Statistical analysis

We evaluated all models by the AUC for 1- to 5-year outcomes. For instance, to compute the 3-year AUC, we considered a mammogram

as positive if it was followed by a cancer diagnosis within 3 years and negative if it had at least 3 years of screening follow-up. Table

S8 describes the distribution of follow-up and cancer times for each dataset. We also calculated Uno’s C-index (59), which offers a generalized AUC across all time points. To address that patients may have multiple examinations, we used a clustered bootstrap approach with 5000 samples to calculate confidence intervals. To assess the significance of the difference between two AUCs, we used the paired DeLong’s test (60) as implemented in the pROC package in R (61). To assess the significance of the difference between two ratios, we

used a two-tailed t test as implemented in R (62). For both tests, we used a predefined P < 0.05 for significance.

Category