Race, restriction, risk and reconviction: findings from an England and Wales medium secure cohort

Abstract

Over-representation of Black people in psychiatric hospital admissions is a long-standing issue; it is unknown whether this group is also at increased risk for discretionary restrictions that extend legal control after leaving hospital. Regression modelling explored associations between restriction order status, reoffending likelihood, index offence severity (violent/non-violent), psychosis, drug abuse and alcohol abuse. Models were generated for Black versus White medium secure patients discharged from all English and Welsh medium secure services during 1997/8. Black people were restricted more often than the White group. Restricted status was anchored on psychosis and violent index offence in the White group, but on psychosis and history of drug problems in the Black group; reconvictions were indistinguishable between the groups. Racial differences reflected the influence of stereotypes on medico-legal decision-making. These findings resonate with the recognised need for legislative reform and provide a template for identifying subtle (but important) forms of systemic racism.

Racial discrimination in mental health services and criminal justice is a global problem; an international meta-analysis (Barnett et al., 2019) showed that racial ‘minority’ groups are subject more often to involuntary hospital admission all over the world. Like elsewhere, UK-minority racial groups are detained for mental health treatment disproportionately; this problem is particularly marked for Black groups (Gajwani et al., 2016). These same groups are similarly over-represented in criminal justice (Tonry, 1994; Uhrig, 2016) and have been shown to receive more punitive court outcomes (Everett & Wojtkiewicz, 2002; Spohn, 2013). Discrimination against Black people has been reported at every stage in the UK criminal justice system; from policing through sentencing severity to prisons and secure mental health services (Coid et al., 2000; Gordon, 1983; Kilty & Joseph, 1999; Rowe, 2012; Souhami, 2011). Black people have a psychiatric hospitalisation rate of 309 per 100,000 whilst this is only 72 per 100,000 in the White UK population (UK Parliament, 2022).

Race and ethnicity are constructs that overlap considerably; they can be difficult to disentangle (García, 2016) and are often conflated (Ansell, 2013). Racial identity reflects a person’s sense of belonging to an identified racial group and reflects both their own internal perception of themselves and externally imposed social perceptions. In contrast, ethnic identity reflects the groups we identify with and who share our cultural characteristics, heritage, language, traditions and values. It must be acknowledged that official systems for recording identity in the UK have tended to conflate race and ethnicity (Byrne et al., 2020). Since the current data were drawn from service user records held in secure hospitals, and those recording the data were doing so ostensibly to identify the ethnicity of service users, it is very likely that those recording the ethnicity in the following data will not have been making subtle distinctions between race and ethnicity and will therefore have tended to conflate these concepts. The current paper will communicate in the parlance of race and racial identity because this is the language being applied in the current debate around the adequacy of UK criminal justice processes and mental health law with respect to reducing racial discrimination.

Increased recidivism rates and over-representation of minority groups in detention have tended to reinforce the impression that minority race status is itself a criminogenic factor (Holloway et al., 2022). Similarly, the elevated rates of hospitalisation for Black people with mental illness has been interpreted as a reflection of racial differences in either the severity of psychiatric symptoms (Gurak et al., 2018) or more extreme violence committed by mentally ill Black people (Coid et al., 2000). A third possibility has also been considered: that racial stereotypes play a part in professional consideration of the risks that people pose to others in both mental health care (FitzGerald & Hurst, 2017; Spector, 2001) and criminal justice (Kovera, 2019). Racial stereotypes tend to bias decision makers to act in more punitive and risk-averse ways when dealing with Black people (Spohn, 2015; Steffensmeier et al., 1998). Apart from excess detention, there is also a widely held belief that the mental health problems (and associated behaviours) are more chronic and extreme in Black people, and these behaviours warrant more frequent use of diagnostic labels such as ‘severe mental illness’, schizophrenia or psychosis (Thomeer et al., 2023).

It has been argued that racial bias in psychiatric diagnosis rates result from ethnocentricity in the definition of mental illness, especially in the case of psychotic diagnoses (Bredström, 2019; Sashidharan & Francis, 1999). Alternatively, some authors have attributed high rates of psychotic diagnosis in Black populations to cultural stereotyping or unconscious racism (Williams & Etkins, 2021), but these views are hotly contested with a range of hypothetical alternatives that include genetics (Zwickeret al., 2018), comorbid substance misuse (Beckmann et al., 2020) and racial differences in exposure to developmental hardship (Van Winkel et al., 2013). It has also been argued that experiences of racist victimisation and fractured social identity can contribute to the development and expression of psychotic symptoms (e.g. sense of differentness, negative self-awareness, paranoid ideation regarding persecution, self-questioning and self-esteem instability) in minority and immigrant populations (Lazaridou et al., 2023; McIntyre et al., 2021). Sadly, ethnic minority groups also report more adverse experiences even once they are in contact with mental health services (Barbee, 2002; Mensah et al., 2021).

