Time-dependent ROC curves in genetically determined subgroups

The ultimate goal of genetic association studies is to identify and map the gene(s) responsible for a given disease. This paper discusses a new and simple statistical method for detecting a genetic association, based on time-dependent Receiver Operating Characteristic (ROC) curves. This method resorts to the Heagerty approach based on Bayes theorem, and uses the Kaplan-Meier or the Akritas estimator. An application to the real problem of examining possible interaction between glycaemia and a risk “genotype” on survival is presented using the Framingham database. Analysis assessed area of chromosome 1 (from 192 to 233 cM) and evaluated the role of fasting blood glucose on survival, at 4 and 8 years of follow-up, according to the presence/absence of allele 242 (marker 23). The allele 242 showed ability in predicting survival. Kaplan-Meier and Akritas estimators provided comparable results. At 4 years of follow-up, area (SD) under ROC curve, in absence of allele 242, was 0.85 (0.09) and 0.81 (0.07) using respectively the Kaplan-Meier or Akritas estimator. The ability showed by the time-dependent ROC curves in real data suggests that the proposed method may be valuable to detect difference in genetic subgroups. Further studies need to better clarify the usefulness of this method in other real applications.