Introduction
Forensic psychiatry (FP) has the complex task of caring for mentally disordered offenders: assessing and treating mental illness and simultaneously protecting the public interest where individuals are deemed dangerous [
1,
2]. FP has taken on different forms across the world and has developed to different extents [
2]. Its scope stretches across several settings: courts, general psychiatric settings, communities, prisons and dedicated forensic psychiatric hospitals. In some places, specialist FP is not available at all [
3]. Naturally, FP is heavily intertwined with and dependent on the criminal justice system given its target population, and it is usually up to the justice system to weigh forensic psychiatrists’ expert opinion, decide on criminal responsibility and whether and where a custodial sentence must be served [
4].
While the management of forensic patients includes rehabilitation and reintegration into society, protecting the public from potentially dangerous individuals remains an important consideration. Forensic inpatients have lengths of stay (LOS) spanning months or, more commonly, years, increasing in the past decades [
5]. These admissions can have LOS that are shorter, equivalent, or sometimes longer than times in detention for imprisoned offenders for the same charge [
6]. A high-security bed in the United Kingdom (UK) will be occupied by a forensic service user for an average of 70 months and a medium-security for 26 months [
7]. A 2018 survey of 23 medium- and 3 high-security hospitals in England identified 23.5% of inpatients as “long-stayers”, defined as inpatients staying for more than 5 years in medium-security and more than 10 years in high-security services [
8]. This phenomenon is not limited to the UK; in Brazil, the average LOS in forensic hospitals was 6 years [
9], whereas in the Netherlands, the duration was greater at 8 years [
8]. The longer LOS of forensic admissions is also apparent within specific diagnoses. In a 2008 study, patients with schizophrenia spent more days per year hospitalised when they had a forensic admission compared with those who had a non-forensic admission [
10].
Forensic mental health services come at a great cost owing to the complexity of care. In the 2019 Scottish government inpatient census, the total number of patients being cared for in a forensic ward in Scotland was 412 [
11]. in a population of 5,479,900 people; this highlights the small patient population admitted to such units. According to the Centre for Mental Health Care in the UK, the average cost per annum in England for low-security and medium-security beds is £153,300 and £176,295, respectively, while the cost for high-security beds ranges from £271,560 to £357,335 depending on the specific service [
12]. High-security beds in Scotland come at comparable costs, approximately £6195 a week and £322,140 per annum in 2009/10 [
13]. Overall, secure care services cost the NHS £821 million in England in 2018/19, corresponding to 10.9% of all public expenditure on mental health services [
14]. This high cost is not limited to the UK. In the Netherlands, a forensic bed costs an average of 388 euros per day, adding up to 141,620 euros per patient per year [
15]. In Japan, the cost of a forensic bed is US$186,019 per year [
16], a cost 4.4 times higher than a non-forensic involuntary admission [
17]. This cost, however, must be understood within the broader healthcare landscape and it is essential to recognise the value these services provide. Alternatives such as prolonged high-security imprisonment may incur comparable or even higher expenses, not to mention potentially inferior outcomes in terms of rehabilitation and recidivism rates.
While high cost is one reason to aim for a shorter LOS, the harms of longer admissions must be considered. A lengthy admission can adversely impact patient outcomes and quality of life and reduce the likelihood of future independent living by the individual [
18]. Furthermore, it can lead to institutionalisation and social withdrawal [
19] and greatly limit autonomy. Forensic environments are by default highly restrictive, with great emphasis placed on security of the individual and others [
20], so a longer admission can result in sometimes unnecessary [
21] prolonged deprivation of civil liberties for the individuals [
22].
However, at present, there is no widely accepted definition of what constitutes a “long LOS” in forensic services; the Butler Committee on Mentally Abnormal Offenders interim report published in 1975 [
23] and the Glancy report published in 1973 [
24] by the Department of Health and Social Security and the Home Office recommended a maximum 18-month to 2-year stay for medium secure units before an alternative placement is sought. However, these were mere recommendations; research across the UK has repeatedly demonstrated that the LOS in medium- and high-security forensic hospitals often exceeds these numbers [
5,
25‐
27]. It is also worth noting that the original Glancy Report envisaged that
“a significant number of patients are likely to require secure accommodation for longer than 18 months to two years”. Moreoever, there is no definition of LOS (prolonged or otherwise) in the forensic inpatient setting described in the most recent “Standards for Forensic Mental Health Services” by the Royal College of Psychiatrists and the Forensic Quality Network for Forensic Mental Health Services published in 2019 [
28].
