Notions of psychopathic personality and personality disorder have a lengthy history within criminal justice and mental health settings. Issues of how to assess and intervene with this group of individuals also continue to present significant challenges for both services. Those categorised as psychopathic or personality disordered may behave in ways that are seriously damaging both to themselves and to others. There has also been frequent criticism and, at times, castigation of services for failing to engage with these groups and protect the public effectively.1

Such pressures have provided impetus to the search for reliable and valid assessments of psychopathy with prognostic value. The Psychopathy Checklists (PCL) is a family of structured assessments developed in the USA by Robert Hare2 ,3 (see box 1). They are based loosely on the criteria for psychopathy set out by Cleckley,4 although it should be noted that significant differences exist. The most commonly used version in practice is the full length Psychopathy Checklist in its revised form (PCL-R). This is a structured checklist that yields an overall score, which is generally divided into two factors. Factor 1 (F1) relates to the selfish, callous and remorseless use of others, factor 2 (F2) relates to a chronically unstable and antisocial lifestyle. Each of the factors can be further divided into facets, with F1 comprising interpersonal and affective facets and F2 comprising lifestyle and antisocial facets.5 ,6

The PCL has seen dramatic growth in its use, particularly in the criminal justice context.7 ,8 Growth was from a very low baseline, largely on the basis of reported promising results from early research.9 The development of use was underpinned by three main claims about the PCL:

1. that it provided a reliable and valid means of identifying those who are unsuitable for current interventions intended to reduce criminal re-offending

2. that it provided an appropriate basis for the development of new interventions for psychopathy

3. that it provided the best means of assessing future risk of violent recidivism.3

These claims though have not been supported by the emerging evidence base.5 ,8 ,10

SELECTION FOR TREATMENT AND THE DEVELOPMENT OF SPECIFIC TREATMENTS

Issues of selection for treatment and the development of specific treatments are closely linked. The notion that those scoring highly on the PCL may suffer counter therapeutic effects from existing treatments creates a need to try to develop approaches that are effective for this group. In criminal justice settings this has become largely systematised, with prisons in England and Wales, for example, routinely excluding those with PCL scores above a specific cut-off point from mainstream interventions. Conversely, those below a specified cut-off may be excluded from access to specialist personality disorder services.11 Specific interventions, such as “Chromis” in the prison service, have been designed exclusively for those scoring above a specific level on the PCL.12 ,13

The evidence for these developments has increasingly been questioned on clinical, economic and ethical grounds.8 ,14 The view that those assessed as above a certain level on the PCL are not amenable to treatment and respond negatively to conventional interventions has become widespread in routine practice. It is a view that can be traced back primarily to a study in Canada15 which analysed outcomes from the Social Therapy Unit (STU) at Penetanguishene. Given its impact it is perhaps surprising that it is only recently that it has been subject to rigorous review.16 The Penetanguishene study clearly had a number of strengths, involving a 10 year follow-up of participants who had completed 2 years of therapy. Those who underwent treatment were also compared with an untreated control group. The researchers subdivided their sample on the basis of PCL-R scores into “psychopathic” groups (scoring ⩾25) and “non-psychopathic” groups (scoring below this threshold).

It was reported that the general recidivism rates for those scoring less than 25 was significantly lower for those who had participated in treatment (44% vs 58%; p<0.05). For those scoring 25 and above, the outcomes were poor, with little difference between treated and untreated groups (87% vs 90%; NS). The violent recidivism rate for those scoring less than 25 was significantly lower for the treatment group (22% vs 45%; p<0.05) but for those scoring 25 and above the opposite trend was seen with treated patients offending at higher rates than untreated (77% vs 55%; p<0.05).

