Introduction
Outcomes in forensic mental health services are varied and often poor. In 2007, around 50% of patients detained under the legal category “psychopathic disorder” in the United Kingdom had a stay in a hospital exceeding 10 years.Reference Rutherford and Duggan 1 Lengthy admissions were also identified in one German study that found that some patients stayed as long as 43 years.Reference Ross, Querengasser, Fontao and Hoffmann 2 Further, prospective follow-up studies of discharged mentally disordered offenders (MDOs) have shown a relatively high rate of reoffending, with 1 in 8 men being convicted for another grave offense after discharge from medium security services in the UK.Reference Coid, Hickey, Kahtan, Zhang and Yang 3 This has significant implications in terms of public protection, cost to the taxpayer, and the ethical position of detaining individuals for treatment, which may not be efficacious.
Current methods of predicting outcome include a multidisciplinary assessment of need (ie, criminogenic and clinical factors that require intervention), which often involves the use of structured professional judgment instruments to assess the level of risk, generally in the context of treatment planning.Reference Glorney, Perkins and Adshead 4 , Reference Gudjonsson and Young 5 The Historical Clinical Risk Management (HCR-20)Reference Webster, Douglas, Eaves and Hart 6 scheme is an example of this, and has shown good predictive validity for future violence.Reference O’Shea, Mitchell, Picchioni and Dickens 7 The psychopathy checklist (PCL-R) has grown in popularity as a quasi-risk-assessment tool due to the demonstrated link between high PCL-R scores and both inpatient violence and community reoffending.Reference Hare, Clark, Grann and Thornton 8 , Reference Walters 9 However, while these assessment tools supersede unstructured clinical decision making,Reference Hanson and Morton-Bourgon 10 they still rely on clinical judgment/decision making to draw conclusions. This is particularly relevant when considering the forensic population, many of whom are diagnosed with disorders that are characterized by deceptive behaviors [eg, antisocial personality disorder, taken from the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5): “Deceitfulness: dishonesty and fraudulence; misrepresentation of self; embellishment or fabrication when relating events.”). Further, it is plausible that offenders may wish to present as low risk in order to secure early discharge, adding a further complication for clinicians making assessments of need.
A recent reviewReference Singh, Fazel, Gueorguieva and Buchanan 11 has identified wide variation in the rate of violence observed from those who are classified as “high-risk” using 9 of the most widely used risk assessment tools, both within and between risk assessment schemes. When considering this alongside evidence suggesting that there is very little change in HCR-20 scores across an individual’s stay in a high-security hospital despite them engaging in risk-focused treatment, it calls into question the clinical utility and sensitivity of such tools. For example, Morrissey et al Reference Morrissey, Beeley and Milton 12 found a change of 1 point or less (possible score range 0–10) in the dynamic clinical and risk scales, across 5 years. Although the clinical scale scores were significantly lower in the group about to be discharged compared to those who were still resident in the hospital, the risk scale and total scores were comparable, suggesting that these scales are either not sensitive enough to capture a reduction in risk, or that clinicians are not regarding this information as useful in their decision making about discharge.
A growing body of evidence has shown than an array of neurobiological factors is associated with violent behaviors in mentally disordered populations,Reference Barkataki, Kumari and Das 13 – Reference Kumari, Das and Taylor 17 and it may be that some of these correlates could also assist clinicians working in forensic services to make decisions about treatment planning, risk, and discharge. Such factors, which are objective and measurable, reduce the likelihood of errors of judgment being made. Consideration of these factors alongside methods already employed could enhance the amount of information available, and thus potentially improve decision making or identify areas of outstanding need. This could theoretically lead to improved outcomes for patients, the public, and the taxpayer via more appropriate treatments being offered, fewer premature discharges, and more efficient services, respectively.
This systematic review aimed to identify and evaluate studies that have assessed objective predictors of outcome in forensic mental health services (ie, did not rely on self-report or clinical judgment) to gain a perspective on how far these correlates have been used by the scientific and clinical community, and to assess the potential usefulness of such markers in further research and subsequently in clinical practice.
Method
OVID-Medline, Embase, and PsychInfo (inception–January 2015) databases were searched using the following 4 terms combined with AND:
1. predict* OR prognos* or marker
2. outcome OR length of stay OR duration of stay OR length of hospitalization OR duration of hospitalization OR reoffen* OR recidiv* OR violen* OR function*
3. mental disorder OR psychiatr* OR mental ill*
4. forensic OR secur* OR incarcerated
A screen of the results for relevance was then conducted on a title/abstract basis. If insufficient information was given in the abstract, the full text was retrieved before making a decision. Studies were assessed for inclusion against the following criteria:
1. All participants were MDOs admitted to inpatient forensic psychiatric services. For the purposes of this review, an MDO is defined as an offender with a diagnosed mental disorder, who is deemed to require treatment in psychiatric services. Individuals residing in prison who have a mental disorder were not included, as it is highly likely that individuals who are deemed treatable within prison (as opposed to secure psychiatric hospitals) are qualitatively different. Further, “specialist” offender groups (adolescents, eg, Letourneau and ArmstrongReference Letourneau and Armstrong 18 ; learning disability, eg, Bastert et al Reference Bastert, Schlafke, Pein, Kupke and Fegert 19 ) were excluded to keep the study samples as homogeneous as possible.
