Seclusion and restraint in adult forensic psychiatric services

In Canada, individuals may be detained in a forensic mental health hospital for two main reasons: 1. For short-term evaluation and treatment to assess their fitness to stand trial or determine if they had a mental disorder at the time of the alleged offense, possibly exempting them from criminal responsibility; and 2. For long-term rehabilitation and reintegration after the court finds them unfit to stand trial (UST) or not criminally responsible due to a mental disorder (NCRMD) [41]. While some forensic patients may be admitted to units with minimum security (i.e., open or locked door, limited unaccompanied time off unit), many begin their forensic hospitalization under medium security (i.e., locked door with restricted time off unit) or high security (i.e., perimeter security, locked door, restricted movement within the facility).

Among adult patients in forensic settings, there is established evidence that younger adults are more likely to be secluded [6,42–44]. Some clinical and behavioral characteristics are also consistent risk factors, including violent and aggressive behavior [6,43,45] and psychiatric diagnoses of schizophrenia and other psychotic disorders [6,42–44], personality disorders [43,44], and substance use disorder [42]. Less extensive research indicates an association with mania, cognitive impairment, lack of insight, and medication non-compliance [7]. There is inconsistent evidence regarding gender differences [6,7,43]. Further, some additional administrative and behavioral characteristics that are risk factors for coercive measures in general psychiatric settings have not yet been examined in forensic settings, including elopement attempts, legal status (e.g., NCRMD vs. UST), and recent hospital admission, which is a time of intensive assessment in forensic hospitals and a prime opportunity to assess and mitigate risk factors.

Most research on coercive measures in forensic settings concerns seclusion, not restraint [6], and relies on bivariate designs rather than multivariable analyses that can tease out the strongest risk factors to inform the most effective interventions. Moreover, no previous studies have analyzed a broad spectrum of clinical, demographic, and administrative factors that can be used in the crucial first few days of forensic admission to identify at-risk individuals for preventative interventions. In addition, there have been calls for increased attention to the experience of racialized and Indigenous individuals in forensic healthcare systems [46–48] in light of reported racial bias within correctional and forensic systems whereby Indigenous people [49,50] and racial minorities [6,43,49] may be disproportionately subjected to seclusion and restraint.

The present study aimed to identify and validate factors associated with seclusion and restraint that can be assessed upon admission, using a large sample of men and women admitted to forensic hospitals in Ontario, Canada. Specifically, we examined the relative contributions of previously established demographic and clinical risk factors, as well as inconclusive or understudied factors, in predicting the use of coercive measures within the first three days of admission. We hypothesized that younger age, diagnoses of psychotic disorder, substance use disorder, and personality disorder, along with violent and aggressive behavior, would be significant predictors. Additionally, we examined the role of additional demographic (male sex, immigration background, Indigeneity), administrative (legal status), and clinical factors (elopement, lack of insight, medication non-adherence, mania, and cognitive impairment), hypothesizing that these would also contribute to the likelihood of seclusion and restraint in forensic hospitals.

Data sources

Data were extracted from the Ontario Mental Health Reporting System on the Resident Assessment Instrument – Mental Health (RAI-MH) [51], which evaluates demographics, mental and physical health symptoms, substance use, behaviors, service utilization, and various functioning domains [52,53] on all adult psychiatric in-patient hospitalizations in Ontario. RAI-MH is a standardized clinical tool used to regularly assess psychiatric in-patients, with adequate reported inter-rater reliability (> 80% average percent agreement across all items) [53] and convergent validity for clinical scales [54]. We accessed the anonymized data on 13.10.2023 for research purposes, and information that could identify individual patients during or after data collection was not available or accessed. The variables used in the study were mandatory reporting requirements for Ontario hospitals and must be completed within 72 hours of admission to a mental health bed. A complete dataset with no missing information was obtained, ensuring that missingness did not pose a concern for data analysis.

Sample

The study included patients for whom full admission RAI-MH assessments (minimum 3-day admission) were completed from April 1, 2013, to March 31, 2023. We accessed data from 9001 forensic patient admissions to the 10 Ontario forensic psychiatric institutions designated facilities under part XX.1 of the Criminal Code of Canada [55]. We identified duplicate admissions based on the encrypted Ontario health identification number, and for cases with multiple admissions, we used the first admission per fiscal year studied. Patients without an Ontario health identification number are assigned a non-unique number and can include out-of-province residents or recent newcomers; 433 admissions fell under this category, and to help identify unique cases, we selected the first admission per fiscal year by sex, Indigenous-self identification, age, and hospital. We first deleted 1313 duplicates based on the unique identifier and then deleted 53 duplicates of the 433 cases with the same ID. We retained 7635 cases for analysis.