A number of significant and poorly recognised weaknesses were present in this study. It involved a retrospective file review to determine PCL-R scores with a matching, rather than a random, allocation to treatment and control groups. This created the possibility of unknown confounding factors being introduced into the study.16 The nature of the therapy undertaken at the STU is also worthy of comment and would generally be seen as out of keeping with current practice. It involved the use of drug therapies including methedrine, LSD, scopolamine and alcohol. It also included the use of marathon naked encounter group sessions, along with a high degree of compulsion to engage in therapy.16

Subsequent studies though have reported similarly adverse outcomes from treatment of those scoring highly on the PCL.17 In this study, prisoners with a high score on F1 of the PCL-R (⩾9) were reported to have higher reconviction rates if they attended group based interventions than if they did not (85.7% vs 58.8%; p<0.01). This group was also reported to have higher reconviction rates if they attended educational courses (82% vs 61%, significance not reported). From the reporting of this study though, it is unclear whether the treatment group also included those who failed to complete treatment. This is of particular significance given the replicated finding that those who drop out of psychosocial interventions often fare significantly worse that untreated groups.8 ,18 If those with higher PCL scores drop out of treatment at higher rates, then it may be this failure to retain them in treatment, rather than issues of response to treatment, that is central here. Notably there is evidence to suggest that those with high PCL scores (PCL-R ⩾30) are indeed more likely to drop out of treatment (32% vs 4%, respectively).19

Another influential study compared outcomes for sexual offenders with high PCL scores showing “good” and “poor” responses to treatment.20 A survival analysis was conducted for serious recidivism and sexual recidivism, with participants in the study divided into those with overall PCL-R scores of 25 and above and those with scores below 25. These groups were further subdivided into those rated as showing “poor” and “good” response to treatment. The results suggested that the high PCL-R groups performed more poorly overall but that the failure rate was not significantly higher for the high PCL score group when they had been rated as responding well to treatment. While not perfect, this study did involve ratings made blind to PCL-R scores and reported low attrition rates.16

The practice of denying access to mainstream interventions on the basis of PCL scores, and the associated widespread belief among practitioners that the evidence base clearly demonstrates they are not treatable using such interventions, has very significant impacts. The conclusions of this detailed review are therefore perhaps worth stressing16:

“This review has demonstrated that we do not have the evidence to conclude that high scoring psychopaths have a negative response to treatment.” (pp174).

The fact that those who score highly on PCL assessments fare worse that those scoring lower is of itself unsurprising. It does not itself provide support for the conclusion that such individuals are not amenable to, or are made worse by, existing treatments. High scoring individuals are, self evidently, likely to have more severe difficulties and greater needs than those with lower scores. They have also been shown, as a group, to re-offend at higher rates. The appropriate comparison in terms of addressing questions of response to treatment though would be a comparison between randomly allocated treated and untreated psychopathic groups. To date, such studies have not been undertaken and the existing evidence base remains too thin to answer what remains an open question.16

Such findings also draw into question the need to develop separate provision in terms of treatment and indeed whether scores on PCL assessments are an appropriate, or clinically useful, basis for doing so.

ASSESSING RISK

The use of the PCL in risk assessment has also been subject to recent scrutiny with a number of empirical studies being undertaken. These suggest that the construct is a relatively strong predictor of general and violent recidivism.21 Those scoring above specified cut-off levels have been reported to be between five and 14 times more likely to be arrested for a violent crime, with an above chance level of subsequent violence.

Recent empirical studies of the PCL based on mental health services in a number of cities in the USA have reported an odds ratio (OR) for subsequent violence of 2.40, a significantly better than chance level of performance. In the same study however, other, relatively easily obtained, single factors, also yielded better than chance levels of performance. For example, a history of father’s drug use yielding an OR of 2.18.10

The MacArthur Risk Assessment Study considered the clinical utility of the PCL in relation to mental health samples and prediction of subsequent violence. Using receiver operating characteristics (ROC) area under the curve (AUC) analysis, this study reported that the PCL predicted violence at significantly better than chance levels (ROC AUC 0.73), with a PCL-Screening Version (PCL-SV) score of 8, achieving the best balance between sensitivity (AUC 0.72) and specificity (AUC 0.65).10 ,22

The choice of factor solution for the PCL is contentious and clinically complex. Yet much of current practice in terms of risk assessment using the PCL has been based simply on the global score, combined with variable thresholds. In fact, when the commonly used two factor solution is employed, F1, relating to the hypothesised “personality disorder” aspects of the assessment, has been reported to add little to the predictive power in terms of general and violence reconviction. Predictive power here appears to derive mainly from F2, composed largely of items measuring antisocial behaviour, including previous offending. Such predictive power appears to derive largely from previous behaviour, with its well established association with later risk.23 Indeed scores on this factor have been reported to be similarly predictive to the score on the full assessment (AUC 0.74 compared with AUC 0.73), suggesting a high degree of redundancy. It has been convincingly argued that there is little impact in terms of such prediction from the hypothesised emotional and personality characteristics.24^–27

