Background
Methods
Optimizing Cooling Trial
Study design
Depth of cooling | |||
---|---|---|---|
33.5 °C | 32.0 °C | Margin | |
Duration of cooling | |||
72 h | 45 % | 30 % | 37.5 % |
120 h | 30 % | 25 % | 27.5 % |
Margin | 37.5 % | 27.5 % |
Ethical considerations
Trial monitoring
Interim analysis using frequentist approaches
Depth of cooling | |||
---|---|---|---|
33.5 °C | 32.0 °C | Margin | |
Duration of cooling | |||
72 h | 7 % (7/95) | 14 % (13/90) | 11 % |
120 h | 16 % (15/96) | 17 % (14/83) | 16 % |
Margin | 12 % | 16 % |
What could a Bayesian monitoring approach add to an interim analysis?
Bayesian monitoring of the Optimizing Cooling Trial
Component | Specification | Example |
---|---|---|
Prior distributions | • Previous studies on the control rate or treatment effect can be used as prior information • Prior beliefs about the treatment effect should be elicited from experts to inform the strength of the evidence needed to convince them | Two-arm trial of Treatment A versus B (control): • Evidence from three previous trials on rate of outcome under treatment B: 17 %, 25 %, 30 % • Evidence from two studies on treatment effect for different population: RR 0.98 (95 % CI: 0.73–1.3); RR 0.75 (95 % CI: 0.56–1.0) Prior distributions, center (95 % CrI): • Control rate: 25 % (5–55 %) • Skeptical prior for treatment effect: RR 1.10 (0.7–2.0) • Enthusiastic prior for treatment effect: RR 0.85 (0.5–1.0) |
Clinically important treatment effect | • Investigators should specify how big a treatment effect needs to be in order to stop a trial and recommend its use or advise against it | • A relative risk reduction of 15 % or more is needed to recommend treatment A, RR < 0.85 • An absolute increase of 2 % in safety outcome would be unacceptable, RD > 0.02 |
Stopping thresholds | • For each type of monitoring, i.e., safety, efficacy, or futility, the level of confidence to stop the trial early needs to be specified • For most cases, it should be based on a clinically important effect • Efficacy: the trial will stop early if the likelihood of seeing a clinically important effect is very large, even under a skeptical prior • Futility: if the likelihood of a clinically important effect is small even under an enthusiastic prior, the trial would stop early • Safety: the trial would stop if the probability of increasing harm is large enough under an enthusiastic prior | At any preplanned interim analysis, any of these occurrences would make the DSMC consider stopping the trial: • Efficacy under skeptical prior: Pr(RR < 0.85) > 0.99 • Futility under enthusiastic prior: Pr(RR < 0.85) < 0.10 • Safety under enthusiastic prior: Pr(RD > 0.02) > 0.70 |
Bayesian model
Prior distributions
Neutral priors
Enthusiastic priors
Interim monitoring for futility
Interim monitoring for safety
Implementation details
Results
Marginal comparisons of longer cooling and deeper cooling
Posterior distributions for three intervention group comparisons
RR posterior median (95 % credible interval) | Evidence of any benefit Pr(RR < 1.0) | Futility monitoring Pr(RR < 0.90) | ||||
---|---|---|---|---|---|---|
Neutral | Enthusiastic | Neutral | Enthusiastic | Neutral | Enthusiastic | |
32.0 °C for 72 h | 1.23 (0.76–1.92) | 1.19 (0.74–1.87) | 20 % | 25 % | 10 % | 13 % |
33.5 °C for 120 h | 1.31 (0.82–2.09) | 1.27 (0.80–2.03) | 13 % | 16 % | 6 % | 8 % |
32.0 °C for 120 h | 1.60 (0.82–2.97) | 1.50 (0.79–2.83) | 8 % | 11 % | 4 % | 6 % |
RD posterior mean (95 % credible interval) | Futility monitoring Pr(RD > 0.01)a
| Safety monitoring Pr(RD < −0.05)b
| ||||
---|---|---|---|---|---|---|
Neutral | Enthusiastic | Neutral | Enthusiastic | Neutral | Enthusiastic | |
32.0 °C for 72 h | −0.02 (−0.08, 0.03) | −0.02 (−0.08, 0.04) | 11 % | 15 % | 19 % | 16 % |
33.5 °C for 120 h | −0.03 (−0.09, 0.02) | −0.03 (−0.09, 0.03) | 8 % | 9 % | 28 % | 25 % |
32.0 °C for 120 h | −0.06 (−0.15, 0.03) | −0.06 (−0.15, 0.03) | 5 % | 7 % | 61 % | 54 % |