One critical shortcoming of many clinical therapeutic trials is their focus on short-term effects, leaving unanswered questions about how participants fare in the longer term. Trial extension studies are conducted to address this gap, allowing assessment of both durability of treatment benefit (eg, whether disease remission or improvements in function are maintained or diminished over time) and long-term safety (such as incidence of delayed adverse events).1 In therapeutic trials, extension studies can mitigate the ethical burden associated with discontinuation of potentially effective treatments or long-term administration of placebo or inferior interventions among those initially assigned to the control group.2 The prospect of an extension study may also give participants greater incentive to enroll in trials, despite concerns about being assigned to the control group in the initial phase of the research. Because extension studies build on existing trial infrastructure, they also allow additional and potentially valuable information to be derived from trials that already require significant up-front investment.
Despite these advantages, several key methodological challenges arise in the design and analysis of trial extension studies. The evidence from these studies is generally framed in the context of the original clinical trial, lending greater credibility to the quality of the outcomes data as well as any causal inferences about longer-term effects. However, epidemiologically, long-term extension studies are essentially cohort studies and, in many cases, uncontrolled cohort studies given the lack of any comparison group. Thus, when longer-term trial extension studies attempt to draw conclusions about the effects of study interventions, either favorable or unfavorable, the reader is often left wondering “Compared with what?”
Some extension studies follow up participants from only the original experimental treatment group. For example, in a study examining the long-term outcomes of intermittent paliperidone palmitate in adults with schizophrenia, investigators provided rare long-term data about outcomes after a trial in a population that is challenging to study3; however, participants who relapsed, died, or withdrew from the initial trial due to adverse events were excluded from the extension. As the investigators acknowledged in JAMA Network Open,3 this effectively created an enhanced cohort study with survivor bias and without a comparison group, limiting the statistical inferences that could be made. Investigators could not determine whether participants would have had similar outcomes even if they had not continued to be exposed to the intervention.