The source of increased psychotic diagnoses in Black people is complex and will most likely include most or all the above factors, but the negative and pessimistic stereotypical views that people hold regarding those diagnosed with psychotic mental illnesses (e.g. unpredictability and dangerousness) are not disputed (Angermeyer et al., 2011). Amongst people experiencing mental illness, stigmatisation is more common in racial minority groups (Eylem et al., 2020). Black people diagnosed with psychotic illness are therefore subject to at least two sources of negative stereotyping that promote fear, rejection, and exclusion. In fact, it is likely that Black people in forensic psychiatric systems may experience so called ‘triple stigma’ as mentally ill, criminal and Black.

Stereotypical views of Black people in general, and of mentally ill Black people in particular, are not confined to the general population but extend into the operation of social systems that are intended to promote social justice (i.e. police, courts and mental health services) but tend to favour the White population (Walker, 2020). Some have concluded that both the criminal justice and mental health systems in developed countries are systemically racist (Bowling & Phillips, 2002; Nazroo et al., 2020).

UK courts and the Ministry of Justice have the power to impose restrictions (Restriction Orders) on people who are going to be treated in secure services. The conditions set out in restriction orders become active once a person leaves secure care or prison. For instance, a person ordered by a court to be treated in a psychiatric hospital under Section 37 of the Mental Health Act may also receive a restriction order under Section 41 of the Act. The purpose of restriction orders is to ensure public protection through conditional release, continued supervision and possible recall to hospital. Restricted service users can be recalled either for breaching the conditions of their release or if their behaviour is considered by those supervising them to represent a significant increase in risk (e.g. excessive alcohol use or treatment noncompliance).

Ideally, decisions to impose restriction orders should be informed by the nature of anticipated harm (e.g. violence or sexual offences versus non-violent offences) and the likelihood of future problem behaviour (e.g. probability of reoffending or reconviction). Restriction order decisions should also be based on factors shown in the literature to be associated with increased risk for violence such as the presence of early antisocial behaviour (Farrington, 2020), psychotic diagnoses (Whiting et al., 2022), drug use problems and/or alcohol problems (van Amsterdam et al., 2020). Restriction order decision-making should of course be informed by evidence in the same way for different racial groups; racial differences in how these evidenced factors influence restriction order decisions would indicate the influence and nature of stereotypes that might potentially bias decisions towards more punitive options for Black people. ‘Problematic, criminal drug user’ is one stereotype that criminal justice systems have tended to reinforce in relation to young Black people (Provine, 2011). This stereotypical perception of Black people has also been shown to extend into medical decision-making (Moskowitz et al., 2012), and it could plausibly have influenced legal decisions to impose restriction orders.

One of the roles of forensic psychiatrists in criminal proceedings is to make recommendations regarding the need for (and nature of) restrictions intended to manage risk when people leave psychiatric care. Forensic psychiatric decision-making has often been criticised (Desmarais et al., 2010), with some evidence that it can be prone to so-called anchoring heuristics (e.g. Lawrence et al., 2018). Under an anchoring heuristic, decisions are ‘anchored’ on a limited subset of variables, even when a full range of relevant factors are available (Neal & Grisso, 2014). Anchoring heuristics are characterised as a form of cognitive short-cut that people use to simplify information to facilitate decision-making with uncertain outcomes and where decisions need to be informed by multiple items of information (Tversky & Kahneman, 1974). Racial stereotypes have also been characterised as an anchoring strategy employed by people to simplify their social appraisals of members of different racial groups (Tamborini et al., 2007).

Given the public protection function of restriction orders, we assumed that restricted status should have been uniformly associated with factors shown to be valid predictors of future harm (extensive conviction histories, violent index offence, psychotic diagnoses, drug and alcohol problems), and we used this assumption to explore possible racial differences in how the UK criminal justice system evaluated risk to inform restriction order decision-making. Risk assessment traditionally involves consideration of the severity of anticipated harmful events (where the empirical focus has been on future violence), along with estimation of likelihood (Fazel, 2019; Whiting et al., 2021).

It should be noted, however, that the association between past violence and future violence is a contested one, with some studies indicating that general antisocial behaviour is more strongly associated with future violence than past violence (Farrington, 2001). The violent nature of index offences is a potent factor in the care and management of people in forensic psychiatric care and tends to be more influential in shaping long-term decision-making than consideration of broader criminogenic risk (Andreasson et al., 2014; Hill et al., 2023b). We therefore anticipated that index offence violence should be uniformly associated with restricted status in both the Black and White racial groups.

