Popul Res Policy Rev. 2025;44(3):33. doi: 10.1007/s11113-025-09952-4. Epub 2025 May 13.
ABSTRACT
Working age (25-64) mortality in the US has been increasing for decades, driven in part by rising deaths due to drug overdose, as well as increases in suicide and alcohol-related mortality. These deaths have been hypothesized by some to be due to despair, but this has rarely been empirically tested. For despair to explain mortality due to alcohol-related liver disease, suicide, and drug overdose, it must first predict the behaviors that lead to such causes of death. To that end, we aim to answer two research questions. First, does despair predict the behaviors that are antecedent to the “deaths of despair”? Second, what measures and domains of despair are most important? We use data from over 6000 individuals at five waves of the National Longitudinal Study of Adolescent to Adult Health and apply supervised machine learning to assess the role of despair in predicting self-destructive behaviors associated with these causes of death. Comparing predictive performance within each outcome using measures of despair to benchmark models of clinical and prior behavioral predictors, we evaluate the added predictive value of despair above and beyond established risk factors. We find that despair underperforms compared to clinical risk factors for suicidal ideation and heavy drinking, but over performs compared to clinical risk factors and prior behaviors for illegal drug use and prescription drug misuse. We also compare model performance and feature importance across outcomes; our ability to predict thoughts of suicide, drug abuse and misuse, and heavy drinking differs depending on the behavior, and the relative importance of different indicators of despair varies across outcomes as well. Our findings suggest that the self-destructive behaviors are distinct and the pathways from despair to self-destructive behavior varied. The results draw into question the relevance of despair as a unifying framework for understanding the current crisis in midlife health and mortality.
PMID:40376253 | PMC:PMC12075290 | DOI:10.1007/s11113-025-09952-4
AI-Assisted Evidence Search
Share Evidence Blueprint
Search Google Scholar