Bias in AI-guided management of patients with major depressive disorders

Abstract

This study reports the extent of bias, and necessary corrections, in an AI system designed to guide management of Major Depressive Disorder. A previous general-population, race-blind, study had identified 1499 medical history risk factors for predicting response to antidepressants. These risk factors included 700 diagnoses, 550 current medications, 151 medical procedures, and 98 aspects of previous antidepressant use. We used hierarchical analysis to examine whether additional race-specific risk factors should be considered when predicting antidepressant response for African Americans. Depressed African Americans in our study took 1.74 different antidepressants during the study periods. In all 14 antidepressants, response to antidepressants for African Americans was better predicted, if we used race-specific models. The largest improvement in variation explained came in predicting response to Amitriptyline, Mirtazapine, and Nortriptyline. These data suggest clinicians should not rely on their experiences with the general population to prescribe medications for African Americans with Major Depressive Disorder.

Ähnliche Arbeiten