The Centre for Addiction and Mental Health (CAMH) is a 530-bed psychiatric hospital located in Toronto, Ontario, Canada. It includes a forensic program with 190 inpatient beds and approximately 300 community forensic patients serving the Greater Toronto Area with a population of approximately 2.4 million people living in the catchment area. All forensic patients found to be unfit to stand trial (UST) or not criminally responsible (NCR) are under the jurisdiction of the Ontario Review Board (ORB). They undergo a formal review at least on an annual basis, and a comprehensive report is provided to the ORB before it hears evidence and gives its written disposition with reasons. Starting in the 1970s, reports prepared for annual ORB hearings of CAMH inpatients and outpatients and the disposition following the hearings have been collected in a registry. Typically, these forensic reports consist of 6 to 20 pages of text summarizing the patient’s clinical, diagnostic, personal, and legal history. For each report in the registry, a trained research analyst extracted and entered 101 variables into a database (e.g., gender, age, marital status, number of times in locked seclusion) organized into 10 sections (e.g., Demographics, Personal and Developmental History, Course in Hospital/Community While on Disposition Order). For the current study, the completion of analyses using this forensic case registry was approved by the CAMH Research Ethics Board before the start of this analysis.

For this study, we selected the most recent ORB report for all cases coded and entered in the database between April 20, 2012, and August 28, 2017, after excluding 35 patients with a diagnosis of cognitive disorder or dementia. The resulting sample comprises 510 forensic patients who came under the jurisdiction of the ORB between September 18, 1990, and May 12, 2012. At the time of their most recent ORB report, all were 18 years or older; 191 were inpatients on a unit of CAMH Forensic Program, 18 were inpatients on another unit of CAMH, 293 were outpatients, two had absconded, and the current setting was not indicated for six.

We extracted data from the patient’s most recent ORB report using methods previously described (21, 22). In summary, two senior researchers in the forensic program at the hospital where data were collected developed a coding scheme that contained all relevant demographic, clinical, legal, and risk variables of interest, as well as their operational definitions (protocol available through corresponding author). Variables were coded from psychiatric reports and ORB Reasons for Disposition reports. These reports typically contain comprehensive information about patients, including psychosocial history (e.g., childhood, relationships, education, employment history), criminal history (usually obtained from police records), mental health (e.g., diagnoses, hospitalizations), and risk assessments. Observations made during previous and current hospitalizations are recorded in these reports. Two research analysts (one master’s level; one with a bachelor’s degree) were trained on how to use the coding form and the Historical, Clinical and Risk Management -20 (HCR-20). During the training phase, there was a good interrater reliability between each research analyst and each of the two senior researchers for all variables (intraclass correlation coefficient, single measure [ICC1] >0.75; κ coefficient for categorical variables >0.75). Following the establishment of rater reliability, research analysts coded cases independently, and interrater reliability during data collection was again assessed by having the senior researchers code 20% of all cases coded by the research analysts. Again, ICC1 and κ values were always greater than 0.75, as suggested by Fleiss (23).

Diagnoses for participants were extracted from the clinical files and were recorded by the attending psychiatrist responsible for the patient’s care and made according to Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria (24). All clinical diagnoses that had been recorded by the attending psychiatrist were extracted for further analysis.

Legal status of the participants falls under the auspices of the ORB, which has two categories, namely, NCR or UST.

The index offense(s) includes the offense(s) for which there is an NCR or unfit finding.

Charges refer to the recorded offense for which the individual was found criminally responsible or UST.

Victim in the index offense(s) was defined as anyone who was considered to be directly harmed (physically or psychologically) by the patient during the offense.

Victim’s relationship with the participant was categorized as being family (spouse, parent, sibling, children, extended family), copatient or roommate, or care professionals (anyone who works with the patient within the institutional/healthcare setting). A stranger was defined as a victim who was completely unknown to the person.

Locked seclusion in a clinical setting is when a patient is placed in a room designed for this purpose with doors locked to manage risk of harm either to themselves or others. Coding for locked seclusion was based on at least one event occurring during the past year.

Restraints are either pharmacological or mechanical. Coding for restraints was based on at least one such event occurring during the past year.

Conflict with staff or copatients was coded in the registry as present when there was “general argumentativeness” or “not getting along with others.”

