To the Editor: Drs. Szmukler and Hotopf’s comments on our trial of involuntary outpatient commitment raise some important issues that we attempted to address with appropriate caveats within Journal space constraints. As they point out, the results of certain statistical tests contrasting rehospitalization outcomes for patients assigned to outpatient commitment and control subjects over the entire study period were not significant—but that is hardly the whole story. The distribution of hospital outcomes precluded analysis with conventional t tests. Hence, we examined the odds of any hospital readmission using repeated measures analyses and demonstrated that for any 30-day period, the outpatient commitment group had a significantly lower risk of readmission than did the control group (odds ratio=0.64, 95% confidence interval=0.46–0.88, p≤0.01; p. 1972). This analysis used a rigorous test, incorporated all the data for the intention-to-treat group, and considered only the initial randomized group membership—ignoring the duration of the outpatient commitment order. In short, our evidence for a significantly lower risk of readmission in the outpatient commitment group overall did not depend on postrandomization analyses.

As we showed, the lower risk of readmission was much stronger among subjects receiving longer outpatient commitment orders. In our view, failure to report the significant association between the duration of outpatient commitment and the reduced risk of readmission would have neglected an important clinical and policy-relevant finding: that sustained periods of outpatient commitment are more effective. Furthermore, we argue not that more than 180 days of outpatient commitment is a meaningful threshold per se but, rather, that more days of outpatient commitment combined with more intensive treatment improve hospital outcomes. Differences between the subjects in long-term and short-term outpatient commitment began to show up early—long before 6 months—because outpatient commitment orders can expire (or be renewed) within days or weeks after discharge.

We recognize the potential for selection bias in that subjects at highest risk for relapse might be selected differentially for extended outpatient commitment. Indeed, unpublished data have suggested that subjects at higher baseline risk for hospitalization were more likely to receive prolonged outpatient commitment. Therefore, if our results are biased, they are quite probably biased against finding an effect for extended outpatient commitment.

Drs. Szmukler and Hotopf did not see that we did test the interaction between treatment condition and diagnosis in staged, multivariable, repeated measures analyses. The addition of this interaction term resulted in significant improvement in model fit (–2 log likelihood increment in fit=4.83, p≤0.05; p. 1972). Finally, they suggest that we should have reported the "number needed to treat," which we believe could be misleading because it does not account for other nonhospital outcomes associated with outpatient commitment.

The premise of Dr. Segal’s critique of our analysis is that time at risk for hospitalization is an inverse function of the duration of outpatient commitment. According to this premise, subjects who received 6 months of outpatient commitment were only at risk for hospitalization during the remaining 6 months of the study year, i.e., when they were not in outpatient commitment. In fact, our research subjects remained at risk for hospitalization each day of their outpatient commitment. Unlike in Dr. Segal’s parole analogy, the outpatient commitment days did not have to be consecutive; the total period in outpatient commitment could have been interrupted by hospitalizations, and hospital readmission did not have to curtail the total period of outpatient commitment. Indeed, early hospital readmission was, for some patients, an occasion for receiving a renewal of outpatient commitment for an extended period, thereby placing them on the path to eventually spending more than 180 days in outpatient commitment.

Perhaps it would have been clearer to present our finding as a comparison of subjects who received a renewal of their initial outpatient commitment order and subjects who did not. The resulting analysis would have been the same, because the clinical decision to renew the initial outpatient commitment order almost always occurred before 6 months. Therefore, the subjects who did not receive a renewed outpatient commitment order spent 180 or fewer days in outpatient commitment, whereas the subjects who did get a renewed order spent more than 180 days. The decision for renewal might have been biased for other reasons, but it did not affect time at risk for hospitalization.

Hypothetically, if a patient were to have spent a large number of days in the hospital during the year—adding up to more than 6 months—this would have precluded the same patient from spending more than 6 months in outpatient commitment, which would have introduced a selection bias in receiving brief versus extended outpatient commitment. Empirically, however, the hospital stays were of short duration, with control subjects spending a mean of 27.9 days hospitalized during the year. Dr. Segal’s suggested reanalysis would not change our findings.

Outpatient commitment is a highly controversial policy, and, clearly, the data from any study on this topic will be interpreted in a highly politicized atmosphere. We believe that a careful and dispassionate reading of our published findings will help inform this important policy debate.