An assistant professor in the Milken Institute of Public Health and his research colleagues received a $1 million grant to research racial health disparities associated with common medical and surgical procedures, according to a release last week.
The National Institute on Minority Health and Health Disparities awarded the grant to Yan Ma, an associate professor and vice chair of the Department of biostatistics and bioinformatics and the principal investigator for the four-year project, the release states. Ma said he and his research team will use machine learning techniques – or algorithms and statistical models – to understand why people of color are more likely to die from or have complications with common joint, knee and hip surgeries.
“Using machine learning techniques will help researchers eliminate healthcare disparities by creating reliable data sources that could be used to assess health care utilization and health outcomes across underserved communities,” Ma said in an email.
He said large administrative databases like State Inpatient Databases and the National Impatient Sample, which tracks inpatient health care records, lack a “moderate” amount of data describing how health disparities affect different races. Ma said machine learning techniques automatically use algorithms to detect statistical dependencies when inputing missing data.
Ma said ensuring the accuracy of health care data broken down by race within these databases will help researchers track health disparities, document health care quality and measure health care outcomes.
He said the study will provide a “robust” statistical tool for filling in missing data on race and other “key” variables in health disparities research. Ma added that policymakers can use information from these databases to garner feedback from community and consumer advocacy groups about how to address health disparities.
“Beyond the methodological insights provided, this study will help to advance health disparities research by enhancing the ability to quantify, monitor and develop targeted solutions to addressing health disparities in and other procedures,” he said.
Health equity experts said the study will help researchers identify which subsets of the U.S. population health care inequity affects and fill in gaps of missing data on racial health differences among races.
William Boag, a graduate student who studies health equity, racial disparities and AI in the Computer Science and Artificial Intelligence Lab at the Massachusetts Institute of Technology, said quantifying health care disparities using machine learning algorithms will signal how and which populations experience health care inequity.
He said researchers can input data into machine learning algorithms, which can determine in what ways and to what extent health care disparities are occurring, like when measuring prescribing patterns for pain medication for white versus minority patients.
Boag said mandating implicit bias training for physicians and removing bias from algorithms used in hospitals to allocate medical resources and personnel and dictate treatment methods can potentially address health care disparities.
Prejudice is sometimes “built in” to machine learning algorithms, which reflects the implicit bias of individuals who design the algorithms, according to the technology website Tech Target.
“One thing that’s fortunate about algorithms is that the bias is always hard to fix and remedy, but it’s probably easier to update an algorithm than it is to update a human decision-making process,” Boag said.
Brian Powers, a clinical fellow in medicine at Brigham and Women’s hospital in Boston, said machine learning techniques can help researchers draw new conclusions from existing data on the prevalence of health care disparities. He said that if the techniques are not implemented properly, machine learning algorithms can reinforce disparities among races.
Powers added that Ma’s research is a “necessary first step” to addressing health care disparities among races.
“The work by Ma and colleagues should allow for future research that identifies how these disparities arise, and what can be done to eliminate them,” Powers said.
Steve Goodman, a professor of medicine and of health research and policy at Stanford School of Medicine, said researchers may be using machine learning to improve predictions of how race contributes to health care disparities.
“It could be that they are able to get more accurate predictions of that missing data from the data that they have,” Goodman said. “Then they might be proposing to use different kinds of more complex models to improve the prediction of that missing data.”