2. We accepted studies that included an objective predictor of outcome (as defined as a factor that does not rely on clinical judgment or self-report, eg, biological, neuropsychological, demographic factors), with outcome defined as one of the following: length of stay, violent incidents (inpatient or community), functioning, clinician-rated risk/need.
3. Only primary research articles with an abstract were included (eg, not theses or reviews). The reference lists of relevant reviews were examined to identify any papers not returned by the initial search.
4. Studies were only included if they used a prospective, or pseudo-prospective, design (ie, looking forward over time) to assess predictive ability. Studies that reported on the ability of static (ie, demographic) factors to predict outcome were also included; these did not necessarily need to be prospective, as static factors by definition are temporally stable.
5. Studies were excluded if they were reviewing the predictive validity of risk assessment tools. This literature is large and robust and has been reviewed elsewhere.Reference Dolan and Doyle 20 – Reference McDermott and Holoyda 22 Further, these tools require the assessment of a combination of demographic and clinical factors that may relate to risk collectively, but often individual item predictive validity is not given.
6. Articles referring solely to competency to stand trial were also excluded. This intervention involves treating the underlying disorder and educating the individual about the American legal system so that they are able to stand trialReference Zapf and Roesch 23 ; it is not analogous with the typical treatment MDOs receive (ie, the focus is to restore competency).
Data extraction
For each study, predictors associated positively with the outcome variable of interest (eg, associated with an increased likelihood of violence), predictors with a negative association (eg, associated with a decreased likelihood of violence), and examined variables with no relationship (eg, no relationship to violence) were extracted. Studies were examined, and any factors identified by the authors as “statistically significant” were extracted. This included significant differences between relevant groups (eg, between reoffenders and non-reoffenders) and significant positive or negative predictors of outcome. Variables that were examined by the authors but had no significant effects were included in the “no relationship” category.
Predictor variables were then compiled into a spreadsheet, and studies that reported on the same broad predictors for the outcome of interest were recorded. Categories that were conceptually similar but perhaps not described in the exact same terms (for example, “severity of offense” and “a violent or homicide offense”) were combined to reduce the number of discrete predictors.
Results
The search returned 1896 results. See Figure 1 for the flowchart of study selection.
Figure 1 Flowchart of study selection.
Fifty articles, which included data on objective predictors of outcome in forensic mental health services, were retained in the final review. Studies were categorized into 3 broad outcome groups: those reporting on predictors of (1) inpatient violence, (2) length of stay in forensic inpatient services, and (3) community reoffending. Further, the types of predictor could also be delineated into 3 categories. These were (i) demographic (42 studies), (ii) neuropsychological/neurophysiological (4 studies), and (iii) biological (4 studies) predictors. The term “demographic” is used here as a broad, all-encompassing term to refer to static, historical factors, including clinical, offense-related, developmental, institutional, and sociodemographic factors.
Predictors of inpatient violence
Demographic predictors
Thirty-eight separate demographic factors across 8 studiesReference Ball, Young, Dotson, Brothers and Robbins 24 – Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 were identified as predictors of inpatient violence (Table 1). Of these, 16 factors were considered in more than 1 study. Only 1 factor, previous psychiatric admissions, was found to be associated with inpatient violence in the majority of studies that examined it; 2 studies found a positive relationship between number of previous psychiatric admissions and inpatient violence,Reference Ball, Young, Dotson, Brothers and Robbins 24 , Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 whereas 1 study found a null effect.Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 One of these studies assessed seclusion episodes as opposed to inpatient violence directlyReference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 ; however, all seclusion incidents were related to aggressive behavior, apart from 1 episode of self-harm.
Table 1 Demographic predictors and their association with inpatient violence
ASPD: antisocial personality disorder; PD: personality disorder; SUD: substance use disorder.
Table gives number of studies with a positive relationship to inpatient violence, over the number of studies examining this factor.
Factors with * include some studies which suggest a negative effect, ie, the factor in question is associated with reduced inpatient violence (exact figures are presented in Figure 2).
Another demographic factor, young age, was examined by 6 studies, of which 3 found a positive associationReference Hoptman, Yates, Patalinjug, Wack and Convit 27 , Reference Rasmussen and Levander 29 , Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 and 3 found no association.Reference Ball, Young, Dotson, Brothers and Robbins 24 , Reference Dietz and Rada 25 , Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 Similarly, a history of violence was found to be associated with inpatient violence in 2 studies,Reference Ball, Young, Dotson, Brothers and Robbins 24 , Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 and not associated in 3 studies.Reference Rasmussen and Levander 29 – Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31
Other factors examined by 2 or more studies and found to be unrelated to inpatient violence are listed in Figure 2. Notably, a history of substance use,Reference Hoptman, Yates, Patalinjug, Wack and Convit 27 , Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 , Reference Rogers, Watt, Gray, MacCulloch and Gournay 30 , Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 , diagnosis,Reference Ball, Young, Dotson, Brothers and Robbins 24 , Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 and genderReference Ball, Young, Dotson, Brothers and Robbins 24 , Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 , Reference Rogers, Watt, Gray, MacCulloch and Gournay 30 , Reference Thomas, Daffern, Martin, Ogloff, Thomson and Ferguson 31 did not emerge as consistent predictors across studies (Figure 2).
