Algorithms predict need for social determinants of health services

Pilot program at Eskenazi Health seeks to further refine models and improve their accuracy in indicating the need for patient referrals to programs.


Algorithms developed by Indiana University-Purdue University Indianapolis and the Regenstrief Institute have been shown to accurately predict the need for social determinants of health (SDOH) service referrals among patients at a safety-net hospital by leveraging clinical and community-level data.

IUPUI and Regenstrief researchers utilized data from 48 socioeconomic and public health indicators to build the “random forest” decision models predicting the need for mental health, dietitian, social work and other SDOH service referrals for patients at Eskenazi Health in Indianapolis.

“There is a growing recognition that the reason people have poor health and negative outcomes is frequently not a factor of medical care, but it is the broader social and environmental conditions that they find themselves in,” says Joshua Vest, a Regenstrief investigator and director of the Center for Health Policy in the Richard M. Fairbanks School of Public Health at IUPUI.

According to Vest, the goal of the study—funded by the Robert Wood Johnson Foundation—was to internally develop predictive algorithms to identify Eskenazi Health patients who need services that better address SDOH, including poor living conditions, income instability, lack of access to transportation and legal issues. Further, he contends that the study is the first time that anyone has tried to use algorithms to predict the need for these kinds of services.

“The data for this comes from myriad sources, including electronic health records, order entry data and information from notes through natural language processing,” Vest adds. “We also leveraged the Indiana Network for Patient Care, the area’s health information exchange, to pull data from across the entire state. So, we have clinical encounters from any healthcare system effectively in Indiana.”

Also See: Leveraging social determinants of health data remains challenge

In addition, Vest says researchers used public health data based on surveys and vital statistics that were also incorporated into their predictive models.

Results of the study, recently published in the Journal of the American Medical Informatics Association, indicate that the “need for various social service referrals can be predicted with considerable accuracy using a wide range of readily available clinical and community data that measure socioeconomic and public health conditions.”

According to the article, algorithms predicting the need for any, mental health and dietitian referrals yielded sensitivity, specificity and accuracy measures ranging from 60 percent to 75 percent. Specificity and accuracy scores for social work and other SDOH services ranged from 67 percent to 77 percent, while sensitivity scores were from 50 percent to 63 percent.

“Our models fall into what we call the range of clinically useful,” concludes Vest. “They are consistent with other types of prediction models. We, of course, are always trying to make models better.”

Going forward, he says IUPUI and Regenstrief researchers will continue to refine the models to improve their prediction value. Currently, they are piloting the use of the algorithms to evaluate whether they change referral patterns for SDOH services for patients at Eskenazi Health. Vest estimates that about half of the adult population of the hospital system could benefit from some SDOH service.

More for you

Loading data for hdm_tax_topic #better-outcomes...