Lecture 7: Risk assessment and management Flashcards
What impact does risk assessment make to individuals?
Can remove someone’s freedom for months / years on the basis of the outcome
OR
Can discharge people who go to sexually / violently assault members of the general public / kill themselves
What is receiver operating characteristics?
ROC: s a graphical plot that illustrates the diagnostic ability of a classifier system (in this case risk assessment) as its discrimination threshold is varied
It identifies hits, false positives, miss and correct rejections
How can ROC be used in analysis of risk assessment?
Can follow offenders for 2 years after being released and see if the risk assessment was correct
What are the issues of ROC in the analysis of risk assessment?
Almost impossible to measure people labelled as high risk - unethical to release them and see what they can do
What is the AUC?
Area under curve of the ROC
What do different ROC/ AUC scores mean for accuracy of risk assessment?
AUC = 0.5 chance performance
AUC = 1.0 perfect performance
AUC= 0.56 regarded as ‘weak’
AUC = 0.65 regarded as ‘moderate’
AUC = 0.71 regarded as ‘strong’
AUC= 0.90 regarded as best as we could do considering we don’t have perfect outcome measures
Limitations of using ROC as predictors?
Loses the quality of the event e.g how long / how quickly they were violent / the extent of the violence etc
Strengths of using ROC as predictors?
ROC is immune to baseline changes (e.g does not matter if it is rare or common event)
Keeps a continuous nature of assessment scale
What factors predict violence?
Previous violence
Unemployment
Poor relationships
Victims of abuse
Negative attitudes
Poor temper control
Issues of using previous violence as a predictor of future violence?
The majority of people receiving a risk assessment for violent behaviour have already committed a violent crime / act
What is an unstructured clinical judgment?
A professional looks at the available evidence, reads files, interviews and then pronounces on their view of how dangerous the person is
Based on their impression, intuition, gut feeling etc
What is the most widely used risk assessment method?
Unstructured clinical judgement
What are the strengths of unstructured clinical judgement?
- Allows ideographic analysis of offenders’ behaviour and a person and context-specific formulation that informs treatment, prevention, and management
What are the limitations of unstructured clinical judgement?
Allows biases
Overlooks important factors
Not based on empirical evidence of risks
What were psychiatrists previously told to assess in risk assessments?
Bedwetting
Cruelty to animals
Arson
Baxstrom study
Johnnie Baxstrom appealed at being detained in hospital for criminally insane after his sentence finished - the appeal was upheld and New York State decided that 966 other dangerous patients should also be released
Very few committed any offenses after being released - only 20 were later arrested for any violent crime (2% - only 20)
What do the results from the Baxstrom study suggest?
The decisions made by clinicians were hopeless
They were no more dangerous than other patients who were NOT allowed to be released
What do Gunn and Taylor (1993) say about the clinical judgment of future risk?
Whatever the theoretical position, in practice psychiatrists and others are bad at predicting the future violence of inmates released from institutions
Odeh et al (2006) study design
Gave professionals (inc nurses, psychologists, and psychiatrists) info about a patient and asked them to make various risk judgments and the probability and severity of future violence.
They were asked why they made the decisions
Odeh et al (2006) results
They found that the cues were not related to violence and therefore the inter-rater reliability was very poor
One said high risk, others low risk for the same person
Why are clinical judgments so bad?
- They are blind to their outcomes for the majority of cases - a smallish number of those released are violent and they may not hear about it - gives the impression they were right
- Tendency to weigh bizarre or unusual factors (e.g command hallucinations) heavily and neglect criminogenic factors
- Too many variables - research in cognitive psychology (Kahnemann and Tversky) reveals that we can only keep track of a small no. of variables when making decisions
- Tend to make judgments quickly and seek support for this
What is an actuarial assessment?
Factors thought to be predictive of risk are put together using a pre-ordained method and are normally based on a ‘construction sample’
What are the strengths of actuarial assessments?
Avoids individual bias
Does not need clinical skills to formulate
Fast
What are the weaknesses of actuarial assessments?
