Prejudice Based on Race in Health Care: Decision-Making Resources

Prejudice Based on Race in Health Care: Racial bias can rear its ugly head in some surprising ways in medical care. One example: the clinical decision tools that are crucial to how patients are evaluated, diagnosed, and treated today.

These tools use algorithms, or step-by-step methods, usually computerized, to determine factors such as the risk of heart disease, the need for a chest X-ray, and prescription drug dosages. Artificial intelligence can examine health records and billing systems to provide the data sets that are needed.

At first glance, these tools might seem objective. However, research shows that the data analysis used in these algorithms can be significantly skewed against specific racial and socioeconomic groups. This bias can have a profound impact on the quantity and quality of health care people in these categories receive.

A 2019 study found that an algorithm used by many U.S. hospitals and insurers to dole out extra health management support was systemically biased against Black people. The decision tool was less likely to send Black than White patients to care-management programs for complex medical needs when the two ethnic groups were similarly ill.

The bias originated from how the algorithm ranked patients by risk score, based on their medical expenses in the prior year. The premise was that identifying high-cost patients would reveal those with the greatest medical needs. But many Black patients have less access to, less ability to pay for, and less trust in medical care than white persons who are equally sick. In this case, their decreased medical bills did not appear to be a good predictor of their health state.

Care management programs use high-touch approaches to address the complex needs of the sickest patients, such as phone calls, nurse home visits, and prioritizing doctor visits. These programs have been found to enhance outcomes, reduce emergency room visits and hospitalizations, and save medical costs. The programs are expensive and targeted to those with the highest risk scores. Scoring methods that discriminate against the sickest black individuals for this care may be largely responsible for the higher risk of death from numerous diseases.

Some technologies deliberately use race as a criterion, yet algorithms can still be biased even without using race as a variable. Take the eGFR score, which measures kidney health and eligibility for kidney transplants.

A 1999 study that established the eGFR score found that Black persons, on average, had greater amounts of creatinine (a consequence of muscle breakdown) than White people. The scientists attributed the higher levels to greater muscular mass in Blacks. So they altered the scoring so that, in effect, Black persons had to have a lower eGFR score than Whites to be identified as having end-stage renal disease. As a result, Black people had to wait until their renal illness was further advanced to receive treatment.

A student in the University of Washington School of Medicine in Seattle, studying medicine and public health, observed in 2018 that eGFR levels were inaccurate in detecting the severity of renal disease in Black patients. She struggled to get race out of the algorithm, and she prevailed. In 2020, UW Medicine concluded that using race was not an effective variable and did not meet scientific rigor standards for medical diagnostic tools.

Even the most basic medical decision tool that does not contain race might reflect social bias. For example, body mass index (BMI) is a measure of weight relative to height. It is used to diagnose underweight, overweight, and obese people.

In 1985, the National Institutes of Health linked the term obesity to a person’s BMI, and in 1998, an expert panel adopted BMI-based criteria that reclassified 29 million Americans who had previously been considered normal weight or just overweight as overweight and obese.

Today, under BMI criteria, most Black, Hispanic, and White people are overweight or obese. But a 2021 report from the Centers for Disease Control and Prevention (CDC) indicated that the percentage of Americans who may be classed as obese differs by race or ethnic group.

Labeling such large percentages of people as overweight or obese has fostered an atmosphere of weight-shaming and mistrust between patients and doctors. Heavier patients report that doctors ignore the actual health problems or concerns that prompt a check-up, instead attributing all health difficulties to weight and urging weight loss as the solution. As a result, many Black and Hispanic patients avoid health care providers, potentially missing opportunities to prevent or detect problems early.

Besides, it is becoming more and more evident that being overweight or obese is not always a health risk. Obesity is linked to greater rates of some major illnesses, including heart disease, stroke, type 2 diabetes, and some types of cancer. But in other instances, such as post-heart surgery, being overweight or moderately fat (but not morbidly obese) is connected with increased survival.

New obesity guidelines for Canadian clinicians, issued in August 2020, stress that doctors should avoid using BMI alone to diagnose patients. The new rules mean that people should only be labeled as obese if their body weight impacts their physical health or emotional well-being. Treatment should be holistic, not just about weight loss. The guidelines also say that “People living with obesity suffer from significant bias and stigma, which contribute to higher morbidity and mortality irrespective of weight or body mass index.

Other metrics, such as waist circumference, may serve as a substitute for considering an individual’s BMI. And obesity itself may be redefined.” In January 2025, 58 scholars proposed a new definition that would abandon BMI in favor of excess body fat and its health impacts. The scientists defined two types of obesity: preclinical, in which people carry excess fat, but their organs function normally, and clinical, in which excess fat damages organs.

But medical algorithms aren’t the only form of algorithm that might be prejudiced. “This problem is not unique to medicine,” a 2020 article in The New England Journal of Medicine highlighted. The criminal justice system, for example, uses recidivism-prediction technologies to inform choices about bail amounts and prison sentences. One of the most popular tools, the authors wrote, “does not use race per se, but uses many factors that correlate with race and gives higher risk scores to black defendants.”

The increased use of artificial intelligence (AI) – specifically machine learning – has also highlighted problems of bias based on race, socioeconomic class, and other characteristics. In healthcare, machine learning is commonly built on electronic health information. Poor and minority patients may receive fragmented care and may be seen at various facilities. They are more likely to be seen in teaching clinics where data input/clinical reasoning may be less precise. And they might not be able to access online patient portals and document outcomes. Therefore, the records of these patients may have missing or incorrect data. This could lead to algorithms that drive machine learning, leaving out poor and minority patients in the data sets and the care they receive.

Fortunately, the understanding of biases in healthcare algorithms has increased over the past few years. They are analyzing data input and outcomes for bias by race, ethnicity, income, gender, and age. U.S. medical specialty associations are recognizing the consequences of race-based medicine and working to exclude race from clinical algorithms. If there are discrepancies, the algorithms and data sets can be modified to be more objective.

There is no accepted legal or scientific definition for an algorithm. The National Institute of Standards and Technology defines it as “A clearly specified mathematical process for computation; a set of rules that, if followed, will give a prescribed result.

In its broadest definition, an algorithm is a step-by-step method for arriving at an answer to a query or accomplishing a desired result. For example, a cake recipe is an algorithm. In finance, that would be an algorithmic trading system.

Machine learning, a subset of artificial intelligence (AI), is defined by IBM, a pioneer in the subject, as “the set of algorithms that can ‘learn’ the patterns of training data and then generalize from those patterns to make accurate inferences about new data.”

The algorithms doctors use to make certain judgments can be biased along lines of race, class, and more, even when they appear impartial and objective. Thus, algorithms must be analyzed rigorously and not taken on trust. “The term ‘algorithm,’ however defined, shouldn’t be a shield to absolve the humans who designed and deployed any system of responsibility for the consequences of its use,” a 2021 article in the MIT Technology Review noted.