# Adaptive designs in randomized clinical trials: reanalysis of the HOVON87/NMSG18 multiple myeloma trial

## TL;DR

This study reanalyzed a multiple myeloma trial to see if adaptive designs could have detected futility earlier, potentially improving trial efficiency and decision-making.

## Contribution

The paper demonstrates how adaptive trial designs could have accelerated futility detection in a failed phase III trial.

## Key findings

- Using the Pocock spending function, the group sequential design suggested trial termination at the second interim analysis due to futility.
- The sample size re-estimation design indicated continuation without expansion due to an observed HR of 0.86.
- Adaptive designs may reduce follow-up time and improve the efficiency of future randomized trials.

## Abstract

Randomized controlled trials are the gold standard to assess clinical efficacy of novel drugs, but improved survival extends follow-up time and challenges their feasibility. Adaptive trial designs, offering earlier outcome assessments, may enhance efficiency and accelerate decision-making. We retrospectively evaluated whether applying adaptive designs to the randomized HOVON87/NMSG18 phase III trial (inclusion between January 2009 and October 2012), which failed to meet its primary endpoint, could have indicated futility earlier.

We modelled two adaptive trial designs: (1) group sequential and (2) sample size re-estimation. Each design recommends early trial termination for strong efficacy or futility, or continuation if signals are inconclusive. Interim analyses for the group sequential design were defined at 33% and 67% of events required for the final analysis, using O'Brien-Fleming, Pocock, and gamma spending functions to control for false positive or negative results. Sample size re-estimation design was evaluated after 67% of events.

Group sequential design: Hazard ratios (HRs) at both interim analyses were within futility/efficacy boundaries with the more conservative O'Brien-Fleming and gamma spending functions, indicating trial continuation as planned. However, the more aggressive Pocock spending function indicated trial termination at the second interim analysis with a HR of 0.88 exceeding the futility boundary of HR > 0.85. Sample size re-estimation design: At 67% of events, the observed HR of 0.86 was in the unfavourable zone, whereas the promising zone required a HR between ≥0.76 and ≤0.83. The observed HR indicated trial continuation without sample size expansion.

This reanalysis of a MM trial suggests that adaptive designs indicate earlier futility detection and thereby facilitate decision-making. Early signals could reduce follow-up time or potentially offer sample size expansion if the effect is promising but not yet conclusive. Our findings highlight the value of implementing adaptive designs to enhance the efficiency and ethical conduct of future RCTs.

The HOVON87/NMSG18 trial was supported by 10.13039/501100004622Dutch Cancer Society grant 2008-4246, the 10.13039/100008730Norwegian Cancer Society and 10.13039/100006436Celgene.