Mental Health Screening, Treatment, and Institutional Incidents: A Propensity Score Matched Analysis of Long-Term Outcomes of Screening

ABSTRACT

There is little study of the impacts of screening on distal outcomes such as recovery or adverse incidents during incarceration. Furthermore, screening practices vary greatly between jurisdictions; many focus on history taking as opposed to current symptoms.

Method. We conducted an observational cohort study of all admissions (N = 13,281) to Canadian prisons over a 33-month period. We used full matching on propensity scores to explore the association between treatment and health, violent and victimization incidents.

Results. Treatment was associated with lower rates of victimization and violence for all inmates, and health incidents among screen-detected cases. Among inmates with pre-existing risk–who accounted for 90% of health incidents—treatment was associated with increased risk of a first incident but decreased risk of recurrent incidents.

Implications. In-depth symptom-based screening identified a group of inmates who benefited from treatment, but accounted for relatively few incidents. Tensions of prioritizing resources for pre-existing vs, incident mental illness were highlighted by weaker treatment effects for those with pre-existing needs. Increased use of diversion options to keep inmates with mental illness out of prisons, and determining the appropriate intensity of intervention to meet the needs of screen-detected cases could alleviate some of these tensions.

KEYWORDS:

• Screening
• prison
• mental illness
• violence
• self-harm

Acknowledgments

The authors thank Correctional Service of Canada (CSC) for providing the data for this project. CSC had no role in the conduct of this study or the decision/approval to publish. The views are those of the authors, and do not necessarily reflect the position of CSC. Dr. Martin was supported by a Vanier Canada Graduate Scholarship, and Dr. Colman is supported by the Canada Research Chairs program. Dr. Martin was on unpaid education leave from employment with CSC during this study, and has returned to paid employment with the department. All remaining authors have no conflicts of interest to disclose.

Notes

1. An accessible general review on analytic approaches with propensity scores, and their advantages relative to traditional regression adjustment is provided by Austin (2011). Briefly, matching approaches achieve greater balance between treated and un-treated groups than stratification (Austin, 2011; which has been shown to remove approximately 90% of imbalance; Rosenbaum & Rubin, 1984). However, balanced matching (e.g., typically one to one matching) results in excluding participants who do not have a match in the dataset. Full matching on propensity score achieves the benefits of stratification (i.e., retaining all cases) and traditional matching (i.e., greater balance) by matching each treated case to all untreated cases that are similar (i.e., there is not a constant ratio of matched untreated participants to each treated participant; (Austin & Stuart, 2015a). An alternative that is similar to full matching is inverse probability of treatment weighting. This approach can achieve similar reductions in bias to full matching when there are weaker or moderate selection biases at play. However, when there are strong selection biases it can result in more extreme weights that result in poorer performance than full matching (Austin & Stuart, 2015b). The final use of propensity scores is regression adjustment (i.e., including the propensity score as a predictor in the regression model). Regression adjustment increases the risk of model misspecification as it requires a correctly specified model relating the propensity score to the outcome variable (i.e. that all interactions, non-linearity, etc. are correctly specified) in addition to a correctly specified propensity model (e.g. that there are no unmeasured confounders); matching approaches only require that the propensity model is correctly specified (Austin, 2011).

2. The reduction of pre-existing differences can be seen by examining the weighted data for each variable in the right hand side of the table, which shows that after weighting the proportion of inmates receiving each level of treatment was similar in all sub-groups to that of the overall sample (i.e., proportions are often the same in all sub-groups, and never differ by more than 3%). For example, while inmates who screened positive (13%) were more than 6 times more likely to receive chronic treatment than those who screened negative (2%) in the unweighted data, after applying the propensity weights there was the same proportion of inmates who received chronic treatment (8%) regardless of screening results.