Step 1: Pick a framing
Step 2: Enter a 2×2 table
| Disease + | Disease − | Row total | |
|---|---|---|---|
| Test + | 100 | ||
| Test − | 900 | ||
| Col total | 90 | 910 | 1000 |
Results
How to read this
Sensitivity & Specificity are properties of the test itself — they don't depend on how common the disease is.
PPV & NPV depend on prevalence. The same test applied to a low-prevalence population yields a lower PPV (more false positives among the positives).
Likelihood ratios combine the above and don't change with prevalence. Rules of thumb (Jaeschke, Guyatt & Sackett, 1994, JAMA 271(9):703-707): LR+ > 10 is strong positive evidence, LR+ 5-10 is moderate, LR+ 2-5 is weak; LR− < 0.1 is strong negative evidence, LR− 0.1-0.2 is moderate, LR− 0.2-0.5 is weak.
RR, OR, NNT are effect-size measures for the cohort-study framing. RR compares the risk in exposed vs unexposed; OR the odds. NNT is the inverse absolute risk difference — how many patients you need to treat (or expose) to see one more event.
CIs: proportions use the Wilson score interval (Wilson, 1927); RR and OR use log-Wald CIs with Haldane-Anscombe correction for zero cells.