Huband et al [
29] attempted in 2018 to define what constitutes “long stay” but found that reports were inconsistent across documents and thus it was not possible. They did, however, highlight some characteristics that may contribute to a long stay, such as
“seriousness of index offence, history of psychiatric treatment, cognitive deficit, severity of illness, history of violence, and history of substance misuse”. This persistent lack of definition of an appropriate LOS from national and international bodies poses significant challenges in research, as investigators have to come up with their own definitions.
In light of the high cost and complexity of care, it is essential to understand what a prolonged admission is and what factors contribute to a longer LOS in FP. The reasoning is dual: from a public health and economic perspective, the financial burden of forensic admissions is heavy and increasing. From a patient-centred care perspective, the potentially detrimental effects of a prolonged admission, albeit being poorly defined, in such restrictive environments must be mitigated. Understanding contributing factors means that services can develop mechanisms to better address prolonged stays and patients’ needs.
While most factors that contribute to LOS are non-modifiable, such as sociodemographic characteristics and forensic or psychiatric history, special consideration must be given to the modifiable factors. Forensic care takes place in a complex system that involves a range of interventions that could influence LOS, such as pharmacological, psychological [
30], and occupational therapies [
31] or risk management-focused activities.
To date, there has been no systematic review or meta-analysis focusing on the factors influencing LOS as a primary outcome. The main aims of this systematic review were to identify modifiable and non-modifiable factors associated with LOS in forensic psychiatric hospitals and to identify gaps in the relevant literature to formulate recommendations for future research.
Results
The search yielded a total of 1606 records across three databases. After de-duplication, 1176 records were left for screening. Following the title and abstract screening stage, 1122 records were excluded, and 54 records were eligible for full text review. Reasons for excluding full papers were recorded and reported (Fig.
1). Quality appraisal and data extraction were completed for 28 studies in total.
The list of included studies and their characteristics are presented in Table
1. In terms of study design, cross-sectional methodologies were largely overrepresented (23 studies), whereas fewer prospective or retrospective cohort [
40‐
43] and case‒control designs [
44] were identified.
A total of 13 studies were conducted in the UK [
42‐
54], five studies in the USA [
55‐
59] and two studies in Sweden [
40,
60] and the Netherlands [
61,
62]. The rest of the studies were conducted in Australia [
63], Canada [
64], Czechia [
37], Germany [
65], Ireland [
41], and Poland [
66]. The total sample size across all studies was 10,112, with the smallest being 14 [
64] and the largest being 2287 participants [
52].
The level of hospital security was not specified or applicable in seven studies [
37,
40,
53,
61,
63‐
65]. There were three studies that were conducted across more than one level of security (2 studies in medium- and high-security hospitals [
50,
52] and one study in low- and medium-security hospitals [
54]). The rest of the studies were conducted across only one level of security, with 10 studies conducted in maximum- or high-security hospitals [
44,
45,
51,
55‐
60,
62], seven studies in medium-security hospitals [
41,
42,
46‐
49,
66] and one study in low-security hospitals [
43].
Most studies reported mean ages of their samples, which ranged from 29.9 years [
44] to 52.97 years [
62]. The age of participants was not reported in four studies [
42,
51,
59,
65]. The majority of studies included predominantly males, with percentages ranging from 64.3% [
64] to 90.3% [
46] of all participants. Nine studies [
43‐
45,
49,
51,
55,
56,
60,
62] included only male participants. One study examined female patients only [
47]. One study only looked at forensic patients with a learning disability [
48].
The overall LOS was reported in most studies, but others reported means of the groups without giving an overall picture. LOS ranged from 0.32 years (equivalent to 116.9 days) in a study conducted in a maximum-security hospital in the USA [
55] to 14.95 years in a study conducted in both medium- and high-security settings in the UK [
50]. In the latter study, there was no differentiation in the reporting between the medium and high secure settings in terms of LOS.
