51视频

04/20/2023
By Brooke Coupal

Members of the Center of Biomedical and Health Research in Data Sciences (CHORDS) want to improve suicide prevention.

鈥淪uicide has touched everyone鈥檚 lives,鈥� says CHORDS Director Hong Yu, a Miner School of Computer and Information Sciences聽professor. 鈥淚 believe our work has the potential to prevent deaths by suicide.鈥�

On average, 121 adults in the United States died by suicide every day in 2020, with veterans having a 57.3% higher suicide rate than non-veteran adults, according to a published in September 2022.

CHORDS researchers, including Yu, Biological Sciences Asst. Prof. Rachel Melamed, Biomedical and Nutritional Sciences Prof. Katherine Tucker and Weisong Liu, a Miner School research assistant professor and CHORDS assistant director, examined the link between suicide risk among veterans and social determinants of health, such as housing instability, financial problems and violence.聽The research was funded by 聽grants totaling more than $3 million.

Their findings, recently published in 鈥�,鈥� showed how natural language processing can be used to analyze all available information about social determinants of health, leading to better suicide risk assessment and prevention.

Critical Data Often Overlooked

When investigating causes of suicidal behavior, Hong says researchers often turn to structured data from electronic health records, which include billing and disease codes. The problem with focusing on structured data is that it often lacks contextual information about the patients, including the social determinants of health.

鈥淪ocial determinants of health are severely undercoded,鈥� says Yu, who adds that this information is typically found in unstructured data, such as clinical notes from physicians, social workers and other providers.

With the lack of a comprehensive database for that type of information, Yu and her team sought a way to gather unstructured data to better analyze the association between suicide and social determinants of health.

The researchers developed a natural language processing system that could extract social determinants of health from unstructured clinical notes found in the U.S. Veterans Health Administration鈥檚 electronic health records, making it the first large-scale case study to ever do so.聽

From there, Richeek Pradhan, a co-author of the paper and Yu鈥檚 former graduate student at the UMass Chan Medical School, helped create a nested case-control study to compare data between veterans who had died by suicide with those who had not.

Using the unstructured data findings from the natural language processing system and the already available structured data, the researchers found that social determinants of health are linked with an increased risk of suicide, with the strongest association coming from legal problems and violence.聽

By using natural language processing, hospitals can better identify patients at risk of suicide based on their social determinants of health, the researchers say. Providers can then point patients to the appropriate services where they can get help, whether that is lawyers to assist with legal problems, food pantries for those facing food insecurity or shelters to help with housing instability.

鈥淭o reduce suicide, we need to reduce social determinants of health,鈥� Melamed says.

鈥淔or all of the social determinants of health that we focused on, there are specific social services that can help,鈥� Yu adds. 鈥淥ur goal is to improve the quality of care for patients and try to improve their outcomes.鈥�

Yu is leading several other multimillion dollar projects within CHORDS, including one recently funded by a $4 million聽聽grant that looks at social determinants of health and Alzheimer鈥檚 disease among veterans and another to be funded by a more than $1 million grant aimed at developing technologies to help prevent opioid overdoses.

鈥淚 want to keep building CHORDS and make it a research powerhouse,鈥� Yu says.