Newswise — INDIANAPOLIS –The collection of information on nonmedical factors that affect health outcomes, referred to as social determinants of health, often takes place during medical appointments. However, this valuable information is typically recorded in the form of text within the clinical notes authored by healthcare professionals such as physicians, nurses, social workers, and therapists.
A recent study conducted by researchers from the Regenstrief Institute and Indiana University Fairbanks School of Public Health showcased one of the pioneering applications of natural language processing in the context of social determinants of health. The research team developed three novel algorithms utilizing natural language processing techniques, which proved successful in extracting relevant data from textual information present in electronic health records pertaining to housing challenges, financial stability, and employment status.
"Health and well-being encompass more than just medical care. They are predominantly influenced by our behaviors, environment, and social connections," stated Joshua Vest, PhD, a Research Scientist at the Regenstrief Institute and faculty member at the Fairbanks School of Public Health, who led this study. He further emphasized, "Healthcare organizations are increasingly compelled to address social determinants, as factors like financial resources, housing, and employment status significantly contribute to the costs that result in adverse health outcomes. The challenge for healthcare organizations lies in effectively measuring and identifying patients facing social risks, enabling targeted interventions."
"Our research contributes to the advancement of both application and methodology in the field. While natural language processing has been utilized for various medical conditions in previous studies, our paper represents one of the pioneering efforts to apply it specifically to social determinants of health. We demonstrated that a relatively simple natural language processing approach can effectively measure social determinants, eliminating the need for more complex deep learning and neural network models. These advanced models are indeed powerful, but their implementation is challenging, complex, and requires extensive expertise, which may be lacking in many healthcare systems."
Dr. Vest further explained, "We intentionally developed a system that can operate in the background, comprehensively analyze all medical notes, and generate tags or indicators indicating the presence of data suggesting potential concerns related to social determinants of health. Our ultimate objective is to measure social determinants with sufficient accuracy to enable researchers to build risk models and allow clinicians and healthcare systems to incorporate these factors—housing challenges, financial security, and employment status—into routine practice. This integration will not only assist individuals in need but also provide a deeper understanding of the overall characteristics and requirements of the patient population."
Extracting information about social needs can encompass various data types within an electronic medical record, including patient occupation, health insurance coverage, marital status, household size, address (indicating low or high crime area), and frequency of address changes.
In a previous endeavor, Dr. Vest and colleagues, including Shaun Grannis, M.D., Vice President for Data and Analytics at the Regenstrief Institute, developed an application named Uppstroms (Swedish for "upstream"). They successfully demonstrated its ability to use structured data for predicting patients who would benefit from referrals to social services, such as nutritionists.
“Natural language processing-driven state machines to extract social factors from unstructured clinical documentation” is published in JAMIA Open.
RSS