TL;DR:
- AI tools demonstrate potential in identifying non-medical needs affecting patient health.
- Social determinants of health (SDOH) impact patient well-being and healthcare utilization.
- Study targets Alzheimer’s and dementia patients with complex needs.
- Rule-based NLP tool excels in identifying transportation access, food insecurity, and social isolation.
- NLP outperforms deep learning and logistic regression for SDOH identification.
- Limitations include housing concerns and medication-related needs.
- Research led by Dr. Elham Mahmoudi and Dr. Wenbo Wu compares AI techniques.
- Anonymized patient records from 2015-2019 are analyzed.
- Prospective validation against SDOH questionnaire is underway.
- A pilot program is planned to address identified SDOH.
Main AI News:
Amidst the fervor surrounding the integration of artificial intelligence and machine learning into healthcare systems for enhanced efficiency, a groundbreaking study illuminates an alternative avenue: discerning patients’ non-medical requisites that could intricately influence their well-being and access to healthcare services.
These critical social determinants of health encompass a spectrum ranging from transportation and housing provisions to the availability of nourishment and a supportive network of family and friends. They wield considerable influence over a patient’s overall health and utilization of healthcare resources.
This study, a notable stride in healthcare research, narrows its focus on a subset of patients necessitating exceptional attention: those grappling with Alzheimer’s disease or other variations of dementia. The intricacies of their condition render them heavily reliant on external assistance to fulfill tasks like attending medical appointments, participating in social engagements, managing medications and finances, grocery shopping, meal preparation, and more.
Findings from this seminal study underscore the effectiveness of a rule-based natural language processing tool in accurately pinpointing patients beset by unstable transportation access, food scarcity, social seclusion, financial hardships, and manifestations of maltreatment, neglect, or exploitation.
Researchers observed that this rule-based NLP tool, a subtype of artificial intelligence designed to analyze human language patterns, outperformed deep learning and regularized logistic regression algorithms when it came to identifying patients’ social determinants of health.
However, even with its remarkable proficiency, the NLP tool exhibited limitations in identifying housing-related concerns and challenges pertaining to medication affordability and administration.
The driving force behind this groundbreaking research is a team led by Dr. Elham Mahmoudi, a distinguished health economist at Michigan Medicine, the esteemed academic medical center affiliated with the University of Michigan. Collaborating with her is Dr. Wenbo Wu, who undertook this work during the pursuit of a doctorate at the U-M School of Public Health and currently contributes to New York University. Notably, Mahmoudi and her co-authors, hailing from the Department of Family Medicine, have spearheaded this transformative endeavor.
The research collective diligently pitted three distinct AI techniques against each other, honing their capabilities through training on a dataset of 700 patient records. This training imbued these AI tools with the ability to recognize pertinent vocabulary and phrases. Subsequently, the researchers subjected the tools to the examination of 300 patient records, rigorously assessing their performance.
It is imperative to note that the AI tools exclusively perused anonymized content extracted from emergency department and inpatient social worker notes. These notes were generated during the period spanning 2015 to 2019, culled from the electronic health records of 231 dementia patients.
Dr. Mahmoudi and her team are at the cusp of a transformative journey. Their next phase involves the prospective validation of the NLP algorithm through a comparison against the Social Determinants of Health (SDOH) questionnaire now disseminated to primary care patients at Michigan Medicine. This pioneering validation exercise will juxtapose the findings of the algorithm with patients’ responses to inquiries concerning their circumstances.
In the words of Dr. Mahmoudi, “We are also in the process of devising a pilot program to evaluate the viability of an intervention that addresses these social determinants of health, thereby facilitating a connection between individuals with identified needs and the communal resources available.” She further adds, “In the interim, we believe our current findings underline the utility of this algorithm for clinicians, case managers, and social workers to proactively address the social requisites of dementia patients and potentially extend this approach to other vulnerable patient demographics.”
Conclusion:
The study highlights AI’s potential to identify critical non-medical factors impacting patient health, particularly in dementia cases. As AI algorithms evolve, their ability to address complex patient needs could pave the way for more proactive and comprehensive healthcare solutions, benefiting both patients and healthcare providers alike. This could open up opportunities for innovative solutions in the healthcare market that focus on personalized patient care and improved outcomes.
