Using natural language processing to improve suicide classification requires consideration of race

Autopsy rates and the misclassification of suicide and accident deaths

Mortality statistics are critical to determine the burden of disease. Certain causes of death are prone to being misclassified on cause of death certificates. This poses a serious risk for public health and safety, as accurate death certificates form the basis for mortality statistics, which in turn are crucial for research, funding allocation and health interventions. This study uses generalised estimating equations and regression modelling to investigate for which cause of death categories suicide and accident deaths are misclassified as. National mortality statistics and autopsy rates from North America and Europe covering the past forty years were analysed to determine the associations between the different causes of death in cross-sectional and longitudinal models. We find that suicides and deaths by accidents are frequently mutually misclassified. We also find that suicides are frequently misclassified as drug use disorder deaths, in contrast to accident deaths, which are not misclassified as drug use disorder deaths. Furthermore, suicides do not seem to be misclassified as undetermined deaths or ill-defined deaths. The frequency of misclassification shows that the quality of death certificates should be improved, and autopsies may be used systematically to control the quality of death certificates.

Examining the Etiology of Asian American Suicide in the United States

Research highlights racial and ethnic disparities in suicide, but Asian American suicide receives very little attention in the literature. This is the first comprehensive, large-scale, nationally representative study of completed suicide among Asian Americans in the United States. Descriptive and multilevel regression techniques compared the risk factors for completed suicide across 227,786 Asian American, White, African American, Hispanic, and American Indian suicide decedents from 2003 to 2019. Results indicated that Asian American suicide decedents were significantly less likely than their counterparts to have several risk factors for suicide. Asian Americans were less likely to be male, uneducated, and unmarried. Asian Americans were less likely to use alcohol and drugs, to have mental health problems, and to die by firearm, relative to other suicide methods. Asian Americans were less likely to have a history of prior suicide attempts, to have intimate partner problems, and to have criminal legal problems. Conversely, Asian Americans were more likely to reside in places with higher levels of concentrated disadvantage, residential instability, racial and ethnic heterogeneity, and population density. The results underscore the need for race-specific suicide prevention strategies that, for Asian Americans in particular, take into account cultural values and barriers to help-seeking behavior.

Artificial Intelligence and the National Violent Death Reporting System: A Rapid Review

As the awareness on violent deaths from guns, drugs, and suicides emerges as a public health crisis in the United States, attempts to prevent injury and mortality through nursing research are critical. The National Violent Death Reporting System provides public health surveillance of US violent deaths; however, understanding the National Violent Death Reporting System's research utility is limited. The purpose of our rapid review of the 2019-2023 literature was to understand to what extent artificial intelligence methods are being used with the National Violent Death Reporting System. We identified 16 National Violent Death Reporting System artificial intelligence studies, with more than half published after 2020. The text-rich content of National Violent Death Reporting System enabled researchers to center their artificial intelligence approaches mostly on natural language processing (50%) or natural language processing and machine learning (37%). Significant heterogeneity in approaches, techniques, and processes was noted across the studies, with critical methods information often lacking. The aims and focus of National Violent Death Reporting System studies were homogeneous and mostly examined suicide among nurses and older adults. Our findings suggested that artificial intelligence is a promising approach to the National Violent Death Reporting System data with significant untapped potential in its use. Artificial intelligence may prove to be a powerful tool enabling nursing scholars and practitioners to reduce the number of preventable, violent deaths.

Family as a Pathway to Suicidal Behaviors Through Depression Symptoms and Internalized Homophobia

Research consistently highlights how systemic and social factors can adversely impact mental health, and the potential buffering effects of family support, yet raced sexual minorities are vastly underrepresented among these studies. As rates of suicide increase among Black people and remain high among men and those in gender and sexually diverse communities, this study sought to examine to relationships between family dynamics and suicidality among young Black men who have sex with men (MSM) in young adulthood. We used an online survey to conduct a logistic regression to examine family factors (family support, open family communication, other adult support, and other adult value), depression symptoms, and internalized homophobia on suicide attempts. The conceptualization of the study's design and interpretation of the results were informed by minority stress theory and the phenomenological variant of ecological systems theory. The results indicate that higher levels of family support and open family communication were associated with lower levels of suicidality. Implications for future research and applications for healthcare providers and human services professionals who support young Black MSM in emerging adulthood are discussed.

Machine learning data sources in pediatric sleep research: assessing racial/ethnic differences in electronic health record-based clinical notes prior to model training

Introduction

Pediatric sleep problems can be detected across racial/ethnic subpopulations in primary care settings. However, the electronic health record (EHR) data documentation that describes patients' sleep problems may be inherently biased due to both historical biases and informed presence. This study assessed racial/ethnic differences in natural language processing (NLP) training data (e.g., pediatric sleep-related keywords in primary care clinical notes) prior to model training.

Methods

We used a predefined keyword features set containing 178 Peds B-SATED keywords. We then queried all the clinical notes from patients seen in pediatric primary care between the ages of 5 and 18 from January 2018 to December 2021. A least absolute shrinkage and selection operator (LASSO) regression model was used to investigate whether there were racial/ethnic differences in the documentation of Peds B-SATED keywords. Then, mixed-effects logistic regression was used to determine whether the odds of the presence of global Peds B-SATED dimensions also differed across racial/ethnic subpopulations.

Results

Using both LASSO and multilevel modeling approaches, the current study found that there were racial/ethnic differences in providers' documentation of Peds B-SATED keywords and global dimensions. In addition, the most frequently documented Peds B-SATED keyword rankings qualitatively differed across racial/ethnic subpopulations.

Conclusion

This study revealed providers' differential patterns of documenting Peds B-SATED keywords and global dimensions that may account for the under-detection of pediatric sleep problems among racial/ethnic subpopulations. In research, these findings have important implications for the equitable clinical documentation of sleep problems in pediatric primary care settings and extend prior retrospective work in pediatric sleep specialty settings.

Differences in adverse outcomes across race and ethnicity among Veterans with similar predicted risks of an overdose or suicide-related event

OBJECTIVE

To evaluate the degree to which differences in incidence of mortality and serious adverse events exist across patient race and ethnicity among Veterans Health Administration (VHA) patients receiving outpatient opioid prescriptions and who have similar predicted risks of adverse outcomes. Patients were assigned scores via the VHA Stratification Tool for Opioid Risk Mitigation (STORM), a model used to predict the risk of experiencing overdose- or suicide-related health care events or death. Individuals with the highest STORM risk scores are targeted for case review.

DESIGN

Retrospective cohort study of high-risk veterans who received an outpatient prescription opioid between 4/2018-3/2019.

SETTING

All VHA medical centers.

PARTICIPANTS

In total, 84 473 patients whose estimated risk scores were between 0.0420 and 0.0609, the risk scores associated with the top 5%-10% of risk in the STORM development sample.

METHODS

We examined the expected probability of mortality and serious adverse events (SAEs; overdose or suicide-related events) given a patient's risk score and race.

RESULTS

Given a similar risk score, Black patients were less likely than White patients to have a recorded SAE within 6 months of risk score calculation. Black, Hispanic, and Asian patients were less likely than White patients with similar risk scores to die within 6 months of risk score calculation. Some of the mortality differences were driven by age differences in the composition of racial and ethnic groups in our sample.

CONCLUSIONS

Our results suggest that relying on the STORM model to identify patients who may benefit from an interdisciplinary case review may identify patients with clinically meaningful differences in outcome risk across race and ethnicity.

Factors associated with suicide/self-inflicted injuries among women aged 18-65 years in the United States: A 13-year retrospective analysis of the National Inpatient Sample database

BACKGROUND

Suicide is a significant cause of mortality in the United States, accounting for 14.5 deaths/100,000. Although there are data on gender disparity in suicide/self-inflicted injury rates in the United States, few studies have examined the factors associated with suicide/self-inflicted injury in females.

OBJECTIVE

To determine factors associated with suicide/self-inflicted injuries among women aged 18-65 years in the United States.

METHODS

Hospitalizations for suicide or self-inflicted injuries were identified using the National Inpatient Sample database from 2003-2015 using sample weights to generate national estimates. Independent predictors of suicide/self-inflicted injuries were identified using multivariable regression models. Interaction term analysis to identify the interaction between race/ethnicity and income were conducted.

RESULTS

There were 1,031,693 adult women hospitalizations in the U.S. with a primary diagnosis of suicide/self-inflicted injury in the study period. The highest suicide/self-inflicted injury risk was among women aged 31-45years (OR = 1.23, CI = 1.19-1.27, p < 0.05). Blacks in the highest income strata had a 20% increase in the odds of suicide/self-inflicted injury compared to Whites in the lowest socioeconomic strata (OR = 1.20, CI = 1.05-1.37, p <0.05). Intimate partner violence increased suicide/self-inflicted injury risk 6-fold (OR = 5.77, CI = 5.01-6.65, p < 0.05).

CONCLUSION

Suicide risk is among women aged 31-45 years, higher earning Black women, intimate partner violence victims, uninsured, and current smokers. Interventions and policies that reduce smoking, prevents intimate partner violence, addresses racial discrimination and bias, and provides universal health coverage are needed to prevent excess mortality from suicide deaths.

Identifying latent themes in suicide among black and white adolescents and young adults using the National Violent Death Reporting System, 2013-2019

Suicide rates for adolescents and young adults (AYA) have risen dramatically in recent years - by almost 60% for Americans aged 10-24 years between 2007 and 2018. This increase has occurred for both whites and Blacks, with the rise in suicide among Black youth of particular note. Blacks historically exhibit lower rates of suicide relative to whites and thus, less is known about the etiology of Black suicide. To gain insight into the underlying causes of suicide among AYA, we examine medical examiner reports from the National Violent Death Reporting System (NVDRS) from 2013 to 2019 for over 26,000 Black and white suicide decedents ages 10-29. We apply structural topic modeling (STM) approaches to describe the broad contours of AYA suicide in the United States today. Our findings reveal distinct patterns by race. Guns, violence and the criminal justice system are prominent features of Black suicide, whether through the mechanism used in the suicide, either by firearm or other violent means such as fire or electrocution, the existence of criminal or legal problems/disputes, the location of death in a jail, or the presence of police. In contrast, the narratives of white AYA are more likely to reference mental health or substance abuse problems. Access to resources, as measured by county median household income, overlay these patterns. Themes more prevalent among Blacks are more common in poorer counties; those more prevalent among whites tend to be more common in wealthier counties. Our findings are consistent with other studies that suggest Black people experience greater exposure to violence and other traumas, systemic racism and interpersonal discrimination that may elevate the risk for suicidal behavior.

Characterizing Female Firearm Suicide Circumstances: A Natural Language Processing and Machine Learning Approach

INTRODUCTION

Since 2005, female firearm suicide rates increased by 34%, outpacing the rise in male firearm suicide rates over the same period. The objective of this study was to develop and evaluate a natural language processing pipeline to identify a select set of common and important circumstances preceding female firearm suicide from coroner/medical examiner and law enforcement narratives.

METHODS

Unstructured information from coroner/medical examiner and law enforcement narratives were manually coded for 1,462 randomly selected cases from the National Violent Death Reporting System. Decedents were included from 40 states and Puerto Rico from 2014 to 2018. Naive Bayes, Random Forest, Support Vector Machine, and Gradient Boosting classifier models were tuned using 5-fold cross-validation. Model performance was assessed using sensitivity, specificity, positive predictive value, F1, and other metrics. Analyses were conducted from February to November 2022.

RESULTS

The natural language processing pipeline performed well in identifying recent interpersonal disputes, problems with intimate partners, acute/chronic pain, and intimate partners and immediate family at the scene. For example, the Support Vector Machine model had a mean of 98.1% specificity and 90.5% positive predictive value in classifying a recent interpersonal dispute before suicide. The Gradient Boosting model had a mean of 98.7% specificity and 93.2% positive predictive value in classifying a recent interpersonal dispute before suicide.

CONCLUSIONS

This study developed a natural language processing pipeline to classify 5 female firearm suicide antecedents using narrative reports from the National Violent Death Reporting System, which may improve the examination of these circumstances. Practitioners and researchers should weigh the efficiency of natural language processing pipeline development against conventional text mining and manual review.

Critical gaps in understanding firearm suicide in Hispanic communities: demographics, mental health, and access to care

Suicide rates increased by 26.7% among Hispanics from 2015 to 2020, driven at least in part by highly lethal firearm suicide deaths. However, there are critical gaps in characterizing firearm suicide risks and prevention opportunities in Hispanic communities. We examined Hispanic adult firearm suicide decedents reported through the National Violent Death Reporting System from 2013-2019, focusing on demographic characteristics, firearm choices, suicidal thoughts/behaviors, mental health, and mental health treatment, compared with non-Hispanic adult firearm suicide decedents. Only 13.8% of Hispanic firearm suicide decedents were known to be undergoing treatment for a mental health or substance use problem prior to death, compared to 18.8% of non-Hispanic firearm suicide decedents. On average, Hispanic firearm suicide decedents were significantly less likely than non-Hispanic firearm suicide decedents known to have been treated for a mental health or substance use problem. These results may underscore the critical need for public health agencies and policymakers to promote initiatives integrating mental health screening into medical care, reducing mental health stigma among Hispanics, and expanding mental health treatment capacity in Hispanic communities.

The use of advanced technology and statistical methods to predict and prevent suicide

In the past decade, two themes have emerged across suicide research. First, according to meta-analyses, the ability to predict and prevent suicidal thoughts and behaviours is weaker than would be expected for the size of the field. Second, review and commentary papers propose that technological and statistical methods (such as smartphones, wearables, digital phenotyping and machine learning) might become solutions to this problem. In this Review, we aim to strike a balance between the pessimistic picture presented by these meta-analyses and the optimistic picture presented by review and commentary papers about the promise of advanced technological and statistical methods to improve the ability to understand, predict and prevent suicide. We divide our discussion into two broad categories. First, we discuss the research aimed at assessment, with the goal of better understanding or more accurately predicting suicidal thoughts and behaviours. Second, we discuss the literature that focuses on prevention of suicidal thoughts and behaviours. Ecological momentary assessment, wearables and other technological and statistical advances hold great promise for predicting and preventing suicide, but there is much yet to do.

Application of Natural Language Processing (NLP) in Detecting and Preventing Suicide Ideation: A Systematic Review

(1) Introduction: Around a million people are reported to die by suicide every year, and due to the stigma associated with the nature of the death, this figure is usually assumed to be an underestimate. Machine learning and artificial intelligence such as natural language processing has the potential to become a major technique for the detection, diagnosis, and treatment of people. (2) Methods: PubMed, EMBASE, MEDLINE, PsycInfo, and Global Health databases were searched for studies that reported use of NLP for suicide ideation or self-harm. (3) Result: The preliminary search of 5 databases generated 387 results. Removal of duplicates resulted in 158 potentially suitable studies. Twenty papers were finally included in this review. (4) Discussion: Studies show that combining structured and unstructured data in NLP data modelling yielded more accurate results than utilizing either alone. Additionally, to reduce suicides, people with mental problems must be continuously and passively monitored. (5) Conclusions: The use of AI&ML opens new avenues for considerably guiding risk prediction and advancing suicide prevention frameworks. The review's analysis of the included research revealed that the use of NLP may result in low-cost and effective alternatives to existing resource-intensive methods of suicide prevention.