Figure 2 Demographic predictors examined by at least 2 studies and their association with inpatient violence.
Neuropsychological predictors
One studyReference Foster, Hillbrand and Silverstein 32 reported the ability of neuropsychological assessments to predict aggression among 23 male forensic inpatients (n=16 with a principal diagnosis of schizophrenia). Aggressive behavior was monitored over the year following testing using the Overt Aggression Scale.Reference Yudofsky, Silver, Jackson, Endicott and Williams 33 The results demonstrated that poor visuospatial processing [assessed by the Judgment of Line Orientation Test (JLOT)Reference Benton, Varney and Hamsher 34 ], poor cognitive inhibition [scores on the Stroop Color/Word Test (SCWT)Reference Stroop 35 ], and the number of misperceptions of an angry voice in an emotional recognition test could reliably predict the frequency of subsequent aggression. Scores from the JLOT and SCWT were also significantly correlated with the severity of aggression.
A similar studyReference Enticott, Ogloff, Bradshaw and Daffern 36 reported a 5-week follow-up of 10 forensic inpatients. Contrary to expectation, performance on a measure of behavioral inhibition (the Stop TaskReference Enticott, Ogloff and Bradshaw 37 ) was better at a trend level among those who were involved in aggressive incidents compared to those who were not, suggesting that those who were more impulsive were involved in fewer incidents. However, this study was significantly limited by its small sample size and low rate of recorded incidents (12 incidents, conducted by 5 patients), and thus the results must be interpreted with caution. In addition, no information regarding diagnosis is given by this study, leaving questions as to the generalizability of the results to other populations.
A further studyReference Murphy 38 examined clinical outcome, need, and risk, which are all facets sensitive to inpatient violence, in a high-security hospital. Thirty newly admitted men with schizophrenia were assessed on a number of neuropsychological tasks, including an assessment of IQ, processing speed, and working memory using the Wechsler Adult Intelligence Scales,Reference Wechsler 39 in addition to the Trail Making TestReference Reitan and Wolfson 40 and the SCWT.Reference Stroop 35 Further, 2 social cognitive tasks were conducted, the Revised Eyes Task Reference Baron-Cohen, Wheelwright, Hill, Raste and Plumb 41 and a Modified Advanced Theory of Mind Test.Reference Frith and Corcoran 42 Outcome measures included the Health of the Nation Scales–Secure version (HoNOS), the Camberwell Assessment of Need–Forensic version (CANFOR), and the HCR-20, which assess clinical, social, and functional outcome; need; and risk, respectively, at 3-year follow-up. Although a number of nonsocial cognitive tasks showed utility in predicting some outcomes of interest (eg, Trail Making part B was significantly correlated with scales from the HoNOS, the total CANFOR score, and HCR-20 risk management scale), the overwhelmingly most predictive test was the Revised Eyes Test. After controlling for all other variables, the Revised Eyes Test score could significantly predict total CANFOR score, the risk management score on the HCR-20, and the social scale score of the HoNOS.
Thus, patients with schizophrenia who were less able to interpret emotional information from the eyes were likely to have higher ratings of unmet need, poorer social functioning, and a higher level of assessed risk. This may be relevant to the Violence Inhibition Mechanism theory,Reference Blair 43 according to which poor interpretation of negative facial expression removes inhibitory influences that serve to stop violent behavior through negative reinforcement of the unwanted (aggressive) behavior. Poor theory of mind may also reduce the capacity for cognitive empathyReference Mathersul, McDonald and Rushby 44 or understanding typical social rules,Reference Roncone, Falloon and Mazza 45 which could lead to social conflict and potentially violent behavior.
Finally, one demographic study extracted evidence of “cognitive impairment” (present/absent) from patient files, and found that this was a significant predictor of frequent violent behavior among inpatients.Reference Lussier, Verdun-Jones, Deslauriers-Varin, Nicholls and Brink 28 Although there is not detailed explanation of the nature or severity of cognitive impairment in these participants, this study supports the assertion that cognitive dysfunction may be related to aggressive behaviors.