In the context of forensic psychiatry, the nature of psychiatric presentation might also legitimately contribute to risk formulation. The presence of a psychotic diagnosis (especially) could be regarded as a legitimate consideration that should influence decisions to impose restriction orders, because of the widely reported association between psychosis and violence (Fazel et al., 2009; 2010; Wallace et al., 1998; Whiting et al., 2022).

One of the difficulties for determining whether racial bias exists in the administration of restriction orders stems from the absence of research on this issue; the focus has instead been on description of minority ethnic over-representation in adolescent (Chowdhury et al., 2005) and adult services (Coid et al., 2000). No attention has been directed to potential differences in the way risk and race are related to decisions (prior to hospital admission) regarding the future management of patients under the Mental Health Act after they leave services. Specifically, it is currently unknown: (a) whether there are racial differences in the frequency of restriction orders, (b) the extent to which objective indicators of risk (for future violence) are associated with restriction orders, or (c) whether risk is appraised differently between racial groups to inform restriction order decision-making. The main objective of the current analysis was to provide the first examination of racial bias in the application of restriction orders under UK mental health legislation. Securing contemporary data to perform this analysis would have required considerable resources and time, so the current analyses relied upon the best available historical data (patients discharged during 1997/8 and followed up until 2004). We acknowledge that subsequent innovations (e.g. the 2007 revision of the Mental Health Act) may limit the currency of our findings, but our findings provide the first observation against which the effects of subsequent changes to practice and policy might be evaluated. We hope that the findings of the current study might also warrant resourcing of contemporary analysis of the sources and nature of racial bias in medico-legal decisions in the courts regarding people with recognised serious mental health problems.

We expected to find that Black people were over-represented in a historical cohort of medium secure service users. Although this is an important observation, the purpose of the current paper was to go beyond merely describing racial differences in admission rates or restriction order rates and was instead to explore whether racial stereotypes influenced how risk was appraised to inform restriction order decision-making.

Method

Design

This research relied upon secondary analysis of data obtained from a Department of Health (DoH) funded database (Maden et al., 2004). The study used a correlational naturalistic observation of a cohort of service users discharged from all 35 medium secure services in England and Wales between 01 April 1997 and 31 March 1998. The follow-up data included reconvictions for six years post-discharge, and these were obtained from the Home Office’s National Offender Index. Offence histories prior to hospital admission were used to estimate the risk of future reconviction using the Offending Groups Reconviction Scale 2 (Taylor, 1999).

The severity of index offences was coded based on whether the index offence was non-violent versus non-fatal violence (see Table 1). Cases with index offences that resulted in human fatality (murder or manslaughter) were excluded because these would have resulted in mandatory life sentences and/or restriction orders, which precluded involvement of psychiatric/judicial discretion in decision-making.

Table 1.

Racial cross-tabulation and severity coding of index offences of included cases.

Index offence	Black	White	Severity coding
			Violent
Attempted murder	3	9
Kidnapping	0	5
Aggravated burglary	1	3
Wounding	12	34
Grievous bodily harm	10	24
Actual bodily harm	15	22
Assault	2	4
Affray	5	13
Common assault	7	18
Rape	8	9
Buggery	0	4
Total	63	145
			Non-violent
Threats of violence	1	11
Indecent assault	8	21
Robbery	14	29
Arson	2	25
Theft	2	10
Burglary	5	25
Drug possession	3	5
Other offence	5	38
Total	40	164

Note: All cases with fatal outcomes were excluded.

Statistical analysis first addressed whether restricted status was associated purely with criminogenic risk and index offence severity, or whether this outcome was also associated with race. Restricted status was initially regressed against three predictor variables; the racial group of service users (Black or White), criminogenic risk (estimated using the Offender Group Reconviction Scale 2, OGRS–2) and the severity of index offences (no violence vs. non-fatal violence). In a subsequent pair of regression models the association between restricted status, criminogenic risk, index offence severity and psychotic diagnoses was examined separately for Black versus White service users.

Ethics

Ethical approval was originally granted by the London Multi-centre Research Ethics Committee on the 20 February 2004 (MREC 99/06). The need for informed consent was not obtained from all individual participants included in the study because many of the discharged service users would have been untraceable for the purposes of obtaining consent. The current analysis was later approved as a secondary analysis as part of a PhD project (see Hill, 2020). Secondary approval was granted by Cardiff Metropolitan University’s School of Sport and Health Science Research Ethics Committee, reference number (Project Reference Number: PGR-1365). The purpose of the secondary analysis was to provide a baseline for comparison of outcomes with a more contemporary cohort of people discharged by a medium secure service in South Wales; all data were anonymised prior to storage and analysis.

