Positive predictive value
Of the patients a model flags as positive, the share who actually have the condition. It depends on how common the condition is, so the same model has different predictive value in different populations.
A screening tool for referable diabetic eye disease reports 90% sensitivity and 90% specificity from a specialist eye clinic where 30% of patients have the condition. There, per 1,000 patients it flags 270 true cases and 70 false ones, so about 79% of flags are real (270 of 340). Moved into primary care, where 3% have it, the same tool flags 27 true cases and 97 false ones per 1,000, and only about 22% of flags are real (27 of 124). Before rollout, ask for the PPV at your own prevalence and size ophthalmology referral capacity to match.
Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.
Sign in →