Correlates of accuracy in the assessment of psychiatric inpatients' risk of violence

Abstract

OBJECTIVE: The authors evaluated characteristics of patients whom clinicians accurately assessed as being at high or low risk for violence and patients for whom clinicians overestimated or underestimated the risk. METHOD: At admission, physicians estimated the probability that each of 226 psychiatric inpatients would physically attack someone during the first week of hospitalization. Nurses rated assaultive behavior in the hospital with the Overt Aggression Scale. Acute symptoms were rated with the Brief Psychiatric Rating Scale. RESULTS: For the group as a whole, assessed levels of risk were substantially related to later physical aggression (sensitivity = 67%, specificity = 69%). Multinomial logit analysis showed that patients with psychotic disorders such as schizophrenia, organic psychotic conditions, and mania were more likely to be accurately assessed by clinicians as being at high risk (true positives) than to be true negatives or false positives. A recent history of violence was associated with higher estimated risk but did not distinguish true positives from false positives. An admission mental status characterized by low levels of hostility, uncooperativeness, and suspiciousness and high levels of depression, guilt, and anxiety differentiated true negative patients from others, but symptom profiles did not differ among true positives, false positives, and false negatives. Clinical judgments emphasizing gender and race/ethnicity were associated with predictive errors: nonwhite and male patients tended to be false positives. CONCLUSIONS: While clinicians can accurately classify the potential for violence in the majority of patients at admission, systematic errors characterize inaccurate assessments of the risk. Awareness of these patterns may help improve assessment of the risk of violence in clinical practice.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 158 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.