Participants

The sample initially contained data for 1178 service users discharged between 01 April 1997 and 31 March 1998 from all 35 NHS and private medium secure units in England and Wales. Racial identity had been recorded for only 878 of the above cases; 300 cases were subsequently excluded. Thirteen cases were excluded because there was no recorded primary diagnosis. A further 92 cases were excluded because their offence histories could not be retrieved from the Offenders Index. Thirteen cases were excluded because their ethnicity was recorded as African but racial group was unrecorded. Sixty-eight cases were excluded because their index offence resulted in at least one fatality (murder or manslaughter). We finally excluded 151 cases because they were admitted informally (n = 18) or under civil sections of the Mental Health Act (MHA; Section 3, n = 117; Section 2, n = 16) with no option to impose a restriction order.

These various exclusions meant that the initially retained sample included 451 people. The racial composition of the sample at this stage was as follows: White n = 309 (68.5%), Black Caribbean n = 72 (16.0%), Black African n = 28 (6.2%), Black Other n = 4 (0.9%), Indian n = 11 (2.4%), Pakistani n = 1 (0.2%), Bangladeshi n = 1 (0.2%), Other n = 3 (0.7%) and Mixed-heritage n = 22 (4.9%). Indian, Pakistani, Bangladeshi, Other and Mixed-heritage cases were excluded because these groups were unfeasibly small for meaningful statistical comparison and also because there is evidence that minority ethnic groups are all treated differently by the criminal justice system and are also differentially exposed to social disadvantage (Lammy, 2017); it was not safe therefore to assume homogeneity between these groups, and they could not have been meaningfully combined in a ‘Other Race’ group for comparison with the Black service users. This final exclusion left a retained sample of 412 people (309 White, 72 Black Caribbean, 27 Black African and 4 Black Other).

The current analysis was focused on differences in the discretionary application of restriction orders to the Black group of service users relative to the White racial group. The Black Caribbean, Black African and Black Other groups were contrasted in terms of admission age (F < 1), psychotic diagnosis, history of drug problems and history of alcohol problems; there were no significant differences (highest chi squared = 1.16, p = .56). In addition, mean risk for reconviction did not differ between the Black Caribbean (mean = 40.75, SD = 26.77), Black African (mean = 30.13, SD = 21.90) and Black Other (mean = 36.80, SD = 20.63) groups, F(2, 143 = 2.46, p = .09. In the absence of any statistical differences the Black groups (Black Caribbean, Black African and Black Other) were combined into a single ‘Black group’.

The racial distribution of this final sample was 309 (75%) White and 103 (25%) Black people. Two of the Black group and five of the White group had no data related to alcohol problems, whilst two of the Black group and seven of the White group had no record of drug use problems prior to admission; these cases were retained because these omissions were few and unlikely to bias the following analyses.

Primary diagnoses of the included cases were as follows: psychosis was most common with 68.4% of cases, personality disorder (8.1%) and mood disorder (7.4%) diagnoses were next most common, and the remaining diagnoses were infrequent and included substance misuse, learning disability, neurosis and paraphilia, all <2% each.

The White group were on average slightly older (mean age = 25.7 years) than the Black group (mean age = 24.7 years), but this difference was not statistically significant, t = 1.19, p > .05. Psychotic diagnoses were proportionately more frequent in the Black group (95.1% vs. 73.8%); statistical analysis of this difference is presented in the results section that follows. Sex was recorded for all 412 included cases (376 male, 36 female); racial group was independent of service user sex, chi squared = 0.00, p > .05.

Exclusion from the following analysis was independent of racial grouping, chi squared = 0.22, p > .05, psychosis diagnosis, chi squared = 1.31, p > .05, and OGRS–2 score, t(709) = 0.56, p > .05. Exclusion was, however, associated with index offence violence; proportionally more cases with non-violent index offences (32.67%) were excluded than people with violent index offences (8.33%), chi squared = 64.31, p < .05, but racial grouping was not associated with index offence violence in the excluded cases, chi squared = 1.73, p > .05, so exclusion of non-violent cases was unlikely to have confounded the following analyses that included racial group and index offence violence as factors.

Materials

Risk for reconviction was estimated using the Offender Group Reconviction Scale 2 (OGRS–2; Taylor, 1999). OGRS–2 has been shown repeatedly to have acceptable predictive validity for groups of people treated in medium secure services (Coid et al., 2011; Gray et al., 2004; Hill et al., 2023a) and can be regarded as a robust assay of service user future risk for reconviction. Predictive validity of OGRS–2 for reconvictions amongst people discharged by UK medium secure services has been established and is not the subject of enquiry in the current analysis. It is important to note that OGRS–2 was validated on a UK national sample of probationers released from prison contemporaneously with this medium secure cohort’s discharge from services (see Hill, 2020; Hill et al., 2023b).

