Analysis of multiple-period group randomized trials: random coefficients model or repeated measures ANOVA?

Area of Science:

• Statistics
• Clinical Trials
• Biostatistics

Background:

• Linear mixed models in parallel group randomized trials (GRTs) can treat time as continuous (random coefficients, RC) or categorical (repeated measures ANOVA, RM-ANOVA).
• Previous guidance favored RC models over RM-ANOVA for GRTs with >2 periods due to better type I error control.
• This recommendation was based on assumptions that may not hold for all data structures and covariance matrices.

Purpose of the Study:

• To evaluate the performance of different analytic models in multiple-period parallel GRTs.
• To investigate the impact of time-by-group variation on type I error rates.
• To compare RC models, RM-ANOVA (with standard and unstructured covariance), and saturated models.

Main Methods:

• Simulated continuous outcomes for cohort and cross-sectional parallel GRT data under RM-ANOVA and RC mechanisms.
• Assumed time-by-group variation in all simulations.
• Applied RC, RM-ANOVA (with unstructured covariance), and saturated analytic models, with and without specifying time-by-group random effects.

Main Results:

• RC and saturated models maintained nominal type I error rates across all simulated datasets.
• RM-ANOVA with unstructured covariance did not prevent type I error inflation with cohort RC data.
• Analytic models omitting time-by-group random effects showed substantial type I error inflation when such variation was present.

Conclusions:

• Random coefficients (RC) and saturated analytic models are recommended as the default for analyzing multiple-period parallel GRTs.
• Properly modeling time-by-group random effects is crucial for controlling type I error rates.
• These findings provide updated guidance for statistical analysis in longitudinal clinical trials.