A BAYESIAN ADAPTIVE TWO-STAGE DESIGN FOR PEDIATRIC CLINICAL TRIALS.

We develop a novel two-stage Bayesian adaptive trial design for pediatric settings which borrows information from previously completed trials in adults to support establishing substantial evidence of efficacy for the pediatric population in situations where information extrapolation from adults is justifiable. At the time of the stage I analysis, the extent of information borrowing from adult data is determined by assessing compatibility of the observed pediatric data with its prior predictive distribution, derived using the adult trial data. At this time, the trial may be stopped for futility, enrollment may be stopped (with ongoing patients followed up for primary outcome ascertainment), or enrollment may proceed into stage II to reach a prespecified maximum sample size. We provide guidance on how practitioners can approach answering the question "How much information should be borrowed?" through balancing use of the adult data (when compatible with the pediatric data) with the need to ensure the design leads to reasonable recommendations regarding key actions that might be taken regarding the trial (e.g., when to stop early for efficacy). Type I error control is considered secondary to these considerations as type I error rate inflation above typical levels is unavoidable in these settings. We illustrate how the method can be applied using the Pediatric Lupus Trial of Belimumab Plus Background Standard Therapy as motivation.