Biological predictors
Four studiesReference Hillbrand, Spitz and Foster 46 – Reference Spitz, Hillbrand, Foster and Svetina 49 examined biological predictors of inpatient violence, although 3 of these studiesReference Hillbrand, Spitz and Foster 46 , Reference Hillbrand, Spitz, Foster, Krystal and Young 47 , Reference Spitz, Hillbrand, Foster and Svetina 49 were conducted within the same sample. TwoReference Hillbrand, Spitz and Foster 46 , Reference Repo-Tiihonen, Paavola, Halonen and Tiihonen 48 related to serum cholesterol levels, while 2Reference Hillbrand, Spitz, Foster, Krystal and Young 47 , Reference Spitz, Hillbrand, Foster and Svetina 49 were concerned with creatine kinase elevations.
Serum cholesterol
The serum cholesterol levels of 106 forensic inpatients at admission were examined, and subsequent aggressive incidents toward others or themselves over the following 2 years (pseudo-prospective review of medical records) were followed up.Reference Hillbrand, Spitz and Foster 46 The sample was divided into high (≥200 mg/dl) and low (<200 mg/dl) cholesterol groups, and the difference in aggressive incidents (frequency, severity, and type) was investigated. While the 2 groups did not differ with regard to severity or type of aggression, the frequency of aggression in the low cholesterol group was significantly increased. Interestingly, the relationship between cholesterol level and frequency of aggression was nonlinear, with aggression being most frequent within the range 160–170 mg/dl.
A similar investigationReference Repo-Tiihonen, Paavola, Halonen and Tiihonen 48 was conducted in order to determine an optimum cut-off point for predicting aggression using serum cholesterol levels. Using male participants detained in a forensic hospital, the sample was divided into those who had been secluded at least once over a 28-month period (n=195) and those who had not been secluded (n=202). When comparing these groups, the secluded group had significantly lower total serum cholesterol. Using receiver operating characteristic analysis, the optimum cut-off for predicting those who would be secluded for any reason was 5.3 mmol/l. However, for patients who spent a longer duration of their detention in seclusion for aggression/self-harm, perhaps considered the most frequently aggressive patients, the optimum cut off was 4.3 mmol/l. When converted into mg/dl (as used in the previous investigation), this equates to approximately 165 mg/dl, which is highly consistent with the 160–170 mg/dl range found for the most frequently violent patients in the aforementioned study.Reference Hillbrand, Spitz and Foster 46 The difference in cholesterol level between the 2 groups was independent of body mass index and medication.
Creatine kinase
One studyReference Hillbrand, Spitz, Foster, Krystal and Young 47 investigated the predictive utility of creatine kinase (CK) as a marker of aggressive behavior in 164 male forensic inpatients, again using a pseudo-prospective design. CK is an enzyme involved in in-situ energy production in cells.Reference Wallimann, Wyss, Brdiczka, Nicolay and Eppenberger 50 The sample was divided into high or low aggression, based on a median split procedure on scores for the severity, frequency, and type of violence as determined by the Overt Aggression Scale (verbal vs physical). In all 3 comparisons (severity, frequency, and type), the CK levels were significantly higher in those who were more frequently violent, engaged in more severe violence, and in those who used physical as opposed to verbal aggression.
An association between assaultiveness and use of restraint prior to CK levels being determined was also observed. Those who had been assaultive during their admission and those who had been restrained had higher observed CK levels than those who had not. Importantly, a significant interaction between these factors was observed, in that those who were assaultive/restrained and then engaged in subsequent violence had significantly increased CK levels (around a 5-fold increase) during their admission compared to those who were assaultive/restrained and then not violent. This suggests that, of those patients who present management problems during their admission, the likelihood of subsequent aggression can be gauged by assessing CK levels. These findings were irrespective of diagnosis, recent physical exercise, recent accidents, or recent intramuscular medication. However, 2 caveats were noted: (1) these findings were only significant in those patients taking antipsychotic medication, and (2) CK levels were not sensitive to change in aggression, ie, they did not increase prior to an aggressive incident, nor decrease afterward. Despite this, the authors assert that using a >200 U/l cut-off could correctly predict future assaults in 94% of cases, compared to using prior assaultiveness alone as a predictor (64%).
A further study on the same sampleReference Spitz, Hillbrand, Foster and Svetina 49 examined CK as a function of ethnicity and aggression. While the results demonstrated that CK levels were higher in African Americans than in Caucasians, and that African Americans were more likely to be physically aggressive compared to Caucasians, the increased levels of CK observed in African Americans was still significant even when the effect of aggression was covaried out.
