Arrest Outcomes Associated With Outpatient Commitment in New York State

We analyzed officially recorded arrests of individuals attending outpatient clinics in the New York City boroughs of the Bronx and Queens. Participants were recruited for a longitudinal study conducted between January 2003 and January 2006, which focused on self-reported psychosocial outcomes associated with AOT (4). After a complete description of the study, written informed consent was obtained from participants, including consent to conduct searches of records. Institutional review board approval was obtained from the New York State Psychiatric Institute, Bronx Psychiatric Center, Creedmoor Psychiatric Center, Bronx-Lebanon Medical Center, and the New York State Office of Mental Health (NYS OMH).

The sample for this report included 183 participants—86 who were assigned to AOT at some point in their lives and a comparison group of 97 individuals never assigned to AOT.

We used logistic regression estimated via generalized estimating equations (GEE) to compare the odds of arrest during the period of AOT assignment and the six months after it with the odds of arrest before AOT and six months or more after completion of AOT and among individuals never assigned to AOT. Thus our analysis provided both within- and between-person comparisons. GEE was appropriate because it could use all information from individuals who contributed different numbers of months of follow-up and because it could take into account the correlated nature of the responses in the within-person component of the design. Our analysis used an exchangeable correlation matrix in the GEE procedure (12).

We used fixed-effects logistic regression as an ancillary approach because it controls for the stable effects of stable confounding variables even if those confounding variables were not measured (13). The approach removes between-person variance, leaving only within-person variance for analysis. By removing the between-person variance, the effects of stable determinants of individual differences, whether measured or not, are effectively controlled. Because fixed-effects analyses addressed within-person changes in variables (that is, arrest and assignment to AOT), only individuals who experienced changes in these variables could be included in the analysis, leaving 40 persons who contributed 3,815 months of observation. Fixed-effects analyses were conducted only for all arrests combined, because arrests for violent offenses were too rare to produce stable estimates. We compared our estimates of the effect of AOT on arrest from the fixed-effects and GEE analyses to determine whether our conclusion about the effect of AOT was the same in these complementary analytic approaches.

Table 1 is based on data collected in interviews and chart reviews. The sample was mostly male, young to middle-aged, with relatively low levels of education. Reflecting the communities where the clinics were located, the sample was largely African American and Hispanic. The only significant difference between those ever and never receiving AOT was that men were overrepresented in the AOT group (67% versus 54%).

Arrests for individuals never assigned to AOT occurred in 1.2% of the 8,546 months of follow-up. For individuals assigned to AOT, arrests occurred in 2.2% of the 4,182 months before AOT, .9% of the 2,327 months during and six months after AOT, and 1.4% of the 1,835 months occurring more than six months after AOT.

Table 2 shows logistic regression coefficients comparing risks of arrest in the different periods of observation for all arrests and for arrests for violent offenses. As expected, overall arrests were far less common among women, declined with age, and were less common among individuals with higher education. African Americans were more likely to have experienced a contact with police that led to arrest. No differences according to category of psychiatric diagnosis were found. Within-group results showed that the risk of arrest was significantly higher during the period before AOT than during the period of AOT, and although the risk of arrest went up slightly in the period after AOT, this difference was not significant. Moreover, between-group results showed that risk of arrest among individuals who were never assigned to AOT was significantly higher than that of the AOT group while assigned to AOT. Using within-group analyses, we also compared (not shown in Table 2) the period before AOT with the period after AOT, finding the risk in the earlier period to be significantly higher than in the later period. Because arrests for violent offenses were relatively rare, many of the covariates were not statistically significant (Table 2). However, even with small numbers, individuals receiving AOT were at significantly lower risk of arrest than they were before AOT.

Figure 1 portrays results from Table 2 in the form of odds ratios (ORs). As the figure shows, for persons who had ever received AOT, the risk of any arrest was 2.66 times greater (95% confidence interval [CI]=1.58–4.44) and the risk of arrest for a violent offense was 8.61 times greater (CI=1.12–65.97) before AOT than it was while receiving AOT. In addition, compared with individuals during and shortly after the period of assignment to AOT, the comparison group that had never received AOT had nearly double the odds (OR=1.91, CI=1.05–3.48) of arrest. As an additional check, when we included as “receiving AOT” only those months actually assigned to AOT rather than “AOT plus six months after the termination of AOT,” our models showed very similar results.

