How many measurements for time-averaged differences in repeated measurement studies?

Area of Science:

• Biostatistics
• Longitudinal Study Design
• Sample Size Calculation

Background:

• Traditionally, the number of repeated measurements in longitudinal studies is treated as a fixed design parameter.
• Statistical considerations can inform the optimal number of repeated measurements, impacting study efficiency.

Purpose of the Study:

• To investigate the effect of the number of repeated measurements on the required sample size in longitudinal studies.
• To evaluate different correlation structures and their influence on sample size determination.

Main Methods:

• Simulation studies were conducted to assess sample size requirements under varying numbers of measurements per subject.
• The compound symmetry (CS) and autoregressive (AR(1)) correlation structures were analyzed.
• The impact of measurement error on the AR(1) model was also investigated.

Main Results:

• Under compound symmetry (CS) correlation, increasing measurements per subject consistently decreases the required sample size.
• Sample size reduction diminishes significantly (less than 5%) beyond four measurements per subject with CS correlation.
• The AR(1) correlation structure exhibits a counterintuitive property where more measurements may increase sample size requirements.
• Introducing measurement error into the AR(1) model eliminates this counterintuitive behavior, leading to reduced sample sizes with more measurements.

Conclusions:

• The number of repeated measurements is a crucial design choice that significantly impacts sample size in longitudinal studies.
• Practitioners should exercise caution when assuming the AR(1) model due to its potential for counterintuitive sample size implications.
• Incorporating measurement error into models can reconcile the relationship between measurement frequency and sample size requirements.