Background: Deleterious DPYD variants are the primary identified cause of severe fluoropyrimidine-related toxicity. The objective was to improve the sensitivity of current DPYD genotyping recommendations and develop a predictive model for severe toxicity. Methods: This pooled analysis included colorectal cancer (CRC) patients from 7 prospective studies. Full DPYD sequencing was performed. Depending on their minor allele frequency (MAF), relevant DPYD variants were identified using Bolasso, or SKAT coupled with in silico functionality predictions. The primary end-point was 12-week grade 4-5 fluoropyrimidine-related hematological and digestive toxicities. Multivariate logistic regression models were developed. Results: Of 4496 eligible patients, 3437 were analyzed (267 events). The final model included patient/treatment characteristics, variants �2A/�13/p.D949V, common variants rs1801160 (p.V732I �6, OR ¼ 1.93, 95% CI ¼ 1.37 to 2.70 if heterozygous) and rs2297595 (p.M166V, OR ¼ 1.49, 95% CI ¼ 1.13 to 1.96 if heterozygous), and aggregation of all very rare (MAF<0.01) in silicopredicted deleterious variants (OR ¼ 3.20, 95% CI ¼ 1.43 to 7.14 if one allele). Model performance was supported by discrimination (AUC¼0.76, 95% CI ¼ 0.73 to 0.78), calibration, and internal validation. Crude sensitivity of DPYD testing increased from 11.2% for �2A/�13/p.D949V to 47.5% when including the 44 variants. An exploratory analysis of 1138 stage III CRC patients revealed, for the first time, that carriers of deleterious DPYD variants exhibited significantly longer overall survival: HR¼0.65, 95% CI ¼ 0.43 to 0.98 for p.V732I and HR ¼ 0.62, 95% CI ¼ 0.43 to 0.90 for p.M166V. Conclusion: DPYD sequencing identified additional relevant variants and improved the sensitivity of DPYD testing. An online calculator (https://fluoropyrimidine-toxicity-predictor.gustaveroussy.fr/) is provided to estimate the individual probability of developing life-threatening toxicity, based on clinical covariates and extended DPYD genotype.
A Sat, study studied this question.