Skip to main content


Fig. 4 | BMC Medical Research Methodology

Fig. 4

From: Multi-cohort modeling strategies for scalable globally accessible prostate cancer risk tools

Fig. 4

Medians and 95 percentile intervals (2.5 to 97.5 percentile) for comparing the AUC, negative of HLS, and net benefit at the 15% threshold between the five possible prediction methods (numbering according to Table 1: 1-Pooled data, cohort ignored; 2-Pooled data, cohort as random effect, median prediction; 3-Pooled data, cohort as random effect, mean prediction; 4-Meta-analysis, fixed effects by center; 5-Meta-analysis, random effects by center) computed across all 252 choices of five cohorts as test sets with the remaining cohorts as training sets. Positive differences indicate superiority of prediction method listed first for the respective operating characteristic

Back to article page