Statistics for Personalized Medicine: Exploratory Subgroup Analyses in Clinical Research

Statistics in Personalized Medicine: Exploratory subgroup analyses covers the issues of subgroup analyses from a practical and a theoretical/methodological point of view. The practical part introduces the issues using examples from the literature where subgroup analyses led to unexpected or difficult to interpret results which in actual fact have been interpreted differently by different stakeholders. Most applications discussed will be from the area of clinical studies and drug development but similar thoughts should be applicable to other areas where decision making is involved.

One of the main issues with subgroup analyses is that results are sought from smaller patient population. The risk to overlook effects increases in this case if the smaller sample size is not counteracted by smaller variability in subgroups. On the other hand, repeated tests for effects can increase the risk to detect artefacts, i.e., results that occur by chance. Some of these issues may be covered by pre-specification, but clearly not entirely. In the drug development process, regulators may tend to interpret data driven results on safety differently from those on efficacy.

On the technical side Statistics for Personalized Medicine addresses selection and selection bias, variance reduction by borrowing information from the full population in estimating a subgroup effect. To this end, subgroup analysis will be linked to statistical modelling, and subgroup selection to model selection. This connection makes the techniques developed for model selection applicable to subgroup analysis.