Procedure

The data used for the current analysis were sourced from a national cohort study from which some findings have already been published (see Hill et al., 2023a, 2023b; Maden et al., 2004). Data were analysed (SPSS 29, IBM) using binary logistic regression and chi squared to explore racial differences in restricted status and associations between index offence severity, criminogenic risk and restricted status.

Results

The first analysis explored possible association between racial grouping and restricted status. Restriction order decisions were associated with race; Black patients were restricted more often (see Figure 1). The following series of analyses considered possible confounding of this association by: (a) higher risk for criminal justice involvement in the Black group (Holloway et al., 2022), (b) more severe index offences in the Black group (Coid et al., 2000), or higher rates of psychotic diagnoses in the Black group (Bredström, 2019; Misra et al., 2022). Patient conviction histories were divided into pre-admission and post discharge convictions using admission and discharge dates as cut-offs. We used pre-admission OGRS–2 scores (with index offence as current offence) as an estimate for reoffending likelihood at the time of admission and found no statistical difference between the OGRS–2 scores for the Black (M = 37.7, SD = 26.53) and White groups (M = 43.63, SD = 27.89) groups, t(410) = 1.29, p > .05. We conclude that there was no difference between the Black and White groups in terms of risk for future offending and that this potential confound did not contribute to increased rates of restriction orders in the Black group (Figure 1). Next, we examined whether violent index offences were more frequent in the Black group; a significant association between race and index offence severity was detected, chi square = 5.43, p < .05; non-fatal violent index offences were more frequent in the Black group (60.6%) than in the White group (47.5%). A third analysis explored possible elevated levels of psychotic diagnoses in the Black group and revealed (as expected) higher rates of psychosis diagnosis in the Black group (95.4%) than in the White group (73.8%), chi square = 21.34, p < .001. The next analysis explored whether racial grouping was associated with a history of drug problems; it was not, chi square = 1.67, p > .05. Finally, a history of alcohol problems was found to be associated with racial grouping; alcohol problems were more prevalent in the White group (54.0%) than in the Black group (21.6%), chi square = 32.25, p < .001.

Figure 1.

Percentage of cases with and without restricted status in the Black versus the Other racial group.

The next stage of analysis was focused on exploring associations between indicators of risk for violence and restriction order decision-making separately in the two racial groups; the data file was ‘split’ into racial groupings (Black vs. White), and two binary logistic models regressed restricted status against five predictor variables: psychotic diagnosis, reoffending risk (OGRS–2), index offence violence, drug problems and alcohol problems. The resulting pair of regression models are presented in Table 2 and were designed to explore the possibility that decisions to impose restriction orders on the Black and White groups were informed differently by these risk factors. Both the Black and White groups were adequately powered for logistic regression in accordance with the 1 in 10 Rule (Harrell, Lee, & Mark, 1996); with five predictor variables and where the event probability (of a restriction order) was relatively high.

Table 2.

Binary logistic regression coefficients for restriction order outcomes (restricted/not restricted), with index offence severity (violence/no violence), reconviction likelihood (0–100%), psychosis diagnosis (yes/no), drug use problems (yes/no) and alcohol problems (yes/no) as predictors.

	B	S E	Wald	Sig	R2
White group					.09–.12
Index violence	1.024	0.27	14.35	.001
OGRS–2	0.005	0.005	1.09	.30
Psychosis	0.825	0.28	8.44	.004
Drug problems	0.169	0.30	0.32	.57
Alcohol problems	0.225	0.27	0.68	.41
Black group					.15–.25
Index violence	0.144	0.602	0.057	.811
OGRS–2	0.024	0.013	3.327	.068
Psychosis	3.200	1.296	6.099	.014
Drug problems	1.782	0.766	5.419	.020
Alcohol problems	0.240	0.791	0.092	.762

Note: Analyses were conducted separately for the White and Black groups. OGRS–2 = Offender Group Reconviction Scale 2.

Table 2 is supplemented by Figure 2 which presents the percentage of patients given restriction orders cross tabulated with psychosis diagnosis, index offence violence, drug problems and alcohol problems for the Black and White racial groups. The absence of association between restricted status and OGRS–2 scores for both the Black and White groups suggested that decision makers consistently neglected past offence histories when deciding whether to impose restriction orders on both racial groups; the absence of association here was confirmed by similar mean OGRS–2 scores for restricted and not-restricted cases in both the Black (restricted M = 39.1, not restricted M = 41.5); t(102) = 0.36, ns, and White (restricted = 44.3, not restricted =  42.3), highest t(307) = 0.61, ns groups. In contrast, restriction order decisions were consistently associated with psychotic diagnosis (see Figure 2), with restriction orders being given more often to people diagnosed with psychosis in both groups (White group chi squared = 11.61, p < .001: Black group chi squared = 11.66, p < .001).

