Deconstructing demographic bias in speech-based machine learning models for digital health

Health systems rely on commercial prediction algorithms to identify and help patients with complex health needs. We show that a widely used algorithm, typical of this industry-wide approach and affecting millions of patients, exhibits significant racial bias: At a given risk score, Black patients are considerably sicker than White patients, as evidenced by signs of uncontrolled illnesses. Remedying this disparity would increase the percentage of Black patients receiving additional help from 17.7 to 46.5%. The bias arises because the algorithm predicts health care costs rather than illness, but unequal access to care means that we spend less money caring for Black patients than for White patients. Thus, despite health care cost appearing to be an effective proxy for health by some measures of predictive accuracy, large racial biases arise. We suggest that the choice of convenient, seemingly effective proxies for ground truth can be an important source of algorithmic bias in many contexts. Show more

In 2 meta-analyses on gender differences in depression in nationally representative samples, we advance previous work by including studies of depression diagnoses and symptoms to (a) estimate the magnitude of the gender difference in depression across a wide array of nations and ages; (b) use a developmental perspective to elucidate patterns of gender differences across the life span; and (c) incorporate additional theory-driven moderators (e.g., gender equity). For major depression diagnoses and depression symptoms, respectively, we meta-analyzed data from 65 and 95 articles and their corresponding national data sets, representing data from 1,716,195 and 1,922,064 people in over 90 different nations. Overall, odds ratio (OR) = 1.95, 95% confidence interval (CI) [1.88, 2.03], and d = 0.27 [0.26, 0.29]. Age was the strongest predictor of effect size. The gender difference for diagnoses emerged earlier than previously thought, with OR = 2.37 at age 12. For both meta-analyses, the gender difference peaked in adolescence (OR = 3.02 for ages 13-15, and d = 0.47 for age 16) but then declined and remained stable in adulthood. Cross-national analyses indicated that larger gender differences were found in nations with greater gender equity, for major depression, but not depression symptoms. The gender difference in depression represents a health disparity, especially in adolescence, yet the magnitude of the difference indicates that depression in men should not be overlooked. (PsycINFO Database Record Show more