A Power Simulation Guide for Choosing Longitudinal Endpoints and Analysis Methods for Continuous Data

Background: The choice of analysis method and primary endpoint can have a major impact on the results of a clinical trial. The power of these analysis methods is affected by the response patterns, response distributions, autocorrelations, missing values, and the number of times the endpoint is evaluated. Guidelines are needed to help investigators select statistical methods that give investigators the best chance of finding an intervention effect if one truly exists. Methods: Monte Carlo methods were used to determine how different endpoints are affected by changes in study assumptions. Copula simulation methods were applied to prior Alliance hot flash and CIPN studies to determine which endpoints would provide the highest power for future studies. Results: Longitudinal analyses had higher power if there was a plateau response, there were fewer missing values, responses were collected more often, and there was lower correlation between responses from the same subject. Having skewed data did not greatly affect power. Prior hot flash and CIPN studies showed skewed data, plateau response rates, and very high autocorrelations. Conclusion: Hot flash and CIPN studies are more likely to find true effects if the analysis method uses data at all of the time points and adjusts for baseline. Collecting data at more time points and reducing the missing data rates also increase study power.