Risk Assessment and Management Flashcards
Receiver Operating Characteristics
Used to express how many good predictions we are making
Use extensive file history to conclude on Yes or No predictions - yes = violent, no = not violent
Follow up later to see what happened, Yes/no
Difficult to know how well we predicted as our risk assessments indicating violence prediction often try to stop from happening - don’t actually know what would have happened
Expressed in a ROC graph - plot hits against false positives
Hit
Predicted to be violent = Yes
Was violent = yes
False positive
Predicted to be violent = yes
Was violent = no
Correct rejection
Predicted to be violent = no
Was violent = no
Misses
Predicted to be violent = no
Was violent = yes
ROC curve
How much the curve bulges from the centre line is how good we are at making hits - called the AUC
AUC
0.50- chance performance – same number as hits and false positives
0.56 - weak
0.65 - moderate
0.71 - strong effect
0.90 - best possible
Pros of ROC
Immune to baseline changes
Can compare with very different base rates
Cons of ROC
Lose quality of an event - e.g. severity, type of violence
Predictive factors for violence
Previous violence
Unemployment
Poor relationships
Victim of abuse
Negative attitudes
Poor temper control
Types of risk assessment
Clinical judgement
Actuarial measured
Structured professional judgement
Unstructured clinical judgement
Based on impression, intuition and experience
Administered by clinicians
Most widely used
Allows for idiographic analysis - intensive study of one person
Allows clinician bias
Unstructured clinical judgement - reliability
Very poor interrater reliability when assessing risk
Poor predictivity
Why are they unreliable?
Clinicians are blind to outcomes - if they actually become violent or not
The tendency to weigh bizare or unusual factors more heavily (e.g. grand delusions) rather than the more actually correlative crimiogenic factors
There are too many variables for a clinician to substantially keep track of
We tend to make judgements quickly and later seek support for them
Actuarial Assessment
A preordained way of coming to a conclusion
Uses factors thought to be predictive of risk using a pre-ordained method - decide if people show these risks or not to inform of what to do next – can add up to a score or lead to a path
Avoids individual bias only to an extent - still subjective if people fit into risk or not
Quick
Lack in idiographic information - possibly reductionist
VRAG - Quinsey et al.
Example of an actuarial assessment
Developed from men who left a Canadian psychiatric unit - released to community, minimum secure units or halfway houses
Measured a range of potential predictors
Highly correlated predictors were used to make 12 items of risk
Scoring 12 items add up to put people into 1 of 9 risk categories
VRAG predictive validity / reliability - Snowden et al.
Pseudo-prospective study - only looked at case notes to review VRAP score (when blind to outcomes) and later checked accuracy by using home office records
High predictive accuracy over 1 year, lasted for around 3 years but went down slightly
Rates lower than Canadian sample but still sig.
Clinician individual differences
Different levels of competence and experience
Tend to focus on short-term outcomes, rather than long-term
Tend to focus on more obviously dangerous patients - can miss others less apparent (part of weighing bizarre factors more heavily)
Harris - AUC of clinician vs. VRAG
Clinicians gave risk rating for patients they knew well and compared with VRAG - prediction of homicides
Clinician AUC - .59
VRAG AUC - .80
– VRAG much better for predicting homicides
Why have actuarials not been widely adopted?
May de-individualise the patient
Clinician belief that their status is at-risk
Clinician belief that risk assessment is not a priority of their job but rather treating
Broken leg problem - actuarials do not account for new factors (such as a broken leg) that would greatly reduce ability/chances of violence
Atheoretical - only provides a number of risk but doesn’t actually give us any idiographic information or why the risk is that
Why have actuarials not been widely adopted?
May de-individualise the patient
Clinician belief that their status is at-risk
Clinician belief that risk assessment is not a priority of their job but rather treating
Broken leg problem - actuarials do not account for new factors (such as a broken leg) that would greatly reduce ability/chances of violence
Atheoretical - only provides a number of risk but doesn’t actually give us any idiographic information or why the risk is that
Structured Clinical Assessment
Combines clinical judgement and actuarial measures
A clinician scores a set of pre-set items and makes a decision of risk based off that score
Avoids bias to an extent - still slightly interpretative
Contains idiographic information
But is time consuming and need clinical skills to interpret
HCR-20
10 historical items (derived from VRAG)
5 present clinical items
5 future risk management items
HCR-20 efficacy in predicting violence + self-harm
Followed for 3 months
Noted all incidences of violence through nursing and incident reports/forms
– very good as not relying on inconsistent crime stats that miss a lot of unreported crime
Classified violence by type - verbal, physical, property
HCR was highly predictive of violence on all 3 types of violence
– AUC of .79 - .83
Prospective studies
Measure from present to future - highly predictive and can confirm/deny our predictions of risk
Can measure confounds and how they change/influence over time
Violence risk is long-term, good as it reflects this
Prospective studies
Measure from present to future - highly predictive and can confirm/deny our predictions of risk
Can measure confounds and how they change/influence over time
Violence risk is long-term, good as it reflects this
HCR-20 and type of MI
Most effective predictor for learning disabilities and schizophrenia
Least effective predictor for personality disorders - perhaps not very strong against manipulation/malingering?
HCR-20 limitations
Hasn’t been tested thoroughly in diverse populations
Doesn’t specify specific forms of violence, e.g. sexual, domestic