Predictors of length of stay
Demographic predictors
A total of 44 diverse predictors were examined in relation to length of stay, with 25 of these being examined by more than one study (Figure 3). The factor that most studies examined was severity of offense. Unsurprisingly, 9Reference Ross, Querengasser, Fontao and Hoffmann 2 , Reference Baldwin, Menditto, Beck and Smith 51 – Reference Andreasson, Nyman and Krona 58 out of 10Reference Moran, Fragala, Wise and Novak 59 studies found that a more “severe” offense was related to a longer length of stay. This is supported by 2 studies that examined the effect of a restriction order on length of stay (administered to patients in the UK who are considered to be particularly high risk), which both showed a lengthening effect.Reference Andreasson, Nyman and Krona 58 , Reference Brown and Fahy 60 Three studiesReference Long and Dolley 53 , Reference Rice, Quinsey and Houghton 54 , Reference Green and Baglioni 57 found that having a psychotic disorder was associated with a longer length of stay, although 1 study found the opposite (shorter stay),Reference Moran, Fragala, Wise and Novak 59 and 1 found no significant effect.Reference Andreasson, Nyman and Krona 58 In addition, 3 studies found no effect for “diagnosis” on length of stay (which included psychosis)Reference Edwards, Steed and Murray 52 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Steadman, Pasewark, Hawkins, Kiser and Bieber 56 ; however, it is notable that in 2 of these studies, there was a very small proportion of offenders not diagnosed with a psychotic illness, suggesting limited sensitivity to find an effect. Two out of 3 studies that examined absconding during hospitalization found that this was associated with a longer stay.Reference Ross, Querengasser, Fontao and Hoffmann 2 , Reference Castro, Cockerton and Birke 61
Figure 3 Demographic predictors examined by at least 2 studies and their association with length of stay.
Previous offenses was found to be unrelated to length of stay in all 6 studies that examined this,Reference Ross, Querengasser, Fontao and Hoffmann 2 , Reference Baldwin, Menditto, Beck and Smith 51 , Reference Edwards, Steed and Murray 52 , Reference Skipworth, Brinded, Chaplow and Frampton 55 – Reference Green and Baglioni 57 , which provides strong evidence that it is the severity, as opposed to the extent, of offending which is implicated in how long MDOs remain in services. Other examined factors for which no clear association emerged are detailed in Figure 3 and Table S2 (available online in the Supplementary Material).
Neuropsychological/neurophysiological and biological predictors
No studies examining the effect of neuropsychological/neurophysiological or biological variables on length of stay were identified.
Predictors of community reoffending
Demographic predictors
Community reoffending, encapsulating re-arrest, readmission, recidivism, etc, was the outcome of interest in the majority of the articles (n=25). Again, a large and diverse number of factors (total 66) was considered across studies (Table 2), with 27 factors only considered in a single study. The most frequently examined predictor was previous offending, which was examined by 18 studies.Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Edwards, Steed and Murray 52 , Reference Rice, Quinsey and Houghton 54 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Castro, Cockerton and Birke 61 – Reference Tennent and Way 74 Of the studies that examined previous offending, 67% found an association with reoffending. Young age at admission or discharge was investigated in 15 studies,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Hillbrand 26 , Reference Edwards, Steed and Murray 52 , Reference Rice, Quinsey and Houghton 54 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Baxter, Rabe-Hesketh and Parrott 62 – Reference Maden, Rutter, McClintock, Friendship and Gunn 67 , Reference Phillips, Gray and MacCulloch 69 , Reference Duncan, Short, Lewis and Barrett 73 , Reference Quinsey and Maguire 75 , Reference Zonana, Bartel, Wells, Buchanan and Getz 76 with 67% finding a positive effect, while the effect of a shorter length of stay was examined in 12 studies,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Friendship, McClintock, Rutter and Maden 66 – Reference Philipse, Koeter, van der Staak and van den Brink 68 , Reference Quinsey, Rice and Harris 70 , Reference Duncan, Short, Lewis and Barrett 73 – Reference Rice, Harris and Quinsey 77 and 50% found that it was associated with reoffending.
Table 2 Demographic predictors and their association with recidivism (reoffending/rearrest/readmission)
SUD: substance use disorder; PD: personality disorder; CD: conduct disorder; SES: socio-economic status.
Table gives number of studies with a positive relationship to reoffending/rearrest/readmission, over the number of studies examining this factor. Factors with * include some studies which suggest a negative effect, ie, the factor in question is associated with a reduced likelihood of recidivism (exact figures given in Figure 4).
Figure 4 Demographic predictors examined by at least 3 studies and their association with reoffending/rearrest/readmission.
