Q&A: AI identifies key social determinants of health for patients with dementia
Click Here to Manage Email Alerts
Key takeaways:
- Artificial intelligence accurately identified social determinants of health in five domains of interest.
- However, using AI is only the first step in addressing patients’ social needs, an expert said.
Artificial intelligence could help clinicians address the social needs of patients with dementia, according to Elham Mahmoudi, PhD.
Mahmoudi, an associate professor of health economics at the University of Michigan, and colleagues developed a rule-based natural language processing algorithm to identify social determinants of health among specific vulnerable populations.
Healio spoke with Mahmoudi to learn more about the new technology and how it can help primary care providers address social determinants of health.
Healio: What does this technology do? What is its intended purpose?
Mahmoudi: We used natural language processing to identify housing, transportation, food, medication insecurities, social isolation, abuse, neglect, exploitation and financial difficulties from the ED and inpatient social worker medical notes for patients diagnosed with Alzheimer’s disease and related dementia at Michigan Medicine. Many of these vulnerable patients live alone at home and may not have the resources or support to address their social needs. Our objective was to develop an algorithm that can reliably identify the social needs of these patients. Clinicians and social workers can proactively use this information to address the social needs of vulnerable patient populations.
Healio: Will you briefly describe your findings and their importance?
Mahmoudi: We developed and compared the rule-based algorithm with two other methods (eg, deep learning and regularized logistic regression). Our rule-based approach was able to accurately identify social determinants of health for all seven domains of interest except housing and medication. Hospitals may use this algorithm to identify the specific social needs of patients and, by acting upon this information, proactively try to prevent adverse health events such as potentially preventable emergency visits, hospitalizations or premature mortality among vulnerable patient populations.
Healio: What are the clinical implications of your findings?
Mahmoudi: Research has shown that social determinants of health are important predictors of health, well-being and adverse health events. However, health care systems in the U.S. do not systematically collect and update this information via patients’ medical records. Our study shows that, through the use of AI, we can reliably identify this information by searching medical notes in electronic medical records. This, however, is only the first step. To make this information useful, health care systems need to develop and maintain connections with patients and community resources to address patients’ social needs in a sustainable way. Preventable emergency visits and hospitalizations are expensive and may cause harm. Many of these adverse health events and the high costs associated with them could be avoided if health care settings address patients’ social determinants of health needs in a timely manner.
Healio: What are the pros and cons of using AI like this in medicine? How should the program be improved?
Mahmoudi: Using AI technology enabled us to efficiently search through patients’ medical notes to identify their specific social determinants of health needs. Otherwise, identifying and validating this information by solely relying on a manual search of medical notes was inefficient, unreliable and simply impossible. We do not plan to use this information for commercial purposes. I would be less enthusiastic about the use of commercial and nonregulated AI technology. This technology is in its infancy. We need to rigorously continue evaluating our algorithms to not only improve the accuracy, reliability and generalizability of the findings, but also to find alternative approaches to identify social determinants of health where AI technology or medical notes do not deliver good results.
Healio: What is the take-home message for PCPs?
Mahmoudi: Social determinants of health, which include food, housing and transportation insecurities, are salient social aspects of health. PCPs and other medical providers should work to identify, understand and address both the social and medical needs of patients in order to improve population health. This is not an easy task because health care systems in the U.S. are not designed to address social determinants of health. To make this effort sustainable, cost-effective and value-added, health care systems need to build a reliable communication infrastructure with patients and community resources. Addressing the social determinants of health requires collaboration among PCPs, community resources and patients. AI is just a tool to identify social determinants of health. Our main objective is to effectively address these social needs.
Healio: Is there anything else you would like to add?
Mahmoudi: Now that we have developed and validated our algorithm, our next steps involve making the algorithm generalizable so different health care systems can modify and use it based on their specific needs. We are also in the process of examining the feasibility and efficacy of addressing the social determinants of health — found via our algorithm — of our vulnerable patient populations through the Department of Family Medicine’s established connections with community resources. We hope, through the use of this technology, we can improve the health and well-being of our patients in Michigan Medicine.