Study variables

Outcome variables included any seclusion and any physical or mechanical restraint documented during the first three days of admission. Seclusion was defined as present when a patient was locked in a room that confined them and from which they cannot leave freely, including being placed in a special seclusion room (n = 1323) or confined to their room (n = 1457) [56]. Restraint was defined as either mechanical restraint, whereby a patient was being restrained in bed or through wrist restraints only (n = 193), or physical restraint, whereby the patient was physically held to restrict their movement for a brief period to restore calm to the individual (n = 167)]; although mechanical and physical holding are different restraint strategies, they had similar durations (i.e., modal use < daily) and due to the relatively low proportion of the sample, it was necessary to combine these cases to capture all restraint, similar to Hui and colleagues findings [57]. The study included 10 forensic in-patient facilities in Ontario, which comprised one 160-bed maximum secure hospital, as well as facilities featuring 470 medium secure beds and 271 minimum secure beds.

Demographic variables included age in years at the time of admission, sex (male, female, other), whether the patient had immigrated to Canada, and Indigenous self-identification (based on patients’ self-identification as First Nations, Metis, or Inuit) [52].

Our administrative variable included forensic status at assessment. This RAI-MH variable consisted of nine categories, which we combined into three categories: 1 = Fitness-related status (fitness assessment, warrant of committal-unfit, treatment order, keep fit order and inter-hospital transfer-unfit); 2 = Not Criminally Responsible (NCR) – related status (NCR assessment, warrant of committal-NCR and inter-hospital transfer-NCR); and 3 = Other (e.g., court-ordered disposition recommendations).

Clinical variables included psychiatric diagnoses, insight into mental illness and medication compliance indicators, and several scales measuring symptoms and behavior [52]. Up to three diagnoses were documented using DSM-IV (n = 2,501) or DSM-5 (n = 5,134). We recoded all diagnoses to indicate whether each individual had a diagnosis within each of six categories: 1 = schizophrenia spectrum and other psychotic disorders, 2 = substance use disorders, 3 = mood or anxiety disorders, 4 = neurocognitive disorders, 5 = personality disorders, 6 = neurodevelopmental disorders.

The RAI-MH includes a measure to assess a patient’s awareness of their mental health issues and contributing factors [52]. This item was recorded as full, limited, or no insight, which we categorized as 0 = full or limited insight and 1 = no insight. History of Medication Adherence is a measure to determine if the patient took their medication as prescribed in the month prior to admission. The item was coded as 0 = at least 80% adherent or no medication prescribed, and 1 = less than 80% adherent or adherence unknown. The RAI-MH includes various behavioral symptom measures, including elopement attempts/threats. Elopement occurs when an individual attempts to leave or leaves the unit without staff knowledge or formal discharge (unauthorized leave). Elopement attempts/threats were recorded from (0) not exhibited in the last 3 days to (3) symptoms that occurred every day during the last three-day period [40,52]. We coded this item as 0 = not present and 1 = present.

We included four clinical scales derived from RAI-MH items [51,53,58]. The Aggressive Behavior Scale (ABS) is composed of four items (verbal abuse, physical abuse, socially inappropriate/disruptive behavior, resistance to care), each scored from 0 (not exhibited) to 3 (exhibited daily) with a total score ranging from 0 to 12 [59,60]. The Violence Sum (VS) sums three items (violent acts, intimidation, ideation), each scored from 0 (never) to 5 (in the last three days), and the total score ranges from 0–15 [58]. The Mania scale measures the frequency of seven symptoms (inflated self-worth, hyperarousal, irritability, pressured speech, labile affect, sleep problems due to hypomania, increased sociability/hypersexuality) with a total score ranging from 0–20 [51]. The Cognitive Performance Scale includes four items that assess short-term memory and recall, daily decision-making, communication abilities, and self-performance in eating. Scores on this decision tree algorithm-based scale range from 0 (indicating intact cognitive function) to 6 (indicating very severe impairment) [53].

Data analysis

We calculated descriptive statistics for admission, seclusion, and restraint frequencies, as well as for sociodemographic, administrative, and clinical variables. For continuous variables, including age, the mean (M) and standard deviation (SD) are reported. For length of stay, the mean (M), standard deviation (SD), median, and interquartile range (IQR) were provided. Categorical variables were summarized using the number (N) and percentage (%). We conducted unadjusted binary logistic regression analyses with independent variables to examine individual association with seclusion and restraint. To account for the clustering of responses within facilities/hospitals, we employed logistic Generalized Linear Mixed Model (GLMM). This approach allowed us to estimate the strength of the association between coercive measures (seclusion and restraint) and patient-level factors, including sociodemographic, administrative, and clinical factors, accounting for clustering within hospitals. We analyzed seclusion and restraint separately. All statistical analyses were completed using SPSS Version 29.

