A brief report from a comparison of the HCR-20 and the HCR-V3
concerning internal consistency and clinical utility.
Gunnar Eidhammer, Lars Erik Selmer and Stål Bjørkly
The SAFE pilot project group, Centre for Research and Education in Forensic Psychiatry, Oslo
University Hospital, Norway. (www.forensic-psychiatry.no)
Introduction
The HCR-20 has been in clinical use in Norway since the first version came in 1995. However, the big
breakthrough came with the Norwegian translation of the second version (V2) in 2002. Based on
many years of research and feedback from clinicians, the researchers of HCR 20, have developed a
third version of the HCR 20, the HCR-V3. We report from our experience with This report describes
findings from: (i) a small-scale intraclass correlation test of V2 and V3, and (ii) a comparison of the
clinical utility of the two versions.
(i) Intraclass correlation study
Participants and setting
The ratings were carried out in a forensic medium security unit in Norway. Two psychiatric nurses,
compared the V2 and the V3 by assessing 20 forensic mental health patients. The raters have
extensive experience in forensic mental health care; have further education in violence risk
assessment and management and have used the HCR-20 for about 15 years each.
characteristics are presented in Table 1
Table 1
Patient characteristics (n=20)
Diagnosis
Most serious violence
Paranoid Schizophrenia
Homicide
Paranoid Schizophrenia
Homicide
Paranoid Schizophrenia
Homicide
Paranoid Schizophrenia
Severe violence
Paranoid Schizophrenia
Severe violence
Paranoid PD
Double homicide
APD
Violence and threats
Psychosis caused by substance abuse
Arson. Homicidal threats
Patient
Schizophrenia
Severe violence
Schizophrenia
Violence and threats
Substance abuse caused psychosis
Homicidal threats
Paranoid schizophrenia
Violence and homicidal threats
APD/psychosis
Violence and threats
Paranoid PD
Violence
Substance abuse caused psychosis
Violence and homicidal threats
Paranoid Schizophrenia
Violence
Schizophrenia
Knife stabbing
Schizophrenia
Rape
Paranoid Schizophrenia
Arson,
homicidal
threats
towards
police officers
Schizophrenia
Violence in institution
Procedure
The raters assessed half of the patients each. First they made a complete assessment of the patients
with the V2. After that the same procedure was followed with V3 for the same patients. Assessment
data was gathered from multiple sources: Patient files, observations and consulting colleagues. All
information was obtained and every score was rated independently. The items in HCR–20 are scored
0, 1, 2, and nk (not known). In the HCR-V3 the coding is y (yes), p (possibly), n (no), o (omit).
To obtain data for statistical analysis a common score scale of 0, 1, 2 was chosen to transform V3
ratings into scores. This procedure was chosen after consulting with one of the authors of the HCR
group (K.D). The data was analyzed by SPSS Version 16
Results
In the analysis we used a two-way mixed effects model where people effects are random and
measures effects are fixed. All values are average measures. The principal aim was to compare V2
and V3 scores
Table 2
Intraclass correlation coefficients for comparison of V2 and V3 in 20 forensic patients (Presence
scores):
Cronbach's alpha (95%CI)
H-items
.854 (.726 - .940)
C-items
.586 (.256 - .809)
R-items
.812 (.662 - .913)
All items of HCR-20 and V3
.842 (.708 - .935)
C-items: p < .001; Other items: p < .000
There are no Relevance scores in V2. Still, we analyzed the internal consistency between V2 scores
(Presence) and V3 Relevance scores. The Cronbach's alpha value (.839) for the total sum score was
almost identical to the one found for the Presence scores (.842).
In sum, we found moderate (C-items) to good (H- and R-items and aggregate scores) estimates of
internal consistency between the two versions of the HCR. This finding indicates that the two
versions reflect common underlying dimensions, and still there appears to be differences between
V2 and V3 ratings for the same patients. The fact that the scores on the C-items yielded lower
internal consistency when comparing the two versions may indicate that the most substantial
difference in the new version pertains to clinical items. We tested the differences of sum scores for
the H, C and R items. After the finding that the sum scores for the H, C and R items were normally
distributed for both V2 and V3 (Kolmogorov-Smirnov test), we conducted a paired sample t-test to
compare V2 and V3 scores. There were significant differences for H-items (t=-2.797, df=19, p<.012)
and C-items (t=-4.040, df=19, p<.001), but not for R-items (t=.218, p<.830).
Since V2 and V3 are indexes and not scales a very high correlation among the items and the factor
structure as a whole would have pointed to deficiencies in the index, and that rather than taping a
broad set of indicators the actual items may be too narrow. Our results indicate that this is not the
case with the HCR V2 and V3.
(ii) Clinical utility
Overall, compared to V2, the V3 contributes to more systematic and detailed violence risk
assessment, with enhanced opportunity to conduct accurate, individual violence risk assessment. The
introduction of the new risk assessment category ("Relevance") to emphasize individual risk factors
of special importance is an asset of the V3. It informs the development of a tailor-made and
recognizable risk management plan to the best for the patient and his/her social and professional
network.
The manual
In our opinion the V3 manual is significantly improved compared to the current HCR version. Each
risk factor is more elaborated on, detailed and explained more thoroughly. "Indicators" and "coding
notes" provides a good guide for the rater.
The 7 steps (gathering information, the presence of risk factors, relevance of the risk factors, risk
formulation, "scenario planning", risk management strategies and summary), provides a very robust
platform and guide for the violence risk work.
Coding
In our point of view the change from numbers to y, p, n as rating decisions makes the V3 more
clinically relevant and dynamic. This mitigates the risk of importing an actuarial approach into the use
of the V3. Violence risk assessment rarely generates a clear-cut, definite answer. We also think that
the new structure in V2reduces the risk of empirically unfounded conclusions substantially.
Presence / relevance
Since the relevance part is new, we needed some time to comprehend it fully. After some exercise,
however, the relevance part indeed turned out to be a major strength in the V3. It helps the raters to
comprehend the patient's background and history more dynamically in the sense that the presence
of a historical risk factor do not necessarily or automatically mean that it is a valid risk factors here
and now on an individual level. The Relevance part contributes to further individualize the clinical
assessment and most importantly gives the opportunity to make risk management plans to mitigate
idiosyncratic risk factors and triggers.
Risk formulations
It is very positive that the authors introduce risk formulations as a separate part of the assessment
procedure. It is our experience that this approach functions very well together with the Early
Recognition Method (ERM) as part of the work with the R scale.
Concluding remarks
We have tested the internal consistency and the clinical utility of the HCR-20 and HCR-V3. According
to our long experience with V2 and our brief use of V3, the latter is a major step forward to improve
risk assessment of violence.