Chapter 9 208 that expanded access programs harbor these characteristics as typically two types of patients are included. The first category are patients who are excluded from the trial due to their baseline condition, e.g., when they are too frail to participate in a trial, but are nonetheless granted access out of ’compassion’. For these patients, the inclusion/exclusion in the trial will probably be driven by a difference in expected outcomes. The second category consists of patients who would have been eligible for the trial but are ’unlucky’, as a trial is already fully enrolled, or as trial sites are geographically out of reach. Although analyzing data of patients of the first category may lead to insights into the generalizability of treatments, it simultaneously may decrease the precision of the estimate and increase the chance of erroneous decision-making. Including the second category of ’unlucky’, trial-like patients may on the other hand increase precision. Hence, an expanded access program may actually resemble the latent class simulation set-up, and we have shown that our method is able to correctly discriminate between these two classes of patients. Expanded access runs in parallel to ongoing trials, and these data hence form a ’current’ external data source. This distinguishes expanded access data from ’historical’ or ’non-current’ external sources and limits the potential bias due to time trends. Furthermore, other scholars have suggested to explicitly incorporate the ’unmet medical need’ or patient burden in trial design specifications - for example by adjusting the controlled type I error rate in diseases with extremely low survival rates (e.g., glioma).As expanded access by definition is only available for patients with a high unmet medical need,44,45 this additional flexibility could be explored through the use of innovative statistical designs. The similarity between data sources should play a decisive role in whether to integrate external and current data and if so, to what degree. The transparency in hybrid two-stage methods using propensity scores allows one to inspect the balance of covariates across data sets before proceeding with the analysis. As such, it provides a quantitative addition to the qualitative measures suggested by Pocock.4 The availability of a causal interpretation of the estimates, combined with the additional safeguarding in hybrid methods, altogether provides a statistically rational argument to attempt to include expanded access patients into decision-making. The acceptability of evidence synthesized from expanded access data in regulatory decisionmaking remains a topic of debate as these data are used in a qualitative, supportive manner.27,37,46,47 Nonetheless, various regulators have put forth guidance on the (statistical) incorporation on ’real-world evidence’.3,48 In addition to statistical arguments, there are also ethical considerations of incorporating expanded access data: ignoring expanded access data would imply treating patients with investigational medicine without reaping the benefits of additional insights on the safety and efficacy for future patients. Lastly, it denies participating patients the freedom to altruistically advance clinical research. One could therefore argue that more attention should be
RkJQdWJsaXNoZXIy MTk4NDMw