Untargeted plasma metabolomics and risk of colorectal cancer—an analysis nested within a large-scale prospective cohort

Table 2 Balanced error rates for random forest analysis of cases versus controls for samples collected at varying time intervals prior to diagnosis

	< 5 years	5–9 years	10–15 years	> 15 years	All samples
Balanced error rate^a,b	0.43	0.49	0.50	0.54	0.46

Potential confounders (body mass index, smoking status, education level, diabetes, alcohol intake, and recreational physical activity) were included in the models, but none was selected in the built-in variable selection step. ^aBalanced error rate for a two-class problem with expected BER by chance of 0.50. ^bSince there is a 1:1 match between cases and controls (i.e., the data are perfectly balanced), the overall error rate is equal to the balanced error rate

ISSN: 2049-3002