Bias Detection in Emergency Psychiatry: Linking Negative Language to Diagnostic Disparities

Area of Science:

• Psychiatry
• Health Services Research
• Computational Linguistics

Background:

• Emergency departments (EDs) present high-stress environments increasing clinician bias exposure risk.
• Black patients disproportionately receive schizophrenia diagnoses in the ED, a condition linked to stigma.
• Understanding clinician bias in the ED is crucial for equitable psychiatric care and decision-making.

Purpose of the Study:

• To examine the association between clinician bias exposure and psychiatric diagnoses.
• To quantify clinician bias using negative sentence ratios (NSR) in psychiatric notes.
• To investigate the impact of clinician bias on schizophrenia diagnosis, particularly for racial/ethnic minorities.

Main Methods:

• Utilized logistic regression on a dataset of 29,005 patients with anxiety, bipolar, depression, trauma, and schizophrenia diagnoses.
• Quantified clinician bias exposure via the ratio of negative to total sentences in psychiatric notes, processed by a large language model (Mistral).
• Controlled for patient demographics, risk factors, and NSR to predict schizophrenia diagnosis.

Main Results:

• A high NSR significantly increased the odds of receiving a schizophrenia diagnosis.
• Clinician bias exposure, measured by NSR, attenuated the independent effect of patient race on diagnosis.
• Black male patients with high NSR exhibited the highest odds of schizophrenia diagnosis.

Conclusions:

• Sentiment-based metrics derived from clinical notes can effectively operationalize clinician bias exposure.
• Clinician bias contributes to psychiatric diagnostic disparities, extending beyond racial or ethnic factors alone.
• Addressing clinician bias in EDs is essential for mitigating diagnostic inequities in mental health care.