Postlude 254 However, it is important to note that clinical trials have different objectives, which often do not revolve around achieving random sampling or generalizability.8 Note that our novel statistical methodology similarly avoids the focus on generalizability: we expand the trial analysis with expanded access data from the patients that are similar to the trial patients.9 Clinical trials are designed to produce specific outcomes, to conduct an experiment, and careful patient selection is essential to this objective. So, if clinical trials do not primarily address selection bias or generalizability, what is their main focus? Trials are meant to yield estimates of comparisons - relative effects - which do not provide direct information on absolute effects in populations outside of trials.9 Patient characteristics are controlled to reduce variation in the estimate of relative effects. Although it is common and expected that subgroups experience different absolute effects (e.g., men versus women, young versus old), relative effects remain surprisingly stable across subgroup analyses.5,10 The premise of real-world data relies in part on the misunderstanding that trials would yield absolute, rather than relative effects, and that these relative effects would not be generalizable to populations outside the trial. For regulatory purposes, establishing relative effects through randomized clinical trials is usually sufficient. However, relative effects may not be the primary interest for reimbursement bodies,11,12 which require absolute effects to calculate health expenses or budget impacts, or for patients, who may prioritize absolute effects. Modelling can subsequently be used to translate relative effects in absolute terms for groups that were not included in the trial, potentially aided by data of patients not in the trial.13 It is precisely here where the benefits of real-world data come into play: to aid, or to yield estimates of parameters that cannot be estimated from the trial without bias (e.g., treatment adherence).14 Confirming this reasoning, our empirical results have shown that reimbursement bodies hence more frequently employ data from expanded access programs than regulators.15 Beyond binary thinking: multifaceted perspectives A recurring concept in the literature on expanded access is that ‘Expanded access is treatment, not research’.7,16,17 Indeed, the primary intent of expanded access is to provide a treatment option for patients in need. However, this persistent ‘treatment versus research’ mantra presents a false dichotomy. Can methods providing treatment not simultaneously facilitate data collection? With the unprecedented growth of electronic health data, a plethora of ‘real-world data’ sources initially designed to furnish treatment - such as patient charts, electronic health records, and claims and billing databases - can now be used for research purposes. Despite not being originally intended for research, these sources are increasingly being harnessed for scientific ends.
RkJQdWJsaXNoZXIy MTk4NDMw