RISE looks at recent headlines concerning social determinants of health (SDoH).
Uber Health, Socially Determined partner in SDoH initiative
Uber Health and Socially Determined, a social risk analytics and solutions company, have announced a new partnership to address SDoH including transportation, food security, and health care access. The collaboration will help payers and providers identify patients in need of supplemental benefits such as transportation to medical appointments and food and prescription delivery.
“Historically, the onus has been on patients to navigate their own benefits—from figuring out what they’re eligible for, to tracking down those services, to securing reimbursement. We’re turning that model on its head,” said Caitlin Donovan, global head of Uber Health, in the announcement. “Now, plans and providers can turn to Socially Determined to gain insight into what their patients need and then leverage the Uber Health platform to coordinate access to those items and services. This enables health care organizations to take a more strategic, proactive, and impactful approach to patient care, driving better outcomes at scale.”
The partnership aims to support the more than 3,000 health care organizations currently using Uber Health to address patients’ unmet social needs.
CMS approves New York’s section 1115 demonstration amendment to advance health equity, behavioral health
The Centers for Medicare & Medicaid Services (CMS) has approved a “groundbreaking” amendment to New York’s Medicaid section 1115 demonstration, which aims to improve primary care, behavioral health, and health equity, according to the U.S. Department of Health and Human Services.
The approval will allow New York to establish base rates for safety net providers serving vulnerable communities throughout the state, connect individuals struggling with housing or food security with necessary support services, improve access to treatment for substance use disorders, and grow the state’s health care workforce.
RELATED: White House releases first SDoH playbook
“As the nation’s largest insurer, CMS is proud to approve this critical demonstration amendment, which gets to the heart of Medicaid’s role as an innovator,” said CMS Administrator Chiquita Brooks-LaSure in the announcement. “The demonstration’s initiatives will provide a broad swath of health and social supports to underserved communities, improving their health and quality of life. We encourage other states to follow New York’s efforts to address health disparities.”
Study: Generative AI can detect SDoH in EHRs
Generative artificial intelligence (AI), particularly large language models (LMs), could play a key role in identifying SDoH in electronic health records (EHRs), according to a recent study conducted by researchers at Mass General Brigham.
Based on their findings, the study indicates that LMs can pull SDoH data from clinician’s notes, which is a key component to accurately capturing patients’ SDoH.
“SDoH are rarely documented comprehensively in structured data in the electronic health records, creating an obstacle to research and clinical care,” researchers wrote. “Instead, issues related to SDoH are most frequently described in the free text of clinic notes, which creates a bottleneck for incorporating these critical factors into databases to research the full impact and drivers of SDoH, and for proactively identifying patients who may benefit from additional social work and resource support.”
For the study, the research team reviewed 800 clinician notes from 770 patients with cancer who received radiotherapy at Brigham and Women’s Hospital/Dana-Farber Cancer Center from 2015 to 2022. Researchers then used the data to explore how effective LMs are in extracting the SDoH data. Six SDoH categories were included in the study: employment, housing, transportation, parental status, relationship, and social support.
The study models identified 93.8 percent of patients with adverse SDoH, whereas the standard ICD-10 codes only captured two percent.
The researchers also compared their fine-tuned LMs with ChatGPT-family models, GPT3.5 and GPT4, and found the LMs to outperform the ChatGPT models; however, the fine-tuning of the LMs is critical at this time, they explained, noting that future engineering of the models could further improve model performance.
“This is an area for future study, especially once these models can be readily used with real clinical data,” they wrote. “With additional prompt engineering and model refinement, performance of these models could improve in the future and provide a promising avenue to extract SDoH while reducing the human effort needed to label training datasets.”