Often lacks ideographic information
Does not easily suggest risk management
A number is not enough
limited to questions given
person must fit with the sample
static measure - now show change in dangerousness
Describe the Quinsey et al (1998; 2006) study
Based on 618 men from Canadian maximum secure psychiatric unit - all had committed at least one serious antisocial act were released into:
- the community
- minimum security
- half way house
and followed for 10 years to see if they committed a violent act
How did Quinsey et al (1998; 2006) create the VRAG?
They measured a range of potential predictors e.g age, number of previous offenses etc)
Dropped any items that did not predict violence
If 2 items were very correlated (e.g the number of previous charges, the number of previous violent charges) the one with the highest correlation was used
Least-squared regression was used to see which of the remaining variables added independently to risk prediction mode
These 12 items were added to the VRAG
What are the 12 VRAG items?
Lived with biological parents to age 16 ( PROTECTIVE FACTOR -VE score)
Elementary school maladjustment
History of an alcohol problem
Marital status
Total Cromier-Lang scores for criminal charges
Failure on conditional release
Age at index offence
Victim injury
Sex of victim
Personality disorder
Schizophrenia ( PROTECTIVE FACTOR -VE score)
Psychopathy checklist score
What are the VRAG categories and risk probabilities?
Scores range from -26 to + 38
Scores used to bin people into 9 categories
Category 1: None were violent within 7 years
Category 9: 100% were violent within 7 years
What were the results from the VRAG construction sample?
Reliability is very high =0.90
SEM of 4 points
VRAG predicted violent incidences - AUC = 0.76
What are the issues with the results from the VRAG construction sample?
We should expect some slippage in v similar sample as OG can take advantage of random correlations
Different samples might have different risk factors (e.g learning disabled, females, UK)
Explain Snowden et al (2007) VRAG study
Followed 421 male patients discharged from medium secure units in the UK (already released 2 years ago)
VRAG scored by case-note review at the time of discharge
What were Snowden et als (2007) results?
Up to a year the VRAG had great predictive validity (AUC = 0.86) - better than OGRS
BUT longer follow up (after 2-3 years) AUC = 0.76 (similar to OGRS predictive accuracy)
The base rate was very different to the original Canadian sample; good at rank ordering but not so good at absolute levels
Conclusions from Snowden et al (2007)
VRAG has good validity for predicting crime, especially violent crime (AUC = 0.86)
The absolute rates of violence in this group are lower than the Canadian sample
Why did violence rates using VRAG differ from England and Canadian sample?
- use of formal reconviction may miss many violent incidents
- may be more supervision / management
- was not able to correct for any further time spend incarcerated (perhaps for a minor offence)
- we are less violent than Canadians
- less competent detection / prosecution of incidents
Why do clinicians argue against comparisons of clinical vs actuarial?
Clinicians tend to focus on short-term outcomes
Clinicians have different info available
Now clinicians know about the variables that predict violence they do alot better
Clinicians concentrate on the really dangerous - not some single act that may occur years ahead (e.g fight in a pub)
Some clinicians are bad
Explain Harris et al (2002) clinical vs actuarial study
Clinicians who knew the patient well were convened
They had access to all the information
They gave a rating (1 to 7) of the patients security needs
Compared to rating using VRAG
AUC for clinicians = 0.59 (small)
AUC for VRAG = 0.80 (large)
Only VRAG significantly predicted homicides - clinicians were at chance level
What is OGRS?
Offender Group Reconviction Scale
used by UK prison and parole service for all offenders - predicts general offending
What is the RM2000?
Used by UK prison service to predict sexual offenses
What is the COVR?
Classification of violence risk
Developed from MacArthur study - uses a classification tree and easy to get variables
Why are actuarials rarely used?
Many clinicians feel:
- They de-individuate the patient
- Feel that their status is being eroded - if a formula/computer can make the decision then what is the clinician for?
- Risk assessment is NOT their job they are there to treat the person’s illness
What are some issues with actuarials?
- Person has to fit with sample - broken leg problem
- Use static measures - not good at showing any change in dangerousness - but does not have to be so
- tend to be based on common violence (e.g assault) - this does not have to be so
- getting a number is not enough - we really also need to know WHY, WHEN, AND WHAT we can do about it
What is a structured clinical assessment?