Risk factors for seclusion

Table 1 presents the comparison between secluded and non-secluded patients and the results of unadjusted binary logistic regression analyses examining the association between individual sociodemographic, administrative, and clinical factors and seclusion as the outcome variable. Secluded patients were significantly younger than their non-secluded counterparts. Unadjusted odds ratios showed that seclusion was significantly associated with younger male patients and those with a diagnosis of psychotic disorder, exhibiting lack of insight, being less than 80% medication compliant in the previous month, showing elopement behaviors, and with higher scores on the ABS, Violence Sum and Mania scales and higher cognitive impairment. Patients whose forensic status was fitness-related were almost three times more likely to be secluded than those whose status was NCR-related. Substance use disorder or neurocognitive disorder diagnosis and immigration background were protective factors in these bivariate analyses. Indigenous self-identification did not significantly affect seclusion outcomes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic, administrative, and clinical characteristics for not secluded and secluded patients and unadjusted associations (OR, 95% CI).

https://doi.org/10.1371/journal.pone.0328164.t001

The overall logistic GLMM model for seclusion (Table 2) was significant, F(19, 7609) = 62.140, p < .001, indicating that the predictors significantly explained variability in seclusion after accounting for hospital-level clustering effects. As expected, age and our measures of aggression and violence were significantly associated with higher odds of being secluded, whereby younger patients and patients exhibiting higher scores on the ABS and Violence Sum scales were more likely secluded, partially supporting our first hypothesis. Regarding diagnoses, a diagnosis of neurocognitive disorder was negatively associated with seclusion, whereas other diagnoses, including psychotic disorder, substance use disorder, and personality disorder, were not significant predictors,

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Fixed effects for logistic GLMM model predicting seclusion during the first three days of admission with hospital-level clustering effect.

https://doi.org/10.1371/journal.pone.0328164.t002

In this logistic GLMM model, being male was significantly associated with higher odds of seclusion; males were 50% more likely to be associated with seclusion than females. Lack of insight, medication non-compliance, higher scores on the Mania scale, and cognitive impairment were also significantly associated with seclusion. Compared to patients admitted for NCR-related reasons, patients admitted for fitness-related reasons were almost twice as likely to be associated with seclusion, whereas elopement was negatively associated. Indigeneity and immigration status were not associated with seclusion in this model.

After adjusting for patient-level predictors, 14% of the variance in seclusion was explained by hospital-level factors (Intraclass Correlation Coefficient [ICC] = 0.14), representing a moderate clustering effect. The fixed effects (patient-level predictors) explained 21% of the variance (Marginal pseudo R² = 0.21). When accounting for both patient-level predictors and hospital-level clustering effects, the total explained variance increased to 35% (Conditional pseudo R² = 0.35), indicating that hospital-level factors contributed additional explanatory power. Table 2 presents the associations between patient-level factors and seclusion during the first three days of admission, accounting for the hospital-level clustering effect.

Risk factors for physical or manual restraint

Table 3 shows results from our unadjusted bivariate regression analyses with restraint as an outcome. Restrained patients were younger compared to non-restrained patients. Unadjusted odds ratios showed that restraint was more likely applied to female patients and patients with a primary diagnosis of neurodevelopmental, personality, or mood and anxiety disorder. In addition, high scores on our clinical measures for aggression, violence, mania, and cognitive impairment were also associated with restraint, as was having no insight into mental health and being less than 80% medication compliant. Having a fitness-related status at admission compared to an NCR-related status increased odds of restraint more than two-fold and exhibiting elopement behaviors increased odds of restraint more than fourfold. Having an immigration background was not a significant factor, nor was Indigenous self-identification.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Demographic, administrative, and clinical characteristics for not restrained and restrained patients and unadjusted associations (OR, 95% CI).

