Suicide Prediction Models Exacerbate Racial Disparities in Health Care

Depressed woman illustration.

nadia_bormotova / Getty Images

Key Takeaways

  • A study found that suicide prediction models predicted suicide in at-risk folks more accurately for certain racial groups compared to others.
  • Some experts suggest community-based intervention and more research to improve models.
  • A lack of access to healthcare contributes to these disparities.

When someone is at risk for suicide, there's a chance they'll turn to healthcare systems for help. Clinicians can then use statistical prediction models to determine who's most at risk, working quickly to intervene and provide care.

However, a new study suggests that some of these models exacerbate racial and ethnic disparities by more accurately predicting suicide deaths in some groups compared to others.

Specifically, suicide death prediction rates for Black, American Indian/Alaska Native (AI/AN), and patients without a recorded race or ethnicity were less accurate than those for White, Hispanic, and Asian patients. The study was published in JAMA Psychiatry in late April.

"Clinical implementation of these models would exacerbate existing disparities in mental health access, treatment, and outcomes for Black, American Indian, and Alaska Native populations," lead study author Yates Coley, PhD, biostatistician and investigator at Kaiser Permanente Washington Health Research Institute, tells Verywell. "We must test for disparities in accuracy and consider the possible negative consequences, including harm."

In 2018, suicide was the 10th leading cause of death in the United States, having increased 35% in the last 20 years. In the same year, suicide rates among AI/AN males were highest (34.8 per 100,000), followed by those among White, Hispanic, Black, and Asian males. Rates were overall lower for women, but AI/AN women and girls were most affected (10.5 per 100,000) followed by White, Asian, Black, and Hispanic women.

"AI/AN rates of suicide are remarkably high and have remained so for several decades," Lisa Wexler, PhD, MSW, professor at the University of Michigan School of Social Work who researches American Indian/Alaska Native suicide prevention and Indigenous youth resilience, but who was not involved with the study, tells Verywell. "For Black youth, particularly younger girls, suicidal behavior is growing at a fast pace. The difficulties of identifying risk in our models within these two populations signal an important reflection point to address."

Statistical Modeling for Suicide Prediction

Of the more than 1.4 million patients included in the data, 768 suicide deaths were recorded within 90 days after 3,143 mental health visits. In running the analyses, researchers focused on the number of visits of those who died by suicide, finding that suicide rates were highest for patients:

  • With no race/ethnicity recorded (313 visits)
  • Asian (187 visits)
  • White (2,134 visits)
  • American Indian/Alaskan Native (21 visits)
  • Hispanic (392 visits)
  • Black (65 visits)

Regardless of suicide rate or the number of healthcare visits, additional statistical tests found that prediction models were most sensitive to White, Hispanic, and Asian patients, and least sensitive to Black and AI/AN patients, and patients without race/ethnicity recorded.

This means that predictive models developed to assist healthcare systems in judging who's most at risk for suicide may be better at predicting for some groups rather than others, with Black and AI/AN patients at the biggest disadvantage.

The models used the following parameters to predict suicide:

  • Demographic characteristics
  • Comorbidities
  • Prior suicide attempts
  • Mental health and substance use diagnoses
  • Psychiatric medications
  • Prior mental health encounters
  • Responses to Patient Health Questionnaire 9

This data helped predict almost half of the suicides in White patients, but only 7% in AI/AN and Black patients.

"Many of the people who died by suicide accessed outpatient services," Wexler says, pointing out that those who died went to a median of two visits, with some having gone to five. "This means that there is a clear opportunity to intervene in meaningful ways to prevent suicide deaths."

For many experts in the field, these results aren't shocking. "I've been studying suicide for a long time," Kevin Early, PhD, CCJS, CAADC, ICAADC, sociology professor at the University of Michigan-Dearborn, tells Verywell. When he looked at the data, he says, he was not at all surprised. "It is clearly reflective of a deeper issue that is pervasive in American society, and that is inequality, disparity. It's not just disparity economically, politically, socially, but in the medical-industrial complex as well."

