Abstract
The trend of increasing U.S. working-age (25–64) mortality is well-documented. Yet, our understanding of its causes is incomplete, and analyses are often limited to using population data with little information on individual behaviors and characteristics. One characterization of this trend centers on the role of despair as a catalyst for self-destructive behaviors that ultimately manifest in mortality from suicide and substance use. The role of despair in predicting mortality at the individual level has received limited empirical interrogation. Using Cox proportional hazards models with behavioral risk factors and latent variable measures of despair in young adulthood (ages 24–32 in 2008–2009) as focal predictors, we estimate subsequent mortality risk through 2022 (298 deaths among 12,277 individuals; 177,628 person-years of exposure). We find that suicidal ideation, suicide attempts, illegal drug