Skip to main content

Table 2 Performance Evaluation of Known Confounders

From: Machine learning enables detection of early-stage colorectal cancer by whole-genome sequencing of plasma cell-free DNA

Confounder

k-fold CV AUC (95% CI)

k-fold CV Sensitivity at 85% Specificity (95% CI)

Confounder CV method

Confounder CV AUC (95% CI)

Age

0.71 (0.64–0.77)

44% (29–57%)

Binned-age

0.50 (0.50–0.50)

Batch

0.72 (0.69–0.75)

43% (31–53%)

k-batch

0.50 (0.50–0.50)

Processing Date

0.69 (0.64–0.74)

38% (25–49%)

Ordered k-batch

0.48 (0.43–0.52)

Institution

0.87 (0.84–0.90)

74% (72–77%)

Balanced k-batch

0.51 (0.28–0.74)

  1. Performance evaluation of known confounders alone to predict cancer with either k-fold or the CV procedure designed to control for the confounder. Confidence intervals are calculated from bootstrapped distributions of the metric across folds