Figure 2.

Distribution of restriction orders in the Black and White groups, cross tabulated with psychosis diagnosis, index offence violence, patient drug problem and patient alcohol problem.

Restriction order decisions were associated with violent index offences in the White group, but this was not so for the Black group of patients. Inspection of Figure 2 reveals that restricted status was more likely following a violent index offence, and less likely (than unrestricted status) in the absence of a violent index offence in the White group (as would be expected), whilst restricted status was the most frequent outcome following both violent and non-violent index offences in the Black group.

Analysis of reconviction outcomes

There is evidence that reconviction rates for medium secure service users vary widely between different discharge pathways (Hill et al., 2023b). Before comparing reconviction outcomes for the Black and White groups it was necessary to control for exposure to risk of reconviction by comparing the discharge pathways for the two groups. Frequency of discharges to open conditions, further secure care or prison were compared between the Black and White groups; there was no association between group and discharge pathway, chi squared = 1.14, p > .05; the two groups were equivalent in terms of exposure to risk for reconviction, and differences in discharge locations did not therefore confound interpretation of the following reconviction data.

Restriction orders are designed to manage risk of future harm and are imposed when the court believes there is high likelihood of future significant harm. Section 41, for instance, makes remaining in the community following discharge from a Hospital Order (Section 37) conditional on compliance with bespoke requirements that are intended to mitigate risk for each individual case (e.g. where people live, compliance with treatment, or avoidance of substance misuse). Failure to comply with conditions set out in a Section 41 Order can result in involuntary recall to hospital and reinstatement of the accompanying Section 37 Hospital Order. The limited evidence that is available suggests that restriction orders are associated with reduced reconvictions following discharge from secure hospital (Coid et al., 2015; Hill et al., 2023b). Since the Black and White groups were equivalent in terms of risk for reconviction (according to OGRS–2 scores) then it followed that increased frequency of restriction orders for the Black group should have been associated with reduced reconvictions because of the additional controls applied on them following discharge; the following analysis was designed to explore this possibility.

The next set of analyses was designed to explore reconvictions for the Black and White groups during the six-year follow-up. Cumulative percentages of reconvictions were compared at three time points; these are presented in Figure 3. Inspection of Figure 3 suggests that reconviction rates for violence in the two groups were indistinguishable. In contrast, reconvictions for any offence appeared to be most frequent in the White group. Chi-squared analyses showed that racial grouping was independent of any offence reconvictions at 2 years, 4 years and 6 years following discharge (lowest p = .145). Violent reconvictions were categorised in the same manner as index offences (see Table 1); contrast of the violent reconviction rates for the Black and White groups showed no association between racial grouping and violent conviction outcomes at any of the follow-up time-points (lowest p = .183).

Figure 3.

Cumulative percentages of Black and White groups who were reconvicted for any offence (left panel) or violent offence (right panel) during a uniform 6-year follow-up period.

Discussion

The racial composition of this medium secure cohort was consistent with previous reports of over-representation of Black people admitted to UK general (Barnett et al., 2019) and secure (Coid et al., 2000) inpatient care, with 23.1% recorded in the sample as Black at a time when this racial group represented only 2.2% of the UK population (UK Census, 2001). Black racial status was also associated with increased risk for restricted MHA status prior to admission to medium secure services.

The current analysis went beyond prevalence and was the first to explore the potential influence of racial stereotypes in decisions to impose restriction orders on people entering secure psychiatric hospitals through the UK criminal justice system. The current findings suggest that broad criminogenic risk (estimated using OGRS–2; Taylor, 1999) did not inform restriction order decisions for either the Black or White racial groups; a history of alcohol problems was also not associated with restriction order outcomes. Neglect of these two reliable predictors of future criminal behaviour (Hill et al., 2023a; van Amsterdam et al., 2020) was surprising and was the first indication of heuristic anchoring in restriction order decision-making.

As was expected, the presence of a psychotic diagnosis was associated with increased risk for restricted status in both the Black and White racial groups, suggesting that this factor functioned uniformly as an anchor for decision makers when appraising risk for both groups of patients. Violent index offence functioned as an anchoring component when appraising risk in the White group, whilst a history of drug problems functioned as an anchor for assessments of Black group. It seemed that decisions regarding whether to use restriction orders were influenced by two quite different stereotypes: (a) psychotic Black drug users are more dangerous, and (b) non-Black psychotic people are dangerous when they have been violent. This finding is perhaps even more remarkable, since the prevalence of documented drug problems was the same in both groups and because violent index offences were less frequent in the White group.

In our final analysis we considered the possibility that Black people were subject to more restriction orders because they were seen as at increased risk for future offending solely by virtue of their racial grouping (Holloway et al., 2022). We tested this idea directly by comparing the reconviction rates for Black versus White groups during a uniform six-year follow-up period; there were no differences in either general or violent reconvictions, and we can discount the possibility that Black racial status per se was a criminogenic factor. We conclude that (in terms of long-term reconviction outcomes) race was not a legitimate factor for shaping restriction order decision-making.

We excluded cases that involved fatal index offences because these would have been subject to mandatory restrictions; this exclusion was to ensure that psychiatric professionals’ evaluation of risk would have informed decisions regarding the discretionary need for restrictions in all cases. The current analysis included two types of restriction order. Firstly, Section 41 is imposed in conjunction with a hospital order (e.g. Section 37) from court and requires input from two independent psychiatric opinions following assessment that is requested by the court. Section 49 restriction orders also require two psychiatric opinions and are requested by the Ministry of Justice when remand (Section 48) or convicted (Section 47) prisoners are transferred from prison to hospital for mental health treatment. Our analysis therefore focused on an important juncture between criminal justice and psychiatric practice, where decisions about who needed to be restricted following discharge/release were made.

Both increased risk for involuntary admission into secure care and for restricted status in the Black group could be regarded as evidence of systemic racial bias (Nazroo et al., 2020) at the juncture between criminal justice and medium secure services in the UK. One problem with this simple interpretation was that the Black group were more likely to have committed a violent index offence than the White group, and decisions to impose restrictions more often on the Black group may have simply reflected the increased severity of potential harm (Coid et al., 2000) that decision makers were attempting to mitigate. The absence of any association between index violence and restriction order status in the Black group made this confounded interpretation of our findings untenable, and we conclude that racial bias was playing some part in decisions to restrict Black people more often; in fact restricted status was more likely than non-restricted status for the Black group irrespective of whether index offences were violent.

It is important to note that 94% of the Black group and 75.5% of the White group were diagnosed with psychosis. Finding effective ways to reduce overdiagnosis of psychosis in black people (Bredström, 2019; Sashidharan & Francis, 1999) may perhaps be the most effective way to deal with the problem of racial bias in restriction order decision-making. Transdiagnostic as well as trauma-informed approaches and more detailed consideration of phenomenology of psychotic presentation and how this relates to risk might also be a useful way forward (Brucato et al., 2019).

Our findings have implications for the fair application of legislation at the intersection between the criminal justice and mental health systems. The apparent bias that was observed indicates that highly restrictive legislation that has the potential to limit the liberty and autonomy of patients long after their discharge from hospital was applied disproportionally and, for the Black group, was not based upon risk of harm. This is problematic, not least because such practice undermines the core principles of equality and fairness that are meant to underpin justice-related decision-making (Courts and Tribunals Judiciary, 2024). It also has implications for public confidence in the justice system and for those who come into contact with it. Theory and research related to procedural justice indicate that experiencing fair and just legal procedures leads to people viewing the law and authority figures as legitimate, resulting in greater compliance, and, when applied, procedural justice has been associated with positive outcomes for people who use justice-based services (Fitzalan-Howard & Wakeling, 2020; Lind & Tyler, 1988). Addressing and reducing the bias that has been reported in the current study would result in more procedurally just decision-making regarding the application of restriction orders, which could in turn improve outcomes for patients subjected to such legislation. For example, implementation of procedural justice principles has been associated with better mental health and psychological wellbeing (Beijersbergen et al., 2014) and fewer behavioural incidents in institutional settings (Fitzalan-Howard & Wakeling, 2020).

Limitations

The characterisation of index offences, restriction orders, drug problems, alcohol problems, psychiatric diagnosis and offence histories in the current analysis was superficial. All of these variables are complex phenomena in their own right, and the current analysis could be criticised for over-simplification of the possible relationships between all of them. We acknowledge this, but justify our approach because presently there are no available sources of more detailed and nuanced data, and it was in keeping with other research evaluating restriction orders and secure services (e.g. Coid et al., 2000, Hill et al., 2023b). Furthermore, this is the first investigation to explore the possible influence of racial stereotypes in this aspect of clinical heuristics; our intention was to determine whether there were potential racial differences and to describe them; it is clear that more detailed evaluation is warranted. The current analysis was limited by the national audit nature of the original data, and future studies will ideally build more detailed and valid methods for measuring patient characteristics into their design, prospectively, though it must be recognised that resourcing such studies would be prohibitively expensive. We suggest that future research on this issue might consider the influence of factors such as patient offence histories, the phenomenology of acute/chronic psychiatric symptoms and severity/chronicity of patient drug or alcohol problems. Future research might also consider the associations between these factors and the specifics of conditions set out in restriction orders.

The cohort of patients whose data contributed to the current analysis were discharged by services more than 25 years ago. It could be argued that the current findings might only be of historical interest and that much has happened in between, with society-wide shifts in racial tolerance/intolerance, amendment to UK mental health law, improvements in psychiatric training and substantial changes to the structure of the UK secure mental health estate. It might be reasonable to suppose that UK criminal justice and clinical decision-making might have become more culturally aware/fair, or better informed by standardised, evidence-based methods of risk assessment. Since the current cohort of patients were treated in secure services there has been a substantial increase in training designed to address racial inequalities in healthcare. Equality, Diversity and Inclusion (EDI) training is now mandatory for all NHS staff in England. One would hope that the requirement for EDI training and the change in the prevailing culture around racial inequality may have reduced the impact of racial stereotypes in the appraisal of risk and subsequent controls applied to service users; this, however, must remain a supposition in the absence of evidence. Application of the current methods on data from a more contemporary cohort of secure psychiatry service users would be the only way to determine whether changes in cultural sensitivity have been associated with a reduced impact of racial stereotypes on restriction order decision-making. Unfortunately, there is currently no evidence to suggest that racial discrimination in UK criminal justice or mental health has diminished substantially, and as recently as 2022 Black people were compulsorily admitted to psychiatric treatment in excess of four times more often than their White counterparts (UK Parliament, 2022).

The current analysis intentionally omitted cases with index offences that resulted in human fatality, because these would have been subject to mandatory restriction orders and therefore did not involve clinical discretion regarding whether to impose restrictions. The post-discharge management of these cases with ‘high impact’ index offences is arguably of greater public concern than the management of cases without fatalities. For the fatality cases clinicians will still have been involved in formulating recommendations regarding the nature of restrictions and determining the conditions that warranted recall. It is plausible that racial stereotypes may also have influenced the nature and conditions set out in restriction orders, and the current analysis does nothing to address this interesting question for either the fatal or non-fatal index offence cases. Access to and analysis of pre-sentencing reports of people admitted by secure services would allow researchers to address this. There are good reasons to suppose that clinicians might have recommended more punitive measures (Spohn, 2015; Steffensmeier et al., 1998) and conditions for the Black patients, and the current analysis suggests that racial stereotypes and anchoring heuristics will play their part in this decision process also.

Proposed revisions to UK mental health legislation that aim to reduce racial discrimination are confined to Sections 2, 3, 5 and 20 of the MHA (1983, 2007), which cover civil detention for assessment or treatment. Proposed revisions to MHA do not consider sections administered in court in the form of Hospital Orders (Sections 35, 36, 37 and 38), those that regulate transfer of prisoners to mental health care (Sections 47 and 48) or restriction orders that are intended as additional public protections and can be put in place during criminal proceedings (Section 41) or imposed by the Ministry of Justice for those transferred from prison to hospital (Section 49). Patients subject to restriction orders can only be discharged from hospital with approval from the Office of The Secretary of State for Justice, and such patients are subject to extended periods of supervision, control and potential involuntary recall to hospital long after they leave inpatient care. Restriction orders impose onerous constraints on the liberties of those who are subject to them, and their sole stated purpose is public protection. Restriction orders require at least two psychiatric opinions to inform either judiciary or the Ministry of Justice; restriction order decision-making represents an important juncture between criminal justice and mental health services in which psychiatric appraisals of risk and risk-based decision-making play prominent roles. The onus on practitioners therefore is to ensure that restriction orders are used impartially, in an evidence-based way that is not shaped by racial stereotypes.

Ethical standards

Declaration of conflicts of interest

Eleanor Brooks has declared no conflicts of interest.

Daniel Lawrence has declared no conflicts of interest.

Ruth Bagshaw has declared no conflicts of interest.

Charlotte Hill has declared no conflicts of interest.

Anthony Maden has declared no conflicts of interest.

Jason Davies has declared no conflicts of interest.

Joe L. Davies has declared no conflicts of interest.

Andrew Watt has declared no conflicts of interest.

Ethical approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of Cardiff Metropolitan University’s Research Ethics Framework (approval reference PGR-1365), with the London Multi-centre Research Ethics Committee (approval reference MREC 99/06) and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

Informed consent

Informed consent was not obtained from all individual participants. The requirement for informed consent was waived on the grounds that many of the discharged patients would have been untraceable for the purposes of obtaining consent.