英语轻松读发新版了,欢迎下载、更新

Machine learning uncovers social risk clusters linked to suicide across U.S. | Cornell Chronicle

2025-05-19 13:07:24 英文原文

作者:By Bridget Kuehn Weill Cornell Medicine

Using machine learning technology, a new study has identified three distinct profiles describing social and economic factors that are associated with a higher risk of suicide.

Scientists at Weill Cornell Medicine and Columbia University Vagelos College of Physicians and Surgeons led the research that showed suicide rates vary significantly across the three clusters and that the patterns differ geographically across the United States.

The findings, published May 12 in Nature Mental Health, could facilitate more effective prevention strategies and thereby help counter the substantial rise in suicide rates over the past two decades in the U.S.

This is the first study to use unsupervised machine learning to analyze a comprehensive set of social determinants of health such as poverty, poor housing, lack of access to health care, harmful environmental exposures and social factors like high family stress, which can all contribute to suicide risk. While prior prevention efforts largely targeted individual or clinical risk factors, this research emphasizes the importance of broader, community-level social conditions.

Unsupervised machine learning can process massive datasets without labels or guidance to discover hidden patterns and relationships unbiased by researchers’ assumptions or partial data. This method allowed the researchers to characterize the overall social and environmental landscape in 3,018 counties, based on 284 social determinants of health. Three distinct clusters were identified which the researchers correlated with suicide rates from 2009 to 2019, after controlling for sex, age and race/ethnicity.

“Our findings could help public health workers develop more tailored interventions that address the specific and different social determinants of health profiles that each community faces to more effectively lower suicide rates,” said lead author Yunyu Xiao, assistant professor of population health sciences and psychiatry at Weill Cornell Medicine. Yuan Meng, a postdoctoral associate in population health sciences, also contributed to the analysis.

One of the clusters, called “REMOTE,” affected people living in remote rural or mountainous areas that often rely on coal or other energy sources contributing to pollution. These individuals tended to be elderly and living in areas with aged, low-quality housing and abandoned homes. Suicide deaths in this cluster predominantly involved men and frequently included firearms.

The second cluster, “COPE,” included individuals experiencing complex family dynamics and severe environmental and social stressors. For example, single parents or grandparents raising their grandchildren created different family structures, many in poverty. Predominantly living in the southern U.S., these communities face harsher environmental factors, including extreme heat. Suicide deaths in this cluster were more common among middle-aged white individuals.

People living in racially and ethnically diverse metropolitan areas on the coasts were more likely to be part of the third cluster, called “DIVERSE.” Many of these communities have large immigrant populations with extreme income inequality, high housing costs, poorer air quality and difficulties accessing health care despite the presence of hospitals and clinicians in their area. Suicide deaths associated with this cluster were higher among women, youth and Black or Hispanic individuals.

“This research moves beyond the idea that suicide is only an individual or medical issue,” Xiao said. “Instead, it shows that the places we live – shaped by history, policy and economics – play a powerful role in shaping who is at risk.”

“Our findings suggest suicide prevention approaches based on modifying social determinants of health must be region- and population-specific, rather than applying the same intervention strategies across the United States,” said senior author Dr. John Mann, the Paul Janssen Professor of Translational Neuroscience in Psychiatry and Radiology at Columbia University and the New York State Psychiatric Institute.

Interventions for the REMOTE cluster, for example, may focus on reducing social isolation, increasing access to mental health care and addressing gun-related risks, Xiao said. In contrast, community-based interventions addressing economic stress and substance use, including alcohol and opioid overdoses, may help the COPE cluster. For the DIVERSE cluster, improving culturally adapted mental health programs, increasing health care accessibility and adapting measures to enhance air quality could have a positive impact.

By tracking changes over time, the researchers were also able to identify factors that reduced suicide rates previously, such as Medicaid expansion in certain counties, which improved health care access and affordability. These areas saw a decrease in suicides.

Next, Xiao and her colleagues will see if they can connect data on regional social determinants of health, suicide rates and electronic health records to gain an even clearer picture of the factors driving suicide clusters. They also hope to test specific interventions targeting each of the three clusters.

This work was supported by the National Institutes of Health; the National Institute of Mental Health; the National Institute on Drug Abuse’s Center for Health Economics of Treatment Interventions for Substance Use Disorder, HCV and HIV; the National Institute on Drug Abuse; and the Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Researcher Diversity.

If you or someone you know is experiencing thoughts of suicide, call or text the National Suicide Prevention Hotline at 988.

Bridget Kuehn is a freelance writer for Weill Cornell Medicine.

关于《Machine learning uncovers social risk clusters linked to suicide across U.S. | Cornell Chronicle》的评论


暂无评论

发表评论

摘要

A new study using machine learning identified three distinct profiles associated with higher suicide risk based on social and economic factors in the U.S. The research, published in Nature Mental Health, highlights variations in suicide rates across geographic clusters: REMOTE (rural areas), COPE (southern regions with complex family dynamics), and DIVERSE (coastal metropolitan areas). This is the first study to use unsupervised machine learning on a comprehensive set of social determinants of health. The findings could help develop targeted prevention strategies based on specific community profiles, addressing broader social conditions rather than individual risk factors alone.