Interpreting Incremental Value of Markers Added to Risk Prediction Models

摘要: The discrimination of a risk prediction model measures that model's ability to distinguish between subjects with and without events. area under the receiver operating characteristic curve (AUC) is popular measure discrimination. However, AUC has recently been criticized for its insensitivity in comparisons which baseline performed well. Thus, 2 other have proposed capture improvement nested models: integrated continuous net reclassification improvement. In present study, authors use mathematical relations numerical simulations quantify offered by candidate markers different strengths as measured their effect sizes. They demonstrate increase depends on strength model, true lesser degree On hand, only size variable correlation predictors. These are illustrated using Framingham incident atrial fibrillation. conclude AUC, improvement, offer complementary information thus recommend reporting all 3 alongside characterizing performance final model.