Current extensive genetic research into common complex diseases, especially with the completion of genome-wide association studies, is bringing to light many novel genetic risk loci. These new discoveries, along with previously known genetic risk variants, offer an important opportunity to improve health care. We describe a method of quickly evaluating these new findings for potential clinical practice by designing a new predictive genetic test, estimating its classification accuracy, and determining the sample size required to verify this accuracy. We illustrate the approach for the case of Type 2 diabetes. By incorporating recently discovered risk factors into the proposed test, we find a potentially better predictive genetic test. The area under the ROC curve (AUC) of the proposed test is estimated to be higher (AUC=0.671) than for the existing test (AUC=0.580).
If, in the design stage, the proposed test appears to be superior to existing tests, or if it reaches a desired accuracy level, it may be worth further developing for clinical use. For that purpose, we also propose a robust and powerful method of developing predictive genetic tests based on real data. The approach is derived based on the optimality theory of the likelihood ratio. Such theory simply shows that the receiver operating characteristic (ROC) curve based on likelihood ratios is the best. Through simulations and real data application, we compare the proposed approach with the commonly used logistic regression and the classification tree approach. These three approaches have a similar performance if we know the underlying disease model. However, for most of the common disease, we have little prior knowledge of the disease model, in which situation the new approach has the advantage over the logistic regression and the classification tree approaches.
While the latter two approaches address the classification accuracy of the test by using the ROC curve, the third approach we propose focuses on the meaning of the test results in terms of predictive values. The approach is derived based on the concept of the TDT. Through simulation, we show that the predictive value estimated from the extended TDT approach approximates the true values well.