Abstract
The analysis of schizophrenia studies is plagued by inefficiency and bias due to much missing data. Mixed-effect models for repeated measures designs help address these problems, but to gain even more efficiency it is desirable to judiciously use additional longitudinal data in such designs by comparing treatment groups over multiple time points. Simulations were conducted to compare a profile analysis approach to other commonly used analysis methods in the presence of data missing at random. One gains efficiency by using a composite contrast over multiple time points when the treatment effect over the time points is not substantially different.
Notes
*Rejection rate for an interaction test (e.g., test of parallelism) from weeks 3–6 conducted at α = 0.25 and two-sided in nature.
+Rejection rate for comparison of TD-H and PBO conducted at α = 0.025 and one-sided.
#A cross-sectional analysis at just week 6 using ANOVA.
&Profile analysis approach (preliminary test for interaction followed by either a test over weeks 3–6, or week 6).