Fixed-effects analyses supply some degree of control of unmeasured confounding and therefore provided a useful comparison to GEE analyses. Compared with arrests during and six months after participants were enrolled in AOT, arrests before AOT were 2.24 (CI=1.14–4.39) times more likely, a point estimate that was reasonably close to the OR of 2.66 from the GEE analyses. The OR for arrest in the period after AOT indicated a nonsignificant increase compared with the period of AOT (OR=1.82, CI=.94–3.52). This consistency across analyses that differed in strengths and weaknesses (GEE and fixed-effects) increased confidence that AOT was associated with a reduction in arrests.

We found AOT to be associated with a sizeable reduction in the probability of arrest for both overall and violent offenses. Before assignment, individuals who would later be enrolled in AOT were substantially more likely than the same individuals while receiving AOT to experience arrest for both overall and violent offenses. In addition, we found that while individuals were enrolled in AOT the risk of arrest was actually lower than that of a comparison group of individuals never assigned to AOT and that the risk of arrest after AOT ended remained significantly lower than the risk in the period before AOT began.

AOT, as implemented under Kendra's Law in New York State, is a policy that substantially reduces the risk of arrest, including arrests for violent offenses among people with serious mental illnesses. From the vantage point of a general public concerned with violence and who hold prevalent perceptions of dangerousness concerning people with mental illnesses (14,15), this is a very positive and straightforward outcome: Kendra's Law directly results in reduced crime and violence. In addition, the result pushes us to consider a very beneficial trade-off in coercion, with a relatively small exposure under Kendra's Law forestalling a substantially larger exposure delivered by arrest. Our results can be read as an indication that both the general public and people assigned to AOT benefit—the former through a reduction in crime and violence and the latter through a reduction in experienced coercion and all of its untoward consequences.

The idea of a powerful mutual benefit, however, is complicated by the following considerations. First, a majority (53%), of the 86 study participants assigned to AOT were never arrested—not before, during, or after their assignment. It is, of course, possible that AOT forestalled some arrests during and after assignment of people who had never been arrested before assignment. At the same time, given that a majority of the AOT group was never arrested, it is likely that some people assigned to AOT were at extremely small or even no risk of arrest. For these individuals, there was no arrest reduction benefit to procure. It is also true that AOT strives to affect outcomes other than arrest, such as quality of life and reduction in self-harm, and previous research has shown benefits in these other domains (2–5). Still, in considering the trade-off in coercion, we must also consider that some people received the coercion of AOT without the possibility of a benefit in reduced coercion due to arrest. Second, although the coercion embodied in outpatient commitment is less severe than that involved in arrest, an important difference exists in the administration of the two forms of coercion. Arrest, at least in theory, is a specific response to a specific illegal action, whereas outpatient commitment is coercion administered to forestall events—illegal and otherwise—that might happen in the future. From a procedural justice point of view, the two types of coercion are quite different and might reasonably be experienced as such by individuals exposed to them (16,17).

To minimize the influence of out-of-state arrests, we included as follow-up only months relatively close to a time when we knew the individual was living in New York (that is, at time of interview). In addition, we would expect coverage of arrests to be much better during AOT when the individual is mandated to treatment and known to be a New York State resident. As a result, any bias would be toward higher arrest rates during AOT than at other times, but we found just the opposite. Another consideration is that although outpatient clinics in New York City provided a good situation in which to test the effects of AOT, generalization to other populations should be made cautiously.

A strength of our quasi-experimental study is that it included individuals at risk of arrest for violent offenses and observed them for a period totaling more than 1,400 person-years. A limitation of our study is that it was not an experimental study with random assignment to study conditions. Yet the design, the analysis, and the pattern of achieved results increase confidence in our findings for the following reasons. First, one would expect the AOT group to have a greater risk of violence because selection into that group depends, in part, on the risk of violence and arrest. The finding that while participants received AOT their risk of arrest was lower than in the comparison group who had never received AOT runs counter to the bias one would expect in our quasi-experimental design. Second, our study provides results of within-group comparisons (before, during, and after AOT) that are not contaminated by biases associated with nonequivalence between groups. Finally, by controlling the effects of stable confounding variables, the fixed-effects analysis added a stringent test that provided concordant findings. Nevertheless, we caution that an ideal study would include both the analytic and design strengths of our study and random assignment to treatment conditions. As a result, we suggest that readers consider our findings in light of their deviation from such an ideal design.