To the Editor: Joan W. Ellason, M.A., and Colin A. Ross, M.D., (1) reported significant gains by 54 patients with dissociative identity disorder following 2 years of therapy with the greatest improvement occurring in 12 patients whose personalities became integrated. The authors concluded that "patients with this disorder often respond very well to treatment." We believe this conclusion is unwarranted.

The improvement observed in these patients cannot unequivocally be attributed to treatment. First, the study lacked a standard treatment protocol, or at least clearly identifiable treatment components, making it difficult to accept strong claims about treatment efficacy. Second, because there was no control condition, it is impossible to know how much of the improvement may have been due to nontreatment factors. For example, some spontaneous recovery may have occurred. Regression toward the mean is also a problem; since patients often enter therapy when they are particularly distressed, they are, on average, likely to show less distress in the future, independent of treatment (2). This problem is amplified by the fact that the baseline assessment occurred during inpatient treatment. Since hospital admissions for dissociative identity disorder tend to be for emergencies (3), many subjects were probably experiencing an exacerbation of symptoms at baseline.

The greater improvement of integrated patients also requires qualification. First, integration status was determined by some of the same scales that were subsequently analyzed as outcome measures. After certain criteria are used to differentiate groups, showing that the groups differed on those criteria is tautological. Second, performing a large number of statistical tests on correlated psychiatric scales while failing to control family-wise error rate inflates the probability of finding main and interaction effects involving integration. Third, there may exist an unreported significant difference between the number of medications (mean=4.4, SD=2.3) prescribed at baseline for the 29 nonintegrated subjects and the number of medications (mean and SD not reported) prescribed for eight integrated subjects for whom these data were available. By using data reported in a separate article by Ellason and Ross for a subset of 30 subjects from the same group (4), we computed a significant difference (calculated for unequal group sizes and variances) between the number of medications for the 22 nonintegrated (mean=4.5, SD=1.8) and for the eight integrated (mean=2.3, SD=0.8) subjects (t=4.62, df=28, p<0.001). If this result can be confirmed in the updated group, it, along with the greater level of depression reported in nonintegrated subjects, suggests that a higher level, or different type, of pathology may be predictive of nonintegration and less improvement at follow-up.