Male gender,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Edwards, Steed and Murray 52 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Buchanan 63 , Reference Buchanan and Leese 64 , Reference Maden, Rutter, McClintock, Friendship and Gunn 67 , Reference Phillips, Gray and MacCulloch 69 , Reference Duncan, Short, Lewis and Barrett 73 , Reference Zonana, Bartel, Wells, Buchanan and Getz 76 , Reference Rice, Harris and Quinsey 77 race,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Edwards, Steed and Murray 52 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Cohen, Spodak, Silver and Williams 65 – Reference Maden, Rutter, McClintock, Friendship and Gunn 67 , Reference Phillips, Gray and MacCulloch 69 , Reference Zonana, Bartel, Wells, Buchanan and Getz 76 , Reference Maden, Friendship, McClintock and Rutter 78 and being singleReference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Rice, Quinsey and Houghton 54 , Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Cohen, Spodak, Silver and Williams 65 , Reference Phillips, Gray and MacCulloch 69 – Reference Rice, Harris, Lang and Bell 72 , Reference Quinsey and Maguire 75 , Reference Rice, Harris and Quinsey 77 were investigated in 10 studies each, with positive findings indicated in 40%, 20%, and 30%, of studies, respectively. Other frequently examined factors included previous violence (9 studies,Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Friendship, McClintock, Rutter and Maden 66 – Reference Philipse, Koeter, van der Staak and van den Brink 68 , Reference Quinsey, Rice and Harris 70 – Reference Rice, Harris, Lang and Bell 72 , Reference Rice, Harris and Quinsey 77 , Reference Reiss, Grubin and Meux 79 44% positive finding), young age at time of offense (8 studies,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Edwards, Steed and Murray 52 , Reference Cohen, Spodak, Silver and Williams 65 , Reference Philipse, Koeter, van der Staak and van den Brink 68 , Reference Phillips, Gray and MacCulloch 69 , Reference Rice and Harris 71 , Reference Rice, Harris, Lang and Bell 72 , Reference Tennent and Way 74 50% positive finding), employment (8 studiesReference Rice, Quinsey and Houghton 54 , Reference Philipse, Koeter, van der Staak and van den Brink 68 , Reference Quinsey, Rice and Harris 70 – Reference Rice, Harris, Lang and Bell 72 , Reference Tennent and Way 74 , Reference Quinsey and Maguire 75 , Reference Reiss, Grubin and Meux 79 ; 34% found that it was negatively associated with reoffending, and the remainder found no association), previous psychiatric admissions (10 studies,Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Maden, Rutter, McClintock, Friendship and Gunn 67 , Reference Phillips, Gray and MacCulloch 69 – Reference Zonana, Bartel, Wells, Buchanan and Getz 76 10% found positive effect), and substance use (7 studies,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Cohen, Spodak, Silver and Williams 65 , Reference Rice and Harris 71 , Reference Quinsey and Maguire 75 , Reference Zonana, Bartel, Wells, Buchanan and Getz 76 , Reference Howard, McCarthy, Huband and Duggan 80 43% positive finding).
In terms of diagnostic groups, personality disorder (PD) was examined by 9 studies,Reference Coid, Hickey, Kahtan, Zhang and Yang 3 , Reference Philipse, Koeter, van der Staak and van den Brink 68 – Reference Rice, Harris, Lang and Bell 72 , Reference Quinsey and Maguire 75 , Reference Howard, McCarthy, Huband and Duggan 80 , Reference Bailey and Macculloch 81 with 78% of studies finding a positive association with reoffending. Six studies examined psychosis,Reference Rice, Quinsey and Houghton 54 , Reference Baxter, Rabe-Hesketh and Parrott 62 , Reference Philipse, Koeter, van der Staak and van den Brink 68 , Reference Rice and Harris 71 , Reference Tennent and Way 74 , Reference Quinsey and Maguire 75 with 50% finding that this was negatively associated, and the remainder finding no association with reoffending. However, 4 studiesReference Edwards, Steed and Murray 52 , Reference Skipworth, Brinded, Chaplow and Frampton 55 , Reference Friendship, McClintock, Rutter and Maden 66 , Reference Maden, Rutter, McClintock, Friendship and Gunn 67 found that “diagnosis” as a predictor (encapsulating PD and psychosis) was unrelated to reoffending, somewhat weakening these initially strong findings. This differential pattern of results likely reflects the diagnostic homogeneity of these 4 studies, in which the vast majority of patients had psychotic disorders and only small numbers were diagnosed with personality disorder (8%, 8%, 13%, and 9%, respectively), whereas studies that had more variance in diagnostic group, and thus more power to detect significant differences, tended to find positive results. For example, in a sample in which the number of participants with PD or psychosis was approximately equivalent,Reference Bailey and Macculloch 81 PD emerged as a factor associated with reoffending.
In addition, 1 studyReference Quinn and Ward 82 that examined “success of transfer” from high security to medium security found no significant demographic predictors. This outcome was deemed conceptually distinct from any of the 3 main outcome groups (as an unsuccessful transfer could be due to inpatient violence or worsening of symptoms, for example), and thus the predictors were not included in the variable count. A list of the variables examined in this study is included in Table S1.