Some have suggested that as a risk assessment method the PCL simply represents a pooling of risk factors, concerned with previous antisocial behaviours, rather than an underlying personality characteristic.28 There is some evidence to support this view although empirical studies have suggested that the performance of the PCL in predicting violence cannot be fully accounted for in this manner.22

It has been argued that considerable caution is needed when applying the PCL as a form of risk assessment. The meaning of scores at both above threshold and sub-threshold levels will be heavily dependent on an adequate understanding of test specificity, sensitivity and likely population base rates. In turn, the interpretation of scores will also be dependent on the factor scores and the factorial solution or solutions used.24 ,25 ,29^–31

Given the complexity and expense of administering the PCL as a means of risk assessment, it clearly needs to perform significantly better than other measures if its use is to be justified. Yet the results from a large scale prisoner cohort study yielded disappointing results. Based on a cohort of 1396 adult male offenders, the PCL-R performed poorly with few items on the scale demonstrating independent predictive power. Those items reported to be predictive were mainly features of the antisocial and impulsive features of psychopathy. The best performing risk assessment in this study was in fact the Offender Group Reconviction Scale (OGRS-2).32

IMPLICATIONS FOR PRACTICE

Use of the PCL has seen very marked growth over the past decade, with assessments increasingly used to determine access to treatment, the development of interventions and the assessment and management of risk. This has taken place in a context of great enthusiasm based on reports of initially promising results. Much of the discussion about the PCL has subsequently been focused on matters such as levels of qualification and provision of training required for use, rather than more substantive issues such as the clinical utility and cost efficacy of the assessment.

Emerging evidence increasingly casts doubt on both the utility and cost effectiveness of the PCL. Its use as a basis for exclusion from mainstream and specialist services has been implemented in a number of settings. Yet such practice lacks a well founded and convincing evidence base. A review of work with personality disordered offenders33 reported that individuals tended to benefit, albeit to a limited degree, from a broad range of interventions, ranging from psychodynamic approaches to drug therapies. In reality the issue of whether those with such personality disorder benefit from current mainstream treatments remains an open question: it is an area where there is simply a lack of sufficient good quality evidence to make any clear determination and there is a pressing need for good quality randomised trials.

The value of basing services around PCL scores is also unclear. There are at least two problems central to such an approach. One is that it seems likely to exclude large numbers of personality disordered individuals who present serious risks to themselves and others but who fall below the arbitrary PCL threshold used. Secondly, it is not obvious that a composite measure like the PCL provides a particularly useful clinical basis for intervention work. The use of resources to undertake PCL assessments as a means to exclude individuals from services therefore appears misplaced. Here, as well, resources might be more usefully employed to provide independent research addressing questions of which interventions are effective, with whom and when.

The use of the PCL as a basis for developing specific treatments for personality disorders is also problematic. In his extensive work on personality disorder, Livesley33 makes the compelling point that the study of this area needs to be grounded in the study of normal personality development. Yet the PCL is not convincingly grounded in current mainstream research.33 ,34 There are more promising and clinically useful means of assessing personality and personality disorder and it would seem appropriate that these are adopted.24 ,25 ,35^–37

The early promise of the PCL as a means of risk assessment has not been seen in subsequent research. It has though become central to a number of structured risk assessments, as well as being increasingly used as a means of risk assessment itself. To justify its expense it would need to significantly out-perform other less resource intensive methods. Recent meta-analytic studies have suggested the opposite, with the PCL performing more poorly than assessments such as the OGRS-2.32 The application of more sophisticated approaches to risk assessment has the potential to widen this gap even further drawing into question the utility of the PCL in risk assessment.10 ,32

Drawing largely on an initial burst of enthusiasm and reports of highly promising results, the PCL has seen dramatic growth in its use, largely in criminal justice settings but also within mental health services. Such use has run well ahead of the evidence base and the future value of the PCL to mental health practice is unclear.