Q&A: ‘Raw data’ on social determinants of health could help with patient referrals
A machine learning model that uses the most basic raw data may help physicians identify patients who need referrals to social workers and predict their risk for hospitalization.
“There is so much data available on social determinants of health, but the challenge is turning it into something health care providers can use,” Joshua R. Vest, PhD, MPH, a Regenstrief research scientist and professor at Indiana University Richard M. Fairbanks School of Public Health, said in a press release.
Vest and colleagues tested various algorithms on data from more than 200,000 patient encounters within the Eskenazi Health safety net health care system in Indiana, according to a press release. They found that the most accurate prediction model used raw data on housing, income and education. Vest said the algorithm — which they dubbed “Uppstroms” — offers health care professionals a practical way to use social determinants of health data.
“Raw data is a viable solution to creating a referral model,” he said in the press release. “There may be a need to develop more complex models in the future, but this can be a strong starting point for health systems to begin leveraging social determinants of health, especially systems that may have fewer resources.”
In an interview with Healio Primary Care, Vest provided more details about Uppstroms.
Healio Primary Care: How did you create and test Uppstroms?
Vest: Uppstroms is Swedish for upstream. This word reflects our thinking on trying to address social risks and social needs, which are the real drivers of a person’s health and well-being before they become challenges.
When my colleagues and I came to Indiana University and the Regenstrief Institute several years ago, we sat down with our clinical partner’s providers and their leadership and asked, “How can we help? What can we help you accomplish [more] and be more effective in care?”
Through those discussions came a strong desire to address patients’ social risk factors, such as homelessness, housing, food insecurity and those kinds of financial challenges.
However, it was very hard to identify patients before those challenges came up, and because their visits increase health care utilization, clinicians always felt they were being reactive, and not as addressing the problem in a proactive manner.
Healio Primary Care: What data are there to support Uppstroms’ use?
Vest: We drew on over 100 data elements and did several rigorous evaluations among ourselves to make sure that we were identifying the correct elements. Then when we rolled it out in a trial, we did see increases in referrals to social workers, which is exactly what we want to do when this risk stratification model was put out. We also saw cost savings through reductions in hospitalizations.
Healio Primary Care: A previous study suggests that pediatric and family medicine providers “lack comfort screening and addressing social determinants of health.” How does your tool assuage some of the discomfort providers may feel when addressing these determinants?
Vest: We took the approach of referring people to the appropriate providers. We said, “Let's automate the process and use what information we can have access to so we can help identify people without the additional burden on physicians.” That's been the driving approach in Uppstroms’ development. In most health systems and most health care organizations, it is not a physician who intervenes to address somebody's housing challenges or somebody’s unsafe living environment. It is going to be a social worker or a case manager.
Healio Primary Care: Do you think patients would be open to discussing sensitive information like income and education? How can physicians initiate this conversation?
Vest: Some patients are comfortable having these conversations. Other patients, particularly those who have had poor treatment and experienced systematic biases within the health care system, are less likely to have those conversations.
The physician can say to a patient, “I would like you to have a conversation with our social worker,” or “We'd like you to have a conversation with our social workers or dietitians, since they’re the ones who are best equipped to talk to you about other nonmedical issues. These are the folks specifically trained to have those conversations.”
Healio Primary Care: In addition to social work referrals and hospitalization prediction, what else could Uppstroms potentially be used for?
Vest: There are lots of opportunities to improve patient well-being and health through addressing social factors. The field is continuously expanding.
When we started work on Uppstroms, a good portion of our clinical partner’sstaff used in-house services. However, use of the model to connect patients to community-based organizations has increased. We are now interested in expanding our modeling to understand how we can effectively inform and help risk-stratify referrals to community-based organizations.
Healio Primary Care: How can physicians not affiliated with Regenstrief access Uppstroms? Is it compatible with existing electronic health record systems?
Vest: We would love to have more partners. We created Uppstroms to be completely vendor-neutral, so it can be widely implemented. It is a model that requires access and utilization of a lot of data and understanding of health care organization's priorities, workflows and data collection processes so that we can accurately stratify their population. People can visit uppstroms.com to see our research and get in touch with us about potential partnerships.
- Incorporating social determinants of health: Simplest solution may give the best results. https://www.regenstrief.org/article/incorporating-social-determinants-of-health-simplest-solution-may-give-best-results/. Published Oct. 7, 2021. Accessed Nov. 9, 2021.
- Lax Y, et al. BMC Health Serv Res. 2021;doi:10.1186/s12913-021-06975-3.