Neuropsychological and neurophysiological predictors
Six demographic studies examined the effect of IQ on reoffending.Reference Quinsey, Rice and Harris 70 – Reference Rice, Harris, Lang and Bell 72 , Reference Tennent and Way 74 , Reference Quinsey and Maguire 75 , Reference Reiss, Grubin and Meux 79 It is notable, however, that these studies did not conduct a formal assessment of IQ; scores were extracted from patient files, which may have limited the findings in terms of standardizing the assessment tool used. Further variation may also have been introduced in terms of when the assessment was conducted (ie, at admission, during an acute phase of illness, during court proceedings, etc), which was not evident from the reviewed papers. Five of these investigations found no relation to reoffending,Reference Quinsey, Rice and Harris 70 – Reference Rice, Harris, Lang and Bell 72 , Reference Tennent and Way 74 , Reference Quinsey and Maguire 75 while one study found a positive association (ie, those with lower IQ were more likely to reoffend).Reference Reiss, Grubin and Meux 79
Howard and LumsdenReference Howard and Lumsden 83 assessed the relationship between the contingent negative variation (CNV) event-related potential during a go/no-go task and community reoffending in a sample of 44 admissions to a high-secure forensic hospital. The CNV during this task has been correlated with measures of impulsivityReference Howard, Fenton and Fenwick 84 and has been used as evidence of pathological impulsivity in court proceedings.Reference Howard 85 Thus, it can be considered an objective measure of behavioral impulsivity. Based on the CNV results obtained, patients were classified as high or low risk depending on whether their score was 1 standard deviation outside or within a control group’s score, respectively. At 15 years post-testing, criminal records were examined to reveal that 6 of 21 in the high risk group had been convicted of another offense, including manslaughter, burglary, and arson. This compares with only 1 of 23 in the low risk group, convicted of theft. Thus, it appeared that using the CNV during go/no-go was sensitive to differentiating those who may reoffend, and appeared to identify those at risk of committing more serious offenses. The authors assert that the overall predictive accuracy was 63.6% and the relative improvement over chance was 72%.
Biological predictors
No studies examining the effect of biological variables on community reoffending were identified.
Discussion
This systematic review of objective factors relating to outcomes in forensic mental health services is, to our knowledge, the first review of such factors to be conducted.
In terms of demographic factors, the predictors of inpatient violence included previous psychiatric admissions (67% positive finding), with mixed findings for young age (50% found an association with inpatient violence). Demographic factors associated with an increased length of stay included the severity of the index offense (90% positive finding) and having a history of absconding (67% positive finding). Initially, psychosis appeared to be associated with an increased length of stay; however, once studies examining “diagnosis” as a predictor more broadly were considered, this association was weakened, probably due to sample diagnostic homogeneity, as a low number patients included in these studies were not diagnosed with psychosis. Our findings relating to reoffending suggest previous offending, young age at admission or discharge, and personality disorder are relatively robust predictors of recidivism, with the large majority of studies examining each factor indicating a positive association. The majority of studies that examined psychosis found that this had no relationship with future offending, perhaps reflecting the relative efficacy of treatments that are available for psychotic disorders in comparison to personality disorder.
This review may have been limited in its ability to examine demographic predictors of outcome, as it excluded articles relating to risk assessment tools, which focus on this type of predictor. Structured professional judgment tools such as the HCR-20Reference Webster, Douglas, Eaves and Hart 6 include items such as young age, identified by this review to be related to future offending, suggesting that they do hold useful predictive properties. However, many factors identified in this review showed conflicting results. For example, young age was found to be associated, and also not associated, with inpatient violence in an equal number of studies, just as a previous prison sentence was found to increase the length of stay in 2 studies, but found to be unrelated in 2 further studies. This suggests that demographic factors in isolation are not particularly useful to clinicians in assisting decision making, but may perhaps hold more validity when considered in combination (as risk assessment tools advocate).
In addition, demographic factors are static and thus not sensitive to changing risk, which may be picked up by indices of neurological or biological function. A further limitation relating to the demographic results is that combining similar, but perhaps slightly different, demographic factors (eg, “severity of offense” and “a violent or homicide offense”), may have somewhat distorted the true relationship between a given predictor and outcome. Future research should aim to operationalize predictor variable definitions to aid in the understanding of the unique contributions each predictor makes. This criticism also holds in relation to the definitions of outcome. For example, inpatient violence often has broad and differing conceptualizations in research investigations,Reference Harris, Oakley and Picchioni 86 and although the majority of articles included in this review included episodes of both verbal and physical aggression in this outcome category, some excluded verbal threatsReference Hoptman, Yates, Patalinjug, Wack and Convit 27 and some included specific operationalizations such as “throwing food or an object that strikes another person.”Reference Zapf and Roesch 23 Length of stay may also have different implications across countries. For example, in the UK, length of stay is linked to clinical responsiveness. Patients admitted under a hospital order are able to move from hospital to conditions of lesser security once they are deemed to have responded to treatment and reduced their level of risk. However, this may not be the case in other countries, such as the USA, where fixed-length sentences may have been imposed. In this review, one-third of studies examining length of stay were conducted in the USA, with 50% conducted in Europe and 17% in Australasia. To allow greater insights into our findings, information about the location of individual studies has been included in Table S1.