Difficult behavior was codes as “difficult to manage from the point of view of staff.” Examples include unruly behavior and noncompliance with unit rules.

Victimization was coded as “being bullied, picked on, insulted, verbally or physically assaulted.”

Suicidal ideation refers to presence of plans or thoughts of killing oneself.

Homicidal/violent ideation refers to plans, ideas, thoughts, desires, or urges about killing or harming someone.

We used descriptive statistics (frequencies, means, standard deviations, and percentages) to summarize sociodemographic, clinical, and forensic characteristics of ID and non-ID groups. χ² Analyses were used to compare ID and non-ID groups for categorical variables, and Fisher exact tests were used where cell sizes for expected values were five or less. We calculated Mantel–Haenszel odds ratios (ORs) and 95% confidence intervals (CIs) as an effect measure. Pearson r was used for comparing continuous variables. All analyses were conducted using Stata 14.2 (25).

The dataset included 510 participants, 47 of whom (9.2%) had a diagnosis of ID. Four had diagnoses of both ID and autism spectrum disorder (ASD). Three had a diagnosis of autistic spectrum disorder only and were excluded from the analyses as they formed a small group clinically distinct from both the ID and the non-ID groups. Table 2 compares the sociodemographic and clinical comorbidity between the ID and non-ID groups. The majority of participants were males in both groups. Compared to the non-ID participants, ID participants were less likely to have completed at least 12 years of education (OR, 0.47; 95% CI, 0.29–0.77), less likely be married (OR, 0.26; 95% CI, 0.11–0.59) or employed (OR, 0; 95% CI, 0–0), and more likely to have a paraphilic disorder (OR, 5.35; 95% CI, 1.72–16.58).

Table 3 summarizes the forensic characteristics of participants: compared to non-ID participants, ID participants were younger at the time of committing their index offense by about 4 years (P = 0.03); they were more likely to have committed a sexual offense (OR, 3.21; 95% CI, 1.46–7.03); 95% CI, the victims of the index offense were more likely to be professionals involved in their care (6.21; 95% CI, 2.54–15.17); and they were more likely to have been found UST and less likely to have been found NCR (OR, 3.73; 95% CI, 1.73–8.04) and more likely to be in a secure unit (OR, 0.74; 95% CI, 0.58–0.93).

Table 4 describes behavioral incidents among inpatient participants. There was little difference in most behavioral variables or clinical outcomes across most areas, including no difference in need for restrictive care practices.

This study examined the diagnosis of ID in a complete population of all patients in a large forensic hospital using a comprehensive number of measures. Notably, there are very few studies outside Europe that have studied this population. The findings have significant implications for clinicians, clinical leaders, and policymakers about needs of this specific subgroup of patients presenting with forensic issues.

There are a number of limitations of this study. First, the past-year behavioral incidents recorded in Table 4 were collapsed into binary variables, which recorded whether there was at least one occurrence of the behavior of interest. There is likely to have been a wide variation in frequencies within these variables, but we were unable to investigate any relationship between ID and severity of the behavioral problem. Our approach was necessary as precise counts of the occurrence of each variable were not available to us, but further studies to investigate frequency and severity of behavioral problems are recommended. Second, although our data extraction methods have been validated, there is a possibility that incidents may be underreported, particularly less serious incidents, as they may not have been included in the narrative reports from which we extracted data. This underreporting is probably as likely in the ID and non-ID groups and therefore unlikely to introduce bias when carrying out group comparisons. However, overall, we may have underestimated the proportion of people having engaged in these disruptive behaviors. Third, in our analyses, we have carried out multiple group comparisons, which increases the risk of obtaining type I errors. One approach would be to have carried out Bonferroni adjustments; however, the risk of type II errors is inflated, and the approach is not widely recommended (35). Although the risk of type I errors remains, we have instead reported ORs with CIs to aid interpretation of the findings. Fourth, the study involves data collected from the forensic division of one Canadian psychiatric hospital. Therefore, the findings may not generalize to other hospitals or settings. Fifth, standardized screening tools were not used on the entire population, so significant comorbidities such as attention-deficit/hyperactivity disorder were not consistently sought. Lastly, as this was a cross-sectional study, we could not assess and compare longitudinal outcomes in the two groups.