Skip to content
Scintria
ToolsGrant writerFunding
Start free
ScintriaStart free
Simulation Lab

Adaptive trial (interim analyses)

Look at the data part-way through and stop early if the answer is already clear. See the boundaries that keep the false-positive rate honest, the chance of stopping early, and the sample size you can expect to save.

Boundary

More options

Stop early for futility?

Significance level (alpha)

Power

79%

A fixed design of the same size has 80%.

Expected size per arm

148

If the effect is as hoped (maximum 175); 175 if there is no effect.

Chance of stopping early for efficacy

44%

False-positive rate

5.1%

Final boundary

z = 2.00

Stopping boundaries

012341231.96Analysisz

When trials stop, if the effect is as hoped

  • Analysis 13%
  • Analysis 240%
  • Analysis 336%

Teal: stopped for efficacy. Amber: stopped for futility. The rest reach the final analysis without crossing.

For your protocol

The trial will use a group-sequential design with 3 equally spaced analyses (2 interim and one final) and O'Brien–Fleming-type efficacy boundaries (z = 3.47, 2.45, 2.00; nominal two-sided p = 5.2e-4, 0.014, 0.045), preserving an overall two-sided alpha of 0.05. With a maximum of 175 participants per arm and a standardised difference of 0.3, simulation of 20,000 trials gives 79% power (80% for a fixed design of the same size) and an expected sample size of 148 per arm if the effect is as hoped and 175 per arm if there is no effect.

How this simulation works

The trial's test statistic is followed as information accrues (a Brownian motion with drift). At each equally spaced look the z-statistic is compared with the efficacy boundary: O'Brien–Fleming boundaries are very strict early and close to the usual 1.96 at the end; Pocock boundaries are the same at every look. The constants come from the published tables for a two-sided alpha of 0.05 or 0.01. The optional futility rule stops when the trend favours control; it is non-binding, so it does not change the efficacy boundaries. Power, stopping chances and the expected sample size are estimated from 20,000 simulated trials.