TABLE 4. Selected Studies Demonstrating Validity of Clinical Risk Factors in Risk Assessment
| Study |
Major findings |
|
| Lidz, Mulvey, and Gardner (1993) |
Clinical predictions significantly above chance for males (63% sensitive, 60% specificity), but not for females; clinical prediction better than chance even when clinicians not predicting on basis of history, age, or gender |
| McNeil and Binder (1994) |
Five-item scale including clinical, historical, and demographic factors (Violence Screening Checklist or VSC) resulted in 57% sensitivity, 70% specificity |
| Mossman (1994) |
In meta-analysis of 58 data sets, 47 demonstrated better than chance validity of clinical prediction of violence, with combined AUC = 0.73 |
| Mulvey and Lidz (1998) |
Clinicians generally right in predicting seriousness and location of violence, but overestimate role of medication noncompliance and SA; clinicians tend to focus on conditions they can address with treatment |
| Arango et al. (1999) |
Inpatient violence related to clinical variables, not sociodemographic variables or psychiatric history; model based on uncooperativeness, poor judgment, poor insight into psychotic symptoms, and violence in week prior to admission achieved 80% positive predictive power (PPP) |
| Hoptman et al. (1999) |
Psychiatrists' prediction of assault in maximum security forensic facility achieved 54% sensitivity, 79% specificity, and 71% correct classification |
| Skeem, Mulvey, and Lidz (2000) |
Clinicians moderately accurate in predicting violent behavior (55% PPP) based on alcohol use, but do not discriminate well between drinkers who are/are not violent |
| McNeil et al. (2003) |
VSC and clinical scale of HCR-20 significantly associated with inpatient assault |
| Gray et al. (2003) |
Brief Psychiatric Rating Scale best predictor of physical aggression over 3 months among patients admitted to inpatient service, followed by "HCR-15" (i.e., no "R") |
|
