Sample size determination for constrained longitudinal data analysis.

The longitudinal data analysis model proposed by Liang and Zeger (Sankhyā: Indian J. Stat. Ser. B 2000; 62:134-148) uses the baseline as well as postbaseline values as the dependent variables, and the baseline mean responses are constrained to be the same across treatment groups due to randomization. Compared with the conventional longitudinal analysis of covariance, this approach can correctly estimate the variance of within-group mean changes and achieve the specified coverage probabilities. General results on the sample size and power calculations for this model in the presence of missing data are obtained. The sample size relationship between the constrained and unconstrained longitudinal data analysis is established. Simple expressions for sample size calculation are obtained for the compound symmetry and first-order autoregressive correlation structures. The sensitivity of the sample size requirement to the configuration of correlation structure and retention pattern is assessed. The performance of several ad hoc approximations for longitudinal data analysis sample size calculation is evaluated. Simulation studies are conducted to assess the validity of the proposed sample size formulas with deviation from normality. The sample size formulas are also illustrated in detail using real clinical trial data.