Model Development. Post-TRIPOD, the number of studies that included predictors based on significance levels in univariable analysis decreased (pre-TRIPOD: 67%, post-TRIPOD: 44%, Figure 2 and Supplementary Table 8) as well as the number of studies using stepwise methods to retain pre- dictors (pre-TRIPOD: 63%, post-TRIPOD: 48%). In general, a larger number of candidate predictors was used in the post-TRIPOD period (median: 25), compared with pre-TRIPOD period (median: 20). Internal validation was more frequently performed in the post-TRIPOD period (74%) compared with the pre-TRIPOD period (62%). When internal validation was performed, bootstrapping was the most frequently used method in both time periods with an increase from 29% in the pre-TRIPOD period to 41% in the post-TRIPOD period. Chapter 7 The majority of studies presented measures of calibration (pre-TRIPOD: 66%, post-TRIPOD: 87%) and discrimination (pre-TRIPOD: 91%, post-TRIPOD: 100%, Figure 3 and Supple- mentary Table 7).A calibration plot and this increased in the post-TRIPOD period (pre-TRI- POD: 50%, post-TRIPOD: 82%)). Discrimination was primarily assessed with the C-statistic and Area Under the Curve (AUC) methods (pre-TRIPOD: 91%, post-TRIPOD: 100%). Measures of classification were reported in more than half of the studies (pre-TRIPOD: 69%, post-TRIPOD: 58%),mostly assessed with diagnostic test summary statistics (i.e. sensitivity, specificity and positive and negative predictive values) (pre-TRIPOD: 63%, post-TRIPOD: 50%) and to a lesser extent the integrated discrimination improvement (IDI; pre-TRIPOD: 16%, post-TIRPOD: 11%) or the net reclassification improvement (NRI; pre-TRIPOD 25%, post-TRIPOD: 18%).
Appears in 2 contracts