To the Editor: Although the randomized controlled trial of involuntary outpatient commitment reported by Marvin S. Swartz, M.D., et al. (1) is extremely important, we were surprised that it was published in its present form. Some fundamental principles of reporting randomized controlled trials have been violated.

The primary outcome measure was rehospitalization. The result of the study was negative (as was the result of the only other randomized controlled trial of these patients, in New York [2]): there was no difference between the patients in involuntary outpatient commitment and the control subjects. This key result was the only one based on an intention-to-treat analysis, and it was the only unbiased finding of the article. Despite this, it merited only three sentences in the Results section (with no presentation of data) and none in the Discussion section. Instead, the authors spent two pages reporting a complex post hoc analysis based on the duration of involuntary outpatient commitment experienced by the patients. The conclusion from this analysis was that a sustained period of involuntary outpatient commitment was effective when combined with intensive treatment. However, Figure 1 seemed to show differences in hospitalization between groups of patients in short-term and long-term involuntary outpatient commitment long before the first 6 months had elapsed. This suggests that the effect had nothing to do with the treatment received but instead reflected a bias in which involuntary outpatient commitment was selectively extended when it seemed to be helping a patient. The analysis presented loses all the advantages that randomization offers in abolishing selection bias and confounding (3).

The authors made much of a subgroup analysis based on diagnostic categories. This was reported without presentation of the interaction between treatment condition and diagnosis and should be seen as a spurious post hoc analysis. Also unacceptable was the absence of a CONSORT diagram (4) showing how many patients were excluded from the initial cohort of those awaiting a period of court-ordered outpatient commitment and why.

Finally, we make a suggestion concerning a clinically helpful statistic for reporting results of a randomized controlled trial, especially where the treatment is long-term, possibly has serious negative effects, and is ethically contentious. This is the "number needed to treat" (5), the number of patients who need to be given the experimental treatment to achieve one extra good outcome compared to their having received the control treatment. For example, if the results reported were an unbiased comparison, and the proportion of patients in prolonged involuntary outpatient commitment with readmission in the study year was 32% compared with 48% of those not in involuntary outpatient commitment, the number needed to treat would be about six. That is, six patients would require long-term involuntary outpatient commitment to prevent one from being admitted at least once during the year. Since the researchers found no difference between the two treatment conditions, the number needed to treat was presumably far more—or possibly a negative value, indicating potential harm associated with the intervention.

The results of this trial, conducted by an eminent group of investigators, are likely to be widely quoted and have important clinical, legal, and political implications on both sides of the Atlantic. It is disappointing that recent advances in randomized trial methodology (3, 4) have been overlooked by both the authors and your editorial board.