ABSTRACTPurpose.  Statistical models for predicting hematologic toxicity were evaluated based on UGT1A1 polymorphisms and baseline serum bilirubin.Methods.  Blood DNA samples were collected from 113 patients with untreated metastatic colorectal cancer receiving irinotecan (FOLFIRI, n = 36; mIFL, n = 41; CapeIRI, n = 36). The primary endpoint was absolute neutrophil count nadir during first treatment cycle. Linear regression models, with increased R2 implying important additional predictive power, sequentially added age, sex, baseline bilirubin level, and UGT1A1 genotype.Results.  All models demonstrated low R2, suggesting unaccounted variables. UGT1A1 genotype added ∼8–9% during cycle 1 and from ∼7% [mIFL regimen] to 26% [CapeIRI regimen] after cycle 1. Correlation between genotype and overall ANC nadir without regard to treatment was low (R = −0.201, P = 0.035). Patients with genotype 7/7 may have increased risk for severe neutropenia, but data are insufficient to characterize this. Contribution of baseline bilirubin level was negligible.Conclusions.  Ability of UGT1A1 or baseline bilirubin to predict neutropenia is low and depends on regimen.