Each week, we publish layman’s abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.

 

The article featured today is from Statistics in Medicine, with the full Open Access article now available to read here.

 

 

Individuals differ in how they respond to a given treatment –“one man’s meat is another man’s poison.”Ideally, we would like to give the treatment only to those individuals who are likely to benefit from the treatment and are unlikely to experience adverse effects of the treatment. This heterogeneity in treatment response is well recognized in clinical practice. Hence any simplistic summary from a clinical trial, such as the overall treatment effect, is not directly relevant for treating individual patients. A new treatment might only benefit a sub-population of the patients with certain characteristics, while it shows no benefit to others. Accurate evaluations of this heterogeneity attributable to the variation in baseline patient characteristics may provide many potential benefits in terms of facilitating the decision-making in appropriately targeting existing therapies to the individuals. Therefore, identifying the individual-level predictors of treatment response is extremely important for personalized treatment decisions. Often, there are numerous candidate variables, and the task is to find which ones are predictive of treatment response. A common strategy for studying the treatment response heterogeneity in clinical trial is subgroup analysis, which explores how treatment effect varies across subgroups. However, traditional methods like one-variable-at-a-time subgroup analysis ignore the joint effect of the covariates on treatment effect. They fail to identify significant treatment-covariate interactions when the number of variables is fairly large, which is often the case in clinical trial studies.We have developed a general method that overcomes the various limitations of existing methods to assess heterogeneity of treatment response in clinical trials setting. Our method is able to automatically screen a large number of potential predictive variables, with a good trade-off between false-positives and false-negatives to accurately identify the significant predictors of treatment response.Our method is well-suited for doing secondary analysis in clinical trials to identify predictive covariates and biomarkers.The superior performance of our method is corroborated by an application to a large randomized clinical trial data (N=2,569)investigating a drug for treating congestive heart failure.