As health systems move into the world of addressing social needs or increasing connections with social service agencies, informaticians are researching how to use data to predict which patients could benefit from referrals to community-based organizations as well as how to send closed-loop e-referrals from EHRs that contain a limited amount of relevant patient data.
During a Feb. 19 Systems for Action webinar, Joshua R. Vest, Ph.D., M.P.H., director of the Center for Health Policy in the School of Public Health at the Indiana University Purdue University Indianapolis, described research work his organization is doing to implement machine-learning-based risk stratification to identify patients in need of what are called “wraparound” services such as housing, transportation and food assistance.
Vest began by noting that among the challenges facing healthcare organizations, there are increasing demands for accountability and the assumption of risk and a recognition that social factors related to poverty complicate care delivery, foster health disparities, and are important to health status. “The current medical care system is not designed to address these issues,” he added. “It is organized to get people into treatment and get them out, and is not conducive to a long-term, high-touch approach that social issues need. When a patient has a social need, primary care doctors are not the folks tasked with dealing with that need; social workers are trained and their work is structured to address those.”
In setting out to see if better referral systems to social work could have a positive impact, the researchers realized that one benefit they have in Indiana is lots of data. Because Indiana has had a statewide HIE for almost 20 years, it has lots of longitudinal clinical data to study. “We wanted to see if we could use that data to support delivery of services that address social risk and needs,” Vest said.
They built an algorithm leveraging claims, EHR data and other data sources including area measures on safety, transportation, and public health. By calculating measures around needs and risk, the algorithm predicts the need of a referral to social work.
The researchers at IUPUI, in partnership with Indianapolis-based Eskenazi Health, set out to test the value of providing clinicians with risk stratification information about which patients might benefit most from a referral to a social worker or nutritional counselor. Eskenazi Health operates safety net outpatient clinics for an urban population in Indianapolis. All social services are offered on a co-located basis. There are no referrals to outside organizations.
Eskenazi would send the researchers a list of patients scheduled for upcoming appointments. The research team would run the data and rank the patients as high, medium or low risk of needing a referral. Then during daily huddles before seeing patients, the clinicians and on-site social workers could look through that listing. In phase 1 of their study, they looked at nearly 240,000 encounters during the study period rolled out in a step-wedge trial, with pre- and post-observations for each group of clinics. They found that the introduction of daily reports was associated with increases in social work referrals and kept appointments. In the next phase, they are working on adding more data sources and expanding the intervention to include the pediatric population. They also are working to integrate the decision support mechanism for clinicians directly into the EHR workflow.
During the same webinar, John Loonsk, M.D., chief medical information officer at CGI Federal Inc., described some work being done on e-referrals to community-based organizations. He noted that there are some obstacles to the adoption of processes for facilitated referrals, including integrating them into EHRs so they are easily accessible in the context of standing orders and clinical decision support. There is work going on, including the research at IUPIU, to identify those in need of referral and the creation of locator services that can help connect people.
Loonsk, who previously served as director of interoperability and standards in the Office of the National Coordinator for Health Information Technology, is part of a group working at HL7 on a Bidirectional Services eReferral (BSeR) project and FHIR implementation guide. “We are trying to work on the actual referral itself, including the conveyance of appropriate patient data,” he said. One limitation so far has been that many electronic referrals involve sharing a full patient summary, which includes a raft of data that is appropriate in clinical settings, but gives anxiety when shipped to extra-clinical programs that are not supported by full-fledged providers of medical care on the receiving end, Loonsk explained. “Technology has been limiting in this regard and impeding these referrals,” he added.
The BSER project has been working on payloads of referral data tied to chronic disease management programs involving diabetes, obesity, arthritis and others. The referral needs to initiate that involvement by leveraging data for that specific program. Referring someone to a weight loss program does not require their full patient record, Loonsk said. BSER is working on sending specific payloads and getting feedback back about patient progress in that activity. “The goal is to close the loop,” he added, “and increase engagement of the provider.”