Quality of studies
Following critical appraisal, seven studies were judged to be of good quality, nine studies of fair quality and the rest of the studies were poor. The most common issues identified were a lack of comparators or adequate control for confounders and offering no justification for sample size selection. Strong indicators across studies commonly included random selection and representativeness of the sample. Moreover, most investigators had direct access to secure and up-to-date medical records but did not account for reporting bias by the authors of the records. The outcomes of the studies were reported to varied extents, with a minority of the studies only including statistically significant results in their manuscripts. Details of the critical appraisal can be found in Table
4.
Table 4
Critical appraisal of included articles
Andreasson 2014 | 2 | Cohort | 3 | 1 | 3 | Good |
Chester 2018 | 5 | Cross Sectional | 3 | 2 | 3 | Good |
Davoren 2015 | 7 | Cohort | 3 | 1 | 2 | Good |
Duke 2018 | 9 | Cross Sectional | 3 | 1 | 3 | Good |
Edwards 2002 | 11 | Cohort | 3 | 1 | 3 | Good |
Messina 2011 | 19 | Cross Sectional | 3 | 1 | 3 | Good |
Smith 2004 | 26 | Cohort | 3 | 1 | 3 | Good |
Belfrage 2002 | 3 | Cross Sectional | 2 | 2 | 3 | Fair |
Brown 2009 | 4 | Cross Sectional | 2 | 1 | 3 | Fair |
Eckert 2017 | 10 | Cross Sectional | 2 | 2 | 3 | Fair |
Green 1998 | 14 | Cross Sectional | 2 | 1 | 3 | Fair |
Griffiths 2018 | 15 | Cross Sectional | 2 | 2 | 3 | Fair |
Long 2012 | 17 | Cross Sectional | 1 | 0 | 3 | Fair |
Pav 2022 | 22 | Cross Sectional | 2 | 1 | 3 | Fair |
Ross 2012 | 24 | Cross Sectional | 2 | 2 | 3 | Fair |
Wint 1994 | 28 | Case-control | 2 | 1 | 3 | Fair |
Alexander 2011 | 1 | Cross Sectional | 2 | 0 | 3 | Poor |
Colwell 2011 | 6 | Cross Sectional | 1 | 0 | 3 | Poor |
Dell 1987 | 8 | Cross Sectional | 2 | 0 | 3 | Poor |
Esan 2015 | 12 | Cross Sectional | 1 | 2 | 3 | Poor |
Gosek 2020 | 13 | Cross Sectional | 2 | 0 | 3 | Poor |
Hillbrand 1996 | 16 | Cross Sectional | 2 | 0 | 3 | Poor |
McKenna 2019 | 18 | Cross Sectional | 2 | 0 | 3 | Poor |
Moran 1999 | 20 | Cross Sectional | 2 | 0 | 3 | Poor |
Moulden 2020 | 21 | Cross Sectional | 1 | 0 | 1 | Poor |
Rodenhauser 1988 | 23 | Cross Sectional | 2 | 0 | 3 | Poor |
Shah 2011 | 25 | Cross Sectional | 2 | 0 | 3 | Poor |
Verstegen 2017 | 27 | Cross Sectional | 2 | 0 | 3 | Poor |
Studied variables
Overall, a total of 380 variables were examined. Variables that were similar were pooled and grouped. Table
2 shows all 79 factors that were assessed by at least two articles. Although the broad categories of variables were predetermined, the ones displayed are a result of what was identified in the review.
The most studied factor was the presence of a major index offence (defined as homicide, attempted homicide, serious bodily assault, armed robbery, kidnapping and arson), which was examined in 20 articles. This was followed by age (16 articles), psychosis-related diagnoses (excluding schizophrenia) (11 articles), male sex (nine articles), and history of substance use disorder (nine articles).
Some noteworthy variables that were examined by a single study and were not included in the discussion were treatment with clozapine [
66], treatment with > 1 antipsychotic [
66], persistent psychotic symptoms over the past 6 months of admission [
66], age at first psychiatric admission [
65], age at the onset of psychiatric symptoms [
43], and substance abuse during admission [
40].
Predictors of length of stay
The present review demonstrated varying quality of evidence for the different variables studied (Table
3). Most variables had no significant correlation with LOS. Significant correlations could be made for some factors. A lower GAF score (moderate evidence), an index offence of homicide or attempted homicide (moderate evidence), a legal status of criminal section with restrictions (moderate), or a diagnosis of schizophrenia-spectrum disorders (weak evidence) were correlated with a longer LOS.
