Incorporating predictions of individual patient risk in clinical trials

Abstract

A risk prediction model is a statistical technique that gives a predicted probability of a certain event for an individual patient. Prediction models outperform the traditional risk classification systems that work by assigning patients into risk groups based on the presence or absence of particular risk factors, such as stage of disease. As such, risk prediction models have a number of important possible uses in clinical trials. For Phase II studies, prediction models can help adjust comparisons with historical control groups for differences in case mix. For Phase III studies, prediction models can ensure that accrued patients are at sufficiently high risk. This improves statistical power and avoids unethical inclusion of low-risk patients. We also propose that prediction models could potentially be used for applying the results of Phase III trials to individual patients. Clinical decisions could be informed by individualized estimates of treatment benefit, rather than by average treatment effects.