Dissecting racial bias in an algorithm used to manage the health of populations

There is no author summary for this article yet. Authors can add summaries to their articles on ScienceOpen to make them more accessible to a non-specialist audience.

Abstract

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.

Related collections

Most cited references 22

Record: found
Abstract: found
Article: not found

Chronic kidney disease and mortality risk: a systematic review.

Marcello Tonelli, Natasha Wiebe, Bruce Culleton … (2006)

Current guidelines identify people with chronic kidney disease (CKD) as being at high risk for cardiovascular and all-cause mortality. Because as many as 19 million Americans may have CKD, a comprehensive summary of this risk would be potentially useful for planning public health policy. A systematic review of the association between non-dialysis-dependent CKD and the risk for all-cause and cardiovascular mortality was conducted. Patient- and study-related characteristics that influenced the magnitude of these associations also were investigated. MEDLINE and EMBASE databases were searched, and reference lists through December 2004 were consulted. Authors of 10 primary studies provided additional data. Cohort studies or cohort analyses of randomized, controlled trials that compared mortality between those with and without chronically reduced kidney function were included. Studies were excluded from review when participants were followed for < 1 yr or had ESRD. Two reviewers independently extracted data on study setting, quality, participant and renal function characteristics, and outcomes. Thirty-nine studies that followed a total of 1,371,990 participants were reviewed. The unadjusted relative risk for mortality in participants with reduced kidney function compared with those without ranged from 0.94 to 5.0 and was significantly more than 1.0 in 93% of cohorts. Among the 16 studies that provided suitable data, the absolute risk for death increased exponentially with decreasing renal function. Fourteen cohorts described the risk for mortality from reduced kidney function, after adjustment for other established risk factors. Although adjusted relative hazards were consistently lower than unadjusted relative risks (median reduction 17%), they remained significantly more than 1.0 in 71% of cohorts. This review supports current guidelines that identify individuals with CKD as being at high risk for cardiovascular mortality. Determining which interventions best offset this risk remains a health priority.

0 comments Cited 515 times – based on 0 reviews      Review now

Bookmark

Record: found
Abstract: not found
Article: not found

The effect of patient race and socio-economic status on physicians' perceptions of patients

Michelle van Ryn, Jane Burke (2000)

0 comments Cited 202 times – based on 0 reviews      Review now

Bookmark

Record: found
Abstract: found
Article: not found

Glycated haemoglobin, diabetes, and mortality in men in Norfolk cohort of european prospective investigation of cancer and nutrition (EPIC-Norfolk).

Ailsa A. Welch, R. Luben, Jeremy Day … (2001)

To examine the value of glycated haemoglobin (HbA(1c)) concentration, a marker of blood glucose concentration, as a predictor of death from cardiovascular and all causes in men. Prospective population study. Norfolk cohort of European Prospective Investigation into Cancer and Nutrition (EPIC-Norfolk). 4662 men aged 45-79 years who had had glycated haemoglobin measured at the baseline survey in 1995-7 who were followed up to December 1999. Mortality from all causes, cardiovascular disease, ischaemic heart disease, and other causes. Men with known diabetes had increased mortality from all causes, cardiovascular disease, and ischaemic disease (relative risks 2.2, 3.3, and 4.2, respectively, P /=7%, or history of myocardial infarction or stroke were excluded. 18% of the population excess mortality risk associated with a HbA(1c) concentration >/=5% occurred in men with diabetes, but 82% occurred in men with concentrations of 5%-6.9% (the majority of the population). Glycated haemoglobin concentration seems to explain most of the excess mortality risk of diabetes in men and to be a continuous risk factor through the whole population distribution. Preventive efforts need to consider not just those with established diabetes but whether it is possible to reduce the population distribution of HbA(1c) through behavioural means.