Discussion
The primary aim of this systematic review was to identify factors that influence LOS in forensic inpatient settings and synthesise findings across high-quality studies. At the time of this study, no previous systematic review has focused entirely on the factors that influence LOS in forensic settings. A systematic review in 2015 examined factors influencing key forensic outcomes [
67]; however, this review included fewer articles and was not focused on LOS. The authors only extracted statistically significant results, and there was no data synthesis. There was no meta-analysis for reasons similar to those reported above. Similarly, a rapid review conducted in 2018 by Huband et al [
29] attempted to answer a set of different questions, including what constitutes a long stay, what are the characteristics of long-stay patients and what factors predict the LOS. This review utilised a rapid rather than a systematic review methodology and more restrictive inclusion criteria, such as factors that were only explored by multivariate analysis. Although this may control for confounding, it nevertheless excludes other robust statistical analyses. Most importantly, however, both Huband et al. and Sedgwick et al. described their findings narratively and did not conduct a data synthesis.
Main findings
While no meta-analysis was performed, we identified factors of interest that are supported by strong, moderate or weak quality of evidence, as explained above. The primary reason for identifying relevant variables remains to inform better care and discharge planning for forensic patients and allow for targeted treatment and distribution of health services; however, for most of the studies, no link to clinical practice was made.
One of the aims of this review was to identify modifiable factors for the purpose of directing potential future interventions. However, most factors with good quality evidence were non-modifiable. While most studies looked at historical information (e.g., psychiatric history, forensic history), few examined characteristics of treatment or institutional behaviour, and no studies looked into details of daily functioning with the exception of GAF scores. Some clinically interesting variables included refusal of treatment [
59] and involuntary treatment administration [
59], treatment with more than one antipsychotic [
66] and having treatment-resistant psychosis [
37], i.e., non-responsive to at least two different antipsychotics.
Overall, it appears that none of the sociodemographic variables appeared to be associated with LOS, including age, sex, ethnicity, employment, and family status, supported by strong and moderate quality of evidence. This differs from general psychiatric wards, where African-Caribbean patients appear to experience prolonged stays [
68] in the UK. A common difficulty in research remains the consistent measurement of these variables, as while frequently included in studies, authors use different terminology. We explore this complexity in more detail below.
Interestingly, civil legal status was not associated with longer admissions. This finding, however, is limited to psychiatric services in the UK, where the term refers to an involuntary admission under Part II of the Mental Health Act (MHA) 1983 [
69] (amended in 2007 [
70]) either for assessment or treatment. A criminal (or forensic) section refers to an involuntary admission under Part III of the MHA, which may include court-imposed restrictions, as indicated previously in this paper. Admissions under civil detention are largely into general psychiatric hospitals, but these patients can be found in forensic wards at varying degrees [
71] and are transferred usually as a result of increased risk and behavioural disturbance that cannot be managed on general psychiatric wards.
The initial expectation was that civil detention would correlate with shorter LOS due to the absence of an index offence and judiciary involvement. However, on closer inspection, these patients appear to pose greater management challenges, with more frequent episodes of aggression [
72,
73]. This may nullify the positive effect of not having committed an offence.
Similarly, an absence of index offence and institutional aggression during hospitalisation (including the requirement of seclusion and restraint) did not influence LOS. Institutional aggression was quantified by the number of incidents perpetrated by the individual throughout their admission. The common expectation in forensic psychiatric settings is that heightened risk would result in a prolonged stay to protect either the individual or society– in some cases, both. There is, however, a need for further investigation to corroborate these findings.
Conversely, and unsurprisingly, patients in the UK admitted under a criminal section with restrictions tended to stay in the hospital for longer periods of time. A criminal section implies that an individual has been convicted of a crime by court but is identified to have a mental disorder and is in need of medical assessment and/or treatment. A criminal section with restrictions means that the individual cannot be granted leave or be discharged without prior approval by the Ministry of Justice and is usually reserved for more serious offences.
