Genetics versus clinical risk scores for melanoma prediction.

Abstract

Cutaneous melanoma is a common cancer, for which risk stratification has been proposed to aid early detection. To assess the performance of a polygenic risk score (PRS) in predicting the risk of invasive cutaneous melanoma, alone and combined with clinical risk factors. The PRS was derived from the most recent genome-wide association study meta-analysis of cutaneous melanoma, which included 28 849 patients with melanoma and 78 922 control participants from 20 studies from the UK, the USA, Australia and Europe. Then, it was tuned in the Canadian Longitudinal Studying on Aging cohort, which included 528 patients with melanoma and 17 787 control participants. The PRS was then tested independently against 14 self-reported clinical factors and a clinical prediction model (the MP16 model) in the QSkin prospective cohort, which included 16 282 participants with genetic data aged 40-69 years at baseline, among whom 359 were identified with new invasive melanomas during 10 years of follow-up. The PRS outperformed any other single clinical risk factor in QSkin (c-index 0.643). The baseline risk model (age, sex, first 10 principal components) had a c-index of 0.603; adding PRS to the baseline model increased the c-index to 0.670 [likelihood ratio test (LRT) P = 1.68 × 10-21]. Adding PRS to the MP16 model significantly enhanced discrimination [MP16 c-index: 0.713; MP16 + PRS c-index: 0.729 (median LRT P = 5.56 × 10-11)] and predicted more true cases in the first and second top deciles (111 for MP16 + PRS vs. 104 for MP16 in the first decile, and 72 for MP16 + PRS vs. 63 for MP16 in the second decile). Across various screening thresholds (10-50%), the sensitivity and/or specificity is higher and the net reclassification improvements were in the range of 0.01-0.05, comparing the MP16 + PRS model with the MP16 model. Incorporating genetic risk information into existing clinical risk tools significantly improves prediction performance for melanoma.