Optimizing Study Designs to Better Inform Individualized Treatment Decisions
Jeremy Goldhaber-Fiebert, PhD
A paradigm shift is occurring in medicine and public health. Previously, trials were intended to identify the intervention with the greatest expected effect for a group of patients at risk for or diagnosed with a given disease or condition. As new and better interventions were identified this way, practice shifted towards the superior interventions. However, even if an intervention is best on average, some individuals may do better than expected while others may not do as well. Similarly, even if all individuals do about the same, some may have more serious adverse events/harms/unintended effects and others may have fewer. If differences in outcomes and harms with each intervention can be predicted for individuals, then the choice of intervention can be tailored and individualized – each person can get the intervention that most ideally achieves better health outcomes while minimizing harms for that person in particular. This is the new paradigm of individualized care, personalized medicine and health. Individualized care holds great promise, but achieving this promise presents a number of challenges. One particular challenge that we focus on is when to individualize care based on current knowledge (information from completed studies) and when and how to design additional studies that should be conducted before care is individualized. Currently most trials focus on showing effects for the group as a whole. After the trial concludes, additional analyses attempt to predict which subgroups will benefit more from which treatment. Even when these analyses find differences, they may be highly uncertain because the original trial was not made large enough to precisely measure these differences at these subgroup levels. The subgroup findings are suggestive, and it may be tempting to individualize treatment based on them. Yet, because treatments carry both the promise of benefits and risks of harms and side-effects, additional studies may be warranted. But when? And how large of a study? And on which subgroups should the new study focus on? Towards answering these questions, we describe a framework we developed. We then apply this framework using simulation model examples to characterize what are the characteristics of the subgroups and their expected benefits, risks and associated uncertainties as well as the maximum available sample size that determines optimal study designs for individualization decisions. In many cases, we consider optimal study designs diverge strongly (but predictably) from proportional random sampling schemes like those currently used in many randomized trials.