Interim Analyses of Clinical Trials

Q. What is an interim analysis?
A. Analysing the efficacy data for some patients before the corresponding efficacy data is available for all patients. (Safety data is looked at continuously).

Q. What is the purpose of an interim analysis?

A. To make one of the following decisions:

· terminate for futility, thereby saving time and money

· terminate for overwhelming efficacy. Put all patients onto the treatment immediately if the treatment is so effective (and safe) that it would be unethical to deprive the control group of the treatment any longer.

· continue with unchanged protocol

· continue with protocol changes e.g. dosing amount or frequency. In practice this is equivalent to starting a new trial because regulators will be reluctant to accept evidence from the early part of the trial.

Q. What is FDA Guidance for Interim Analysis?

A. See section IV E. Interim Analysis and Early Stopping in FDA, Statistical Principles for Clinical Trials.

Q. What level of clinical significance is required at an interim analysis?

A. That isn't a statistical question.

Q. What level of statistical significance is required at an interim analysis?

A. A level that makes it likely that the target level of significance (e.g. p = 0.05) will be achieved when all patients have been treated and their results analysed.

Q. Does peeking at the data at an interim analysis reduce the significance finally achieved?

A. Yes, that effect is called "alpha spending". To counteract it, a larger sample size is required than if there were no interim analyses.

Q. If the results at an interim analysis are statistically significant, will they also be at the final analysis?

A. If the patient characteristics are random, it's highly likely that the significance level at the final analysis will be higher than at any of the interim analyses.

Q. Does it matter if the level of statistical significance at an interim analysis hasn’t reached the target level, e.g. p = 0.05?

A. If it's not far above p = 0.05 at the interim analysis, there's a good chance that it will be under p = 0.05 by the time all patients have been treated and their results analysed.

There isn't a formula to convert the significance level at an interim readout into the expected significance level at the final readout. The range of likely outcomes can be modelled using Monte-Carlo methods.

Q. What is the relation between confidence intervals at successive interim analyses?

A. The width of the confidence interval for the estimate of an endpoint is inversely proportional to the square root of the sample size.

Example. Endpoint is % reduction in uPCR.

Totally hypothetical data, not to be used for share trading.

Observed at n =72. 42% reduction in uPCR and 95% confidence interval is 42% plus or minus 5% points

n = 72 +144 = 216. The width of the 95% confidence interval will be 5% points * (72/216) ^0.5 = 2.89% points

So the 95% confidence interval is 42% plus or minus 2.89% points

n = 286. The width of the 95% confidence interval will be 5% points * (72/286) ^0.5 = 2.51% points

So the 95% confidence interval is 42% plus or minus 2.51% points

==================

Further Readings

FDA, Statistical Principles for Clinical Trials

https://www.fda.gov/media/71336/download

Jennison, Interim Monitoring of Medical Trials

https://projecteuclid.org/journals/statistical-science/volume-5/issue-3/Statistical-Approaches-to-Interim-Monitoring-of-Medical-Trials--A/10.1214/ss/1177012099.full

Chang, Futility Stopping in Clinical Trials

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7954786/

A. J. SANKOH, PHD

INTERIM ANALYSES: AN UPDATE OF AN FDA REVIEWER'S EXPERIENCE AND PERSPECTIVE

Drug Information Journal, Vol. 33, pp. 165-176, 1999