https://doi.org/10.1371/journal.pone.0328164.t003

The overall logistic GLMM model for restraint (Table 4) was significant, F(19, 7609) = 24.177, p < .001, indicating that the predictors significantly explained variability in seclusion. As expected, age and measures of aggression and violence were significantly associated with higher odds of restraint, whereby younger patients and patients exhibiting higher scores on the ABS and Violence Sum scales were more likely to be associated with restraint, partially supporting our first hypothesis for restraint predictors. Regarding diagnoses, mood and anxiety disorders, neurodevelopmental disorder, and personality disorder were significantly associated with restraint. In contrast, other diagnoses, including psychotic disorder and substance use disorder, were not, in partial support of our hypothesis. Compared to being admitted for NCR-related reasons, fitness-related legal status increased the odds of restraint by 44%, and lack of medication adherence and higher scores on our cognitive impairment measure were also significant predictors in this model. However, neither sex, elopement attempts, insight, immigration status, nor Indigeneity were found to be significantly associated with restraint. The ICC was 0.02, indicating a low clustering effect. After adjusting for patient-level predictors, 2% of the variance in restraint was attributable to hospital-level factors (ICC = 0.02), reflecting a small clustering effect. The fixed effects (patient-level predictors) explained 5% of the variance (Marginal pseudo R² = 0.05). When accounting for both patient-level predictors and hospital-level clustering effects, the total explained variance increased to 8% (Conditional pseudo R² = 0.08), indicating that hospital-level clustering contributed additional variance beyond individual-level factors. Table 4 presents the associations between demographic, administrative, and clinical factors and restraint during the first three days of admission with a hospital-level clustering effect.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Fixed effects for logistic GLMM model predicting restraint during the first three days of admission with hospital-level clustering effect.

https://doi.org/10.1371/journal.pone.0328164.t004

Limitations

We conducted a secondary analysis of anonymized health records using each variable’s available data and existing operational definitions. For example, we were unable to identify diagnoses of bipolar disorder specifically in many cases; however, we mitigated this by including a measure of manic state, an important risk factor for aggression that may be related to restraint and seclusion. Further, analysis of archival data can have variable reliability and data quality; future studies using prospective clinical assessment are needed to replicate the present findings. Some methodological decisions were necessary during the data management; for example, we excluded multiple admissions each year, enabling our analysis to focus on cases but could have inflated our estimates of coercive measures if at-risk individuals were secluded or restrained upon new admissions in subsequent years. Our data pertained to seclusion and restraint within the first three days of admission, which enabled our examination of risk factors that clinicians can assess upon a patient’s admission to a forensic hospital but did not permit an assessment of factors associated with prolonged seclusion. We combined mechanical and physical restraint due to low proportions, even though they may be different strategies and experiences. We acknowledge that our definitions of coercive and restrictive practices may affect prevalence rates and that variability in definitions and reporting in the literature could influence the interpretation of risk factors for such practices. For example, we included only physical restraint (hands-on contact) or mechanical restraint (use of physical devices to restrict movement) in our definition of restraint, whereas Belayneh and colleagues included chemical restraint (medications intended to inhibit an individual’s movement or behavior) [2]. Chemical restraint was excluded in our study due to challenges in defining and identifying its use. PRN medications may serve multiple purposes, and clinical records often lack clear documentation of their intended use. Definitions also vary widely across different settings, which limits standardization. To ensure consistency, we employed a more conservative definition that focuses on physical and mechanical restraint. Nonetheless, certain forms of restraint may be systematically more likely to be applied to patients with specific characteristics. For example, Nicholls and colleagues reported that men were more likely to receive PRN antipsychotic medication than women, who were more likely to be placed in seclusion [67]. Our exclusion of chemical restraint may therefore affect our finding that gender was not a risk factor for restraint. Future research using time spent in seclusion or different kinds of restraint is required to determine whether the risk factors may differ.

Implications

The present study clarifies questions in the literature about whether certain risk factors are consistently associated with coercive measures in forensic settings, such as younger age, aggression, and violent behavior, which were consistent predictors of both seclusion and restraint in the present study. In addition, understudied indicators, including mania, poor cognitive performance, lack of insight, and medication non-adherence, were consistent predictors. Some demographic characteristics also contributed to both models, even accounting for aggression and mental disorder, indicating the need for further examination of the mechanisms of these associations. For example, staff may perceive young men to be more likely to engage in violent behavior and enact seclusion or restraint to prevent potential harm; different research methods, such as studies of staff perceptions of risk and functional analysis [68], may help elucidate some of the mechanisms.

The present findings provide some evidence to support the practice of individual-level assessments and individualized interventions to reduce the risk of seclusion and restraint. Understanding the factors associated with restraint and seclusion within the first three days of forensic admission allows clinicians to identify risk factors in the crucial time period shortly after a patient is admitted and provide timely interventions to limit the onset of the use of coercive measures. Our findings regarding clinical risk factors suggest valuable interventions among at-risk forensic in-patients including evidence-based pharmacological and psychosocial treatments. In particular, evidence-based interventions for the prevention of in-hospital aggression are indicated by our findings and may include both violence risk assessment [69] and assessments specifically to identify individuals at risk of experiencing coercive measures [5], along with aggression management training programs for clinical staff [70], especially in units caring for young men and individuals under fitness-related assessment or treatment orders.