Common themes emerged from the identified neuropsychological and neurophysiological predictors. Impulsivity as assessed by the contingent negative variation event-related potential was associated with future reoffending upon discharge,Reference Howard and Lumsden 83 and SCWT errors (poor cognitive inhibition)Reference Foster, Hillbrand and Silverstein 32 were associated with increased frequency and severity of inpatient violence. Both of these facets could be considered to reflect poor behavioral controls, and thus this may be an area that merits further research in relation to its utility as a marker of violence or reoffending. One study included in this reviewReference Enticott, Ogloff, Bradshaw and Daffern 36 did not support this assertion; however, as previously discussed, it was underpowered, with a very short follow-up period and a low rate of inpatient violence was observed. Poor social cognition emerged from 2 studies as a robust marker of outcome.Reference Foster, Hillbrand and Silverstein 32 , Reference Murphy 38 Misperception of angry voices was found to be associated with inpatient violence, and another study identified poor reading of emotion from the revised eyes task to be the overwhelmingly best predictor of risk and unmet need at follow-up. These results indicate that both cognitive and social-cognitive deficits appear to be associated with outcome, and could be targets for effective treatment.
The strategy of using neuropsychological tests to predict outcome is strengthened by other, nonprospective studies not included in this review. For example, it was shown that scores from the Iowa Gambling TaskReference Bechara, Damasio, Damasio and Anderson 87 could be used effectively to predict whether MDOs had been secluded in the past for either predatory or impulsive violent acts while in secure mental health services.Reference Bass and Nussbaum 88 However, 1 cross-sectional studyReference Fullam and Dolan 89 found no significant association between neuropsychological measures and previous inpatient violence in 82 violent men with schizophrenia (including the National Adult Reading Test,Reference Nelson and Willison 90 the Wechsler Abbreviated Scale of Intelligence,Reference Wechsler 91 Stop Task,Reference Rubia, Russell and Overmeyer 92 and the CANTAB-2 batteryReference Fray and Robbins 93 ), although current and predicted IQ tended to correlate negatively with the number of violent incidents across an individual’s time in the hospital, suggesting that there may be a role for neuropsychological function in the emergence of violent behavior. More prospective studies are required to fully elucidate relationships such as these.
The use of biological markers to assist in clinical decision making also appears to have support from the reviewed studies. Both serum cholesterol and creatine kinase appeared able to predict inpatient violence to a reasonable degree of accuracy. Low serum cholesterol has been linked to higher rates of death from violence or suicide,Reference Golomb, Stattin and Mednick 94 and experimentally lowered cholesterol has been linked to aggressive behavior in animals.Reference Kaplan 95 A putative mechanism of action suggests that low cholesterol reduces the integrity of cell membranes, making serotonin receptors less efficacious, and poor serotonergic transmission has been linked to violent behavior.Reference Kaplan 95 Serum cholesterol as a marker has shown great promise in another prospective study of nonforensic inpatients; total cholesterol had a significant negative relationship with inpatient suicidal and violent behavior, and with 3-month post-discharge violent behavior.Reference Roaldset, Bakken and Bjørkly 96 This is an area for future research and development with strong potential.
A number of other studies, not included in this review due to the samples being referred for “forensic psychiatric evaluation” as opposed to admitted to services, have also investigated biological markers and show some promise. For example, one studyReference Virkkunen, Rissanen, Franssila-Kallunki and Tiihonen 97 found that 27% of variance in reoffending could be explained by low non-oxidative glucose metabolism in a sample of violent offenders referred for evaluation and followed up 8 years later. Another studyReference Stalenheim 98 showed that high levels of the thyroxine hormone triiodothyronine were associated with relapse into offending in another cohort of offenders referred for psychiatric examination. The use of these markers to predict other outcomes such as inpatient violence in individuals specifically detained in forensic mental health services is an area to be explored further.
The use of biomarkers to predict complex behavioral outcomes such as aggression or reoffending requires ethical consideration. Biomarkers in psychiatry have been subject to ethical scrutiny, namely for reasons including over-simplification of multifaceted and complex conditions, and by shifting the focus of “risk” to the individual as opposed to considering the wider societal contributing factors.Reference Singh and Rose 99 These issues are relevant to MDOs, and further work in this area should be mindful of the wider implications of the findings. Certainly at this early stage, putative biomarker predictors should be considered alongside clinical judgment and other predisposing factors such as personality pathology. There are also scientific issues to be resolved before the use of biological markers can be condoned. For example, an acceptable level of sensitivity and specificity would need to be established for any putative marker, and this would need to add incremental validity to any risk assessments that are currently in practice. An idea of the temporal stability would also be required, i.e., over what time frame does this marker suggest a risk? Interactions with medications and the “trait” vs “state” status of any biomarker would be further considerations before widespread use could be advocated.
In conclusion, the findings of this review suggest that using neuropsychological, neurophysiological, and biological markers to inform outcome is a feasible and potentially useful strategy. However, development of such markers is in its infancy, and further research in this field is required to translate these findings to clinical practice. Initial replication of the promising, small-scale studies identified in this review is needed and, if successful, large prospective cohort studies would be essential to establish the merit of such a strategy. Once developed, adding empirical markers such as these to clinical decision-making tools may be a beneficial strategy in the future to improve outcomes for MDOs, who are a group at present experiencing lengthy admissions to psychiatric care and poor outcomes in terms of reoffending. Thus, innovation in this area is essential.
Disclosures
The authors do not have anything to disclose.
Supplementary material
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S1092852915000723