Existing Disparities Make an Impact

While the study draws attention to racial disparities in prediction models, Raymond Tucker, PhD, a psychology professor at Louisiana State University, tells Verywell that more research into specific racial and ethnic groups is needed to improve these models.

"There's a disparity in how we diagnose psychiatric illnesses," he says. This is important, considering one of the prediction models' main parameters was a previous psychiatric diagnosis.

For example, Tucker adds, Black men are overdiagnosed with schizophrenia compared to White men. "So there was a disparity in, and we should not be surprised that there's disparity out," Tucker says.

Coley adds that, while it's difficult to know for sure, she agrees: The fact that suicide rates were highest for people with unrecorded race/ethnicity could highlight historical disparities in healthcare.

"The one particular thing we saw in our data was that people without race and ethnicity recorded had a lower rate of common suicide risk factors," Coley says. In addition to psychiatric diagnosis, the other parameters like prior suicide attempts, medications, and prior mental health encounters could be making the models less accurate for certain groups. These individuals might not be able to access health care or may not opt for it, rendering the parameters irrelevant.

"This is something that we need to do more research into," Coley adds. "But we think that this finding really underscores the need for thorough auditing of prediction models before implementing them into clinical practice."

Mental health stigma can also factor into this data. "Black, American Indian, and Alaska Native patients are less likely to participate in the mental health community than Whites," Early says. "And one of the reasons is because there's stigma."

At the same time, Early offers alternative guesses as to why some did not record their race or ethnicity. "Oftentimes people feel that if I identify, I'm less likely to be treated or to receive adequate treatment," he says.

The Models Need Improvement

Tucker stresses that these models are still important in the clinical setting; they add another tool to patient care.

At the same time, these models need to become more inclusive across racial and ethnic groups. How can healthcare systems, then, implement these necessary tools while making them helpful for everybody, regardless of race or ethnicity?

"We don't think that it should be the responsibility of individual clinicians or individual patients to be concerned with these models," Coley says. "It's the role of health systems who are choosing to use these prediction models to do the evaluation."

For Coley, this study offers a guide to healthcare systems on how to audit their predictive models and make them more equitably applicable. "For the potential benefits of clinical prediction models to be realized in BIPOC populations, there has to be an investment in electronic health record data infrastructure and resources and healthcare systems that serve more racially and ethically diverse populations," Coley says.

Wexler suggests there may be some practical and inexpensive ways to improve the models. "Perhaps a built-in collaboration with people's current support system and in culturally responsive ways—perhaps partnering with families, churches, tribes—that leverage strengths of Black and AI/AN communities," Wexler says.

In addition to reaching out to support systems, to Wexler, predicting and preventing suicide requires engaging community health workers to do culturally responsive health promotion within communities. That could look like working with family members or other parts of someone's support system to ensure they have limited access to firearms, for example. It could also involve working with community leaders from organizations, churches, or tribes, to reduce mental health stigma.

Overall, Early sees the finding as a symptom of a larger, deeply-rooted issue. "It doesn't matter whether or not you're insured as a person of color. You're still less likely to get healthcare," he says. "And even if you have insurance, the quality of healthcare that you have is not going to be as good as it would be if you were not a person of color." To change these embedded patterns and inequalities, Early adds, "What I would like to see changed in America is American culture."

2 Sources
Verywell Health uses only high-quality sources, including peer-reviewed studies, to support the facts within our articles. Read our editorial process to learn more about how we fact-check and keep our content accurate, reliable, and trustworthy.
  1. Coley RY, Johnson E, Simon GE, Cruz M, Shortreed SM. Racial/ethnic disparities in the performance of prediction models for death by suicide after mental health visitsJAMA Psychiatry. 2021;78(7):726-734. doi:10.1001/jamapsychiatry.2021.0493

  2. National Institute of Mental Health (NIMH). Suicide.

By Sarah Simon
Sarah Simon is a bilingual multimedia journalist with a degree in psychology. She has previously written for publications including The Daily Beast and Rantt Media.