Effect of antipsychotic withdrawal on extrapyramidal symptoms: statistical methods for analyzing single-sample repeated-measures data.

Using symptom severity ratings of extrapyramidal side effects (EPS) during a 4-week antipsychotic washout period, we illustrate particular problems associated with repeated measures of symptom severity and demonstrate four analysis methods. The often suggested analysis of variance and multivariate analysis of variance found no mean change in weekly Simpson Angus scores over the 4-week washout despite the fact that 43% had clinically significant EPS prior to drug discontinuation. On the other hand, the Friedman Analysis of Ranks and Cochran-Mantel-Haenszel (CMH) statistics found significant change over the washout period. These two less well-known techniques place fewer restrictions on the data, can be more sensitive to patterns of change, and may be more appropriate for psychiatric data. The CMH method is particularly attractive since it does not require complete data on all subjects as do many other techniques. This minimizes the number of cases lost to missing data and increases the generality of the results.