It attempts to combine the worlds of clinical and actuarial measures
Clinicians’ scores set of preset items on whether present or not
Makes decisions about risk in a clinical manner
What are some strengths of structured clinical assessment?
Avoids individual bias (to some extent)
Contains ideographic information and is excellent at making evidence-based formulations and management plans
What is a limitation of structured professional judgment (SPJ) assessments?
Time consuming
Needs clinical skills to interpret
What is the ‘best’ structured clinical assessments?
HCR20
What are the 3 sections of the HCR20?
Historical (past)
Clinical (present)
Risk management (future)
What are the items in historical part of HCR-20?
Previous violence
Young age at first offence
Relationship instability
Employment problems
Substance abuse problems
Major mental illness
Psychopathy
Early maladjustment
Personality disorder
Prior supervision failure
What are the items in the clinical part of HCR-20?
Lack of insight
Negative attitudes
Active symptoms of major mental illness
Impulsivity
Unresponsive to treatment
What are the items in the risk management part of HCR-20?
Plans lack feasibility
Exposure to destabilisers
Lack of personal support
Non-compliance with remediation attempts
Stress
Gray et al. (2003) study design (HCR-20, OCL-R and BPRS (symptoms) and BHS)
Prospective study in forensic mental health setting.
Assessed at admission, followed for 3 months and all incidences of violence were noted.
Assessed using HCR-20, OCL-R and BPRS (symptoms) and BHS.
Gray et al. (2003) study results
Results show that the HCR-20 has good ability to predict violence whilst in the unit (HCR-20 = 0.81)
(PCL-R = 0.70
BRPS = 0.84
BHS = 0.53)
Gray et al. (2008) follow up study
Traced UK male mentally disordered offenders discharged from MSU’s for at least 2 years (traced through home office)
Assessments were made at time of release (prospective study)
Gray et al. (2008) study results (1/2 year later)
Total: 0.76
Historical: 0.77
Clinical 0.61
Risk: 0.69
Gray et al. (2008) study results (1 year on)
Total: 0.71
Historical: 0.71
Clinical: 0.54
Risk: 0.65
Why did the clinical scale of the HCR-20 NOT work that well in Gray et al. (2008) study?
They were not able to do a clinical interview
What studies have assessed HCR20 in female patients?
Nichols et al. (2004): preliminary findings that HCR20 has some predictive validity in women
DeVogel and De Ruiter (2006): men AUCs ranged from 0.75-0.88, women 0.52-063. Clinical judgment was a lot better than Gray et al (0.80)
What study has assessed HCR20 in learning disabilities?
Gray et al. (2007): psychiatric inpatient care sample
HCR20 very good (AUC > 0.80) - better than other offenders (UK sample)
What study has assessed HCR20 for ethnic minorities?
Snowden et al. (2010)
Psychiatric inpatient
Both VRAG and HCR20 had similiar predictive ability for black patients as white
What study assessed HCR20 in relation to different diagnoses?
Gray et al. (2011): piC sample
- Found that diagnosis is an indicator of future violence (those with PD = most dangerous)
- HCR20 good predictor at those with scz but worse in PD
Why is the HCR20 not as good at predicting violence in those with PD?
Could be statistical issues
Harder to follow
More likely to die due to violence / suicide (therefore statistics will say they havent committed crime)
May just be less easy to predict their behaviour
What are some limitations of the HCR20?
- need more evidence of its utility to specific populations (e.g female, ethnic groups, PD, LD, age)
- need more evidence on specific forms of violence - DV, children
- Need to see if treatment produces changes in HCR20
Homicide is rare and therefore few studies have assessed HCR20 as predictor
When was the HCR20 updated?
2014
What changes were made to the HCR20?
7 STEPS:
1. gather info
2. presence of item
3. relevance of item
4. formulation of risk
5. scenario planning
6. case management plan
7. develop final options
Recent backlash to the HCR20 + risk assessment in general?
Clinicians have argued against risk assessment: say we need to treat people better BUT BOB would still say we need risk assessment
What assessments have been produced for sexual violence, suicide, domestic violence and short term risks?
Sexual violence = SVR20 and RSVP
RoSP = suicide
SARA = domestic violence
START = many short term risks
