * 0.1 die * 0.85 survive = 0.0082 = 0.0165 = 0.1402 * 0.33 incomplete treatment 0.5 detected * 0.05 heal * 0.2 die * 0.75 survive = 0.0250 = 0.1000 = 0.3750 0.5 not detected N Cases Disease + - + 18 40 58 Test -