Can AI help hospitals spot patients in need of extra non-medical assistance?

Within the rush to harness synthetic intelligence and machine studying instruments to make care extra environment friendly at hospitals nationwide, a brand new research factors to a different potential use: figuring out sufferers with non-medical wants that would have an effect on their well being and skill to obtain care.

These social determinants of well being—every little thing from transportation and housing to food supply and availability of household and pals as helps—can play a serious position in a affected person’s well being and use of well being care providers.

The brand new research focuses on a affected person inhabitants with particularly advanced wants: folks with Alzheimer’s illness or different types of dementia. Their situation could make them particularly reliant on others to get them to medical appointments and social activities, deal with drugs and funds, store and put together meals, and extra.

The outcomes of the research present {that a} rule-based natural language processing instrument efficiently recognized sufferers with unstable entry to transportation, meals insecurity, social isolation, monetary issues and indicators of abuse, neglect, or exploitation.

The researchers discovered {that a} rule-based NLP instrument—a sort of AI that analyzes human speech or writing—was far superior to deep learning and regularized logistic regression algorithms for figuring out sufferers’ social determinants of well being.

Nonetheless, even the NLP instrument didn’t do nicely sufficient at figuring out wants associated to housing or affording or taking treatment.

The research was led by Elham Mahmoudi, Ph.D., a well being economist at Michigan Drugs, the College of Michigan’s educational medical middle, and Wenbo Wu, Ph.D., who accomplished the work whereas incomes a doctorate on the U-M College of Public Well being and is now at New York College. Mahmoudi and two different authors are within the Division of Household Drugs.

They and their colleagues in contrast the SDOH-spotting capabilities of three completely different AI methods, first coaching them on a set of 700 affected person information to show them the sorts of phrases and phrases to search for, after which utilizing them on 300 information and judging the outcomes.

The instruments solely regarded on the anonymized contents of emergency division and inpatient social employee notes made between 2015 and 2019 within the electronic health records of 231 sufferers with dementia.

Mahmoudi says the workforce is now working to prospectively validate the NLP algorithm towards the SDOH questionnaire that has not too long ago begun to be supplied to all main care sufferers at Michigan Drugs. That may enable them to match what the pc program finds and what sufferers say in response to questions on their scenario.

“We’re additionally making ready a pilot program that may consider the feasibility of an intervention that addresses these social determinants of well being, and join recognized folks with group sources,” mentioned Mahmoudi. “Within the meantime, we hope our present outcomes present that this algorithm can be utilized by clinicians, case managers and social staff to proactively deal with the social wants of sufferers with dementia, and doubtlessly different weak affected person populations.”

Along with Mahmoudi and Wu, the research’s authors are Kaes J. Holkeboer and Lorrie Carbone LMSW of the U-M Division of Household Drugs, and Temidun O. Kolawole of Johns Hopkins College.