Five Things to Know about AI and Health Care
Radiology Is Hot, Diagnosis Is Still in The Future
Blog Post
Clinicians commonly make predictions about disease risk, treatment effectiveness, and other outcomes in an effort to improve patient care and assist medical decision making. To do so, they often use prediction models that take various factors into account, such as patient age. Recently, the question of whether to use measures of race in clinical prediction models has emerged.
In a new study, LDI Senior Fellow Atheendar Venkataramani, Northwestern University economist Charles Manski, and University of Wisconsin economist John Mullahy unpack the use of measures of race in these clinical prediction models and provide a useful framework to focus ongoing debates.
Numerous arguments for and against the inclusion of race in clinical prediction have been posed. Those opposed to including race assert that it is a social concept, not a biological one, and should not be considered if no causal link exists between race and illness. Some suggest that using race may further worsen health inequities.
But analysis by Venkataramani’s team shows that patients across racial groups can fare better when clinical decision making is informed by race.
Venkataramani and colleagues develop a model of clinical prediction and decision making, allowing them to evaluate how predictions may change when race is included or excluded. They assumed an utilitarian ethical framework which posits that the clinician seeks to maximize the health of each individual patient based on all of the information available during treatment.
The model allows for clinicians to consider as many patient attributes as possible—including gender, socioeconomic status, health history, genetic information, and biomarkers—to improve the accuracy of their predictions. To reflect real-world care, they did not assume that clinicians can predict patient outcomes with certainty.
By comparing different scenarios, the investigators found that observing more patient attributes allowed clinicians to improve their predictions of treatment outcomes, enabling them to choose the most beneficial treatment courses for each individual patient. The framework suggests that, as long as measures of race have some predictive power, outcomes for patients of all races are optimized when patient race is considered.
Even when considering preventive measures, aimed at addressing how systemic factors such as structural racism affect disease risk, their analysis still highlighted the advantages of using measures of race for prediction at the time of a clinical encounter. Findings show that if better measures, such as measures of exposures to structural racism or ancestry or more accurate biomarkers that reflect these life course processes, become available, they replace measures of patient race, if race measures no longer hold predictive power when these additional predictors are included. Overall, it is important to weigh the potential benefits of using race against the potential costs identified by those in favor of excluding race.
Venkataramani and his colleagues point out that it is important to clarify and strive for achieving multiple clinical and population goals; closing the racial health gap and achieving better individual patient outcomes need not be mutually exclusive.
Evolving considerations of how to use race in clinical decision making will require the establishment of clearer policy goals. For example, the U.S. Department of Health and Human Services has recently undertaken efforts to revise Section 1557 of the Affordable Care Act which prohibits discrimination in health care on the basis of race, color, national origin, sex, age, or disability. These efforts include a new proposal that focuses on discrimination and clinical algorithms. Because models that include race can notably impact disparities in treatment or health outcomes, regulators must offer clearer definitions and guidelines to guide clinical decision making.
Ultimately, by conducting this research, Venkataramani and colleagues provide a useful framework to inform current debates regarding the use of race in clinical decision making. While the conversation evolves, their work provides evidence that using measures of race has the potential to benefit all patients across racial groups, and offers a nuanced lens to inform the conversation.
The study, “Using Measures of Race to Make Clinical Predictions: Decision Making, Patient Health, and Fairness” was published in the Proceedings of the National Academy of Sciences on August 22, 2023. The authors are Charles F. Manski, John Mullahy, and Atheendar S. Venkataramani.
Radiology Is Hot, Diagnosis Is Still in The Future
At an Oct. 10 Conference, Experts from LDI and the Wharton Healthcare Analytics Lab Will Discuss the Thorniest Questions About AI
First-of-Its-Kind National Analysis Shows Racial and Geographic Disparities
LDI Fellows Offer Specific Ways to Avoid Disparities and Serve Patients Better
Evidence from Philadelphia During the Pandemic Suggests Ways To Build Trust
Penn LDI Panel Unpacks Barriers and Options for Change