PSU4 VALUE of Information for Adaptive Trials: Proof of Concept Study in MULTI-Arm MULTI-STAGE Trials of Interventions for the Prevention of Surgical Site Infections

Use value of information methods to compare the economic value to the UK National Health Service (NHS) of an adaptive trial design with a conventional design on interventions for the prevention of surgical site infections (SSI). We developed a decision tree to compare the cost-effectiveness of simple dressing, tissue adhesive as a dressing, and no dressing for the prevention of SSIs following surgery from the perspective of the UK NHS. Relative effectiveness of preventing SSI was based on a network meta-analysis of 22 trials. We used the decision tree to estimate the expected value of sample information (EVSI) of a conventional 3-arm trial comparing simple dressing, no dressing, and tissue adhesive in 2500 patients with 2:2:1 allocation ratio. This trial had a 5% type-1 error rate and 10% type-2 rate. We compared this with a 3-arm 2-stage trial with an interim futility calculation on the first 1250 patients deciding whether to drop the no dressing and/or tissue adhesive arms. We used a genetic algorithm to choose the optimal futility bounds for this design while preserving type-1 and type-2 error rates. Optimality was defined as the design minimising the average of the expected sample size (ESS) under the global null hypothesis and the maximum possible sample size. A lower ESS indicates fewer patients unnecessarily randomized and a reduced trial cost. We calculated the EVSI of this optimal design. The conventional design with 2500 patients had a per-person EVSI for of £112 for the NHS. The optimal design had a per-person EVSI of £108 and an ESS of 1753 under the global null hypothesis, representing a roughly 30% reduction in required patients. Adaptive trial designs, and specifically multi-stage designs, can achieve similar EVSI to conventional designs while exposing fewer patients to unnecessary randomization and with lower cost.