The most examined variable across all studies was a major index offence. Unlike the findings of Sedgwick et al [
67] and Huband et al [
29], the evidence suggesting that having committed a broadly defined major offence prolongs the LOS was inconclusive, primarily due to the inconsistency of results across studies. A major index offence is associated with an increased risk of violence and more conservative discharge planning. Nonetheless, it is important to note that despite being inconclusive, there were nine studies [
40,
42,
47,
51,
55,
59,
63,
65,
66] where having committed a major index offence was reported as a predictor of longer LOS, and there is a need for more high-quality studies to decipher this relationship. On the other hand, having committed specifically homicide or attempted to commit homicide was associated with an increase in LOS.
With regard to rating scales, GAF was the only scale to have an association with LOS, and this was negative. This finding appears logical, as a higher GAF score is associated with fewer symptoms and better social and occupational functioning [
74]. GAF has also been found to have a strong predictive validity of one-year treatment outcomes [
75] and thus appears to be a reliable tool to monitor progress and recovery.
In terms of diagnosis, while schizophrenia-spectrum disorders (SSD) were linked to prolonged LOS, the quality of evidence was moderate and did not extend to the specific diagnosis of schizophrenia or other psychotic disorders. This may be due to limited high-quality studies and sample sizes, but it could also be partially explained by the relative dearth of evidence for optimal management of SSD such as schizophrenia [
76] and the widespread recognition of the burden of such a diagnosis [
77]. One key aspect of the impact of diagnosis on LOS, however, would be the existence of comorbidities, such as substance use disorder, which has a very high prevalence among patients with SSD [
78]. This association should be the focus of future research.
One key finding across this review was a consistently higher LOS among patients in the UK, particularly in comparison to the USA. The importance of different organisations and the provision of forensic psychiatry cannot be understated. However, at least when comparing services of similar nature (i.e., high-security settings), such a stark difference could be explained by several factors. In US secure hospitals, the focus is on competency restoration to stand trial, rather than long-term treatment. If competency cannot be restored, then charges must be dropped, and the individuals are either released or admitted under a civil Sect. [
79]. There has been a recent drive to reduce the prison population and a focus on community services in the USA, which may lead to faster movement between services [
80,
81]. Additionally, the influence of a public, nationalised health system in the UK might also be relevant, as is the court diversion system [
82]. Very few studies from other countries were included, and thus, no inferences could be made.
Variation of definitions
One striking finding of this review was the often-extreme variations in definitions of relevant variables. Some of these variations, such as the level of security or lack thereof, were expected, as FP is structured differently across the world and is dependent on local legislation [
83].
This variation included the primary outcome variable, LOS, defined mostly as either LOS at the time of the study or LOS until discharge. As there is no accepted threshold of a “lengthy” inpatient stay, authors have had to devise their own definition which differ across studies. A total of 19 out of 28 studies did not define LOS at all and rather drew comparisons based on the LOS of the included patients at a specific point in time. The most common cut-off point used was the 2-year mark [
42,
44,
46,
51,
60,
65]. Chester et al [
50] and Duke et al [
52] defined a prolonged LOS as more than 5 years in medium secure care and 10 years in high secure care and compared patients in these groups with those who did not meet the criteria. For Alexander et al [
48], the difficult-to-discharge group was determined based on the median LOS of the patients who had been discharged at the time of the study. The need for a universal definition remains a very important point for consideration in forensic research.
The definition of ethnicity also poses a challenge, as there is no universally agreed system of classification, and such a classification would largely depend on local context [
84]. The relationship between ethnicity and outcomes may be affected by minority status, which is not constant globally. In our review, different studies used different ways of reporting or had much broader categorisations, and it is possible that there may be associations missed due to the difficulties in grouping.
The definition and measurement of age was surprisingly varied. The definitions encountered in the included articles were age on admission, age on discharge, and age during the study. In terms of association with LOS, these varying definitions represent different variables and should be studied further to corroborate the accuracy of the conclusion itself.
Another important finding to highlight was the lack of effect size measurement and differences in statistical analysis and reporting across multiple studies.
P values alone are not sufficient to identify an association, and a statistically significant result is much more likely with larger sample sizes [
85]. Effect sizes are thus necessary to understand the extent of the difference between the groups and provide an added layer of security that the result is not only statistically but also clinically significant.
Limitations
This review aimed to answer a complex question. The authors focused on synthesising evidence across different countries, cultures and economic systems, all factors that are imperative in forensic psychiatry. There is thus an inherent weakness in any similar quantitative‒qualitative analysis and one that is unlikely to be resolved in the future.
As with all systematic reviews, and despite best efforts to include all relevant terms and keywords in the search strategies, it is possible that relevant and high-quality studies might have been missed as non-English articles were excluded and the search was limited to three databases. However, the search strategy was initially piloted across all key databases, and those with relevant results were included in the review.
As the present review focused on observational studies, the evidence identified carries several of the limitations associated with this design. Cross-sectional designs in particular—which were overrepresented—have limited capacity to assess causal relationships between LOS and variables of interest. It is, however, a cost-effective and easy way to look at a snapshot of information, and it is preferred where access to detailed medical records is readily available.
Most commonly, forensic mental health services include 3 levels of security. This is the case for the UK [
5], Sweden [
86], Poland [
87], Canada [
88], Germany [
89] and Australia [
90]. Dutch forensic services provide four levels of security, determined by the patient’s legal and clinical status [
8]. In some jurisdictions, all three levels are provided on the same site. However, even where the level of security appears similar, service provision and expectations may not be comparable, as demonstrated by the stark LOS difference among high-secure hospitals in the US and UK. It would be important for future research to focus on specific levels of security that are aligned both in terms of risk stratification and scope of practice, particularly when conducting international reviews.
A key limitation in both the available evidence and this study is highlighted by the lack of a meta-analysis. Studies looking at LOS in forensic settings are extremely heterogeneous in terminology, measurement of variables, statistical analysis, measurement of effect sizes, and even in reporting of results. Even beyond the heterogeneity, most of the included studies explored a large number of variables but without having a prior hypothesis and without reporting effect sizes. This is not the case only in FP research but also in general psychiatry and was highlighted by a 2011 review on the LOS in general psychiatric inpatients in the USA [
91]. It is the authors’ hope that this review can bring this issue forward and encourage future authors to follow a list of recommendations that have been compiled and can be found below.
Lastly, an important observation was data dredging. Data dredging increases the chance of identifying possible associations, particularly statistically significant ones, through introducing multiple variables or multiple categorisations of variables [
92,
93]. Across several studies in this review, the list of variables was lengthy, and among the statistically significant associations identified, some often lacked practical or clinical significance and had poor generalisability outside the study population. Such variables could include inappropriately grouped diagnoses (e.g. intellectual disability and dementia) or extensive categorisation of demographic history (e.g. employment or social history).
Implications for future research and practice
The present body of work adds significant value to the literature, if only for the gaps that have been identified and described at length. A set of variables, including having committed homicide/attempted homicide, a criminal status with restrictions and schizophrenia-spectrum disorders, were found to have evidence of varying quality to suggest that they may prolong LOS. These findings should be re-examined using higher quality research to prove this association and understand how care can be adapted to account for them– perhaps through the development of different pathways for rehabilitation, according to the details of the index offence (aside the levels of security). On the other hand, the importance of GAF scores has been highlighted, as they are negatively associated with LOS and can be a quick and efficient tool to use in daily practice. While the evidence is moderate, it implies that both severity of illness and daily functioning are important aspects to consider across care planning, and future researchers are encouraged to use them as guidance for examining response to rehabilitation.
Regarding heterogeneity, unless a set of commonly examined variables are standardised, it is unlikely that a meta-analysis will be possible in the future. A set has been compiled below, and the authors encourage investigators to consider this in their practice. It is also crucial that modifiable variables, particularly in terms of treatment (pharmacological and non-pharmacological) characteristics, are explored. While the evidence for psychotherapy treatment was inconclusive due to a lack of high-quality studies and consistency, Long et al [
47] identified that engagement reduced LOS, while Moulden et al [
64] reported a reduction in LOS for patients receiving Dialectical Behaviour Therapy (DBT), albeit in a very small sample. Another non-pharmacological treatment option that ought to be explored is occupational therapy, as it plays a significant role in recovery [
94]: Messina [
57] highlighted a shorter LOS among those with higher attendance at therapeutic groups while in the hospital, and there was evidence in this review to suggest that the same applies to higher GAF scores [
40,
55,
57], which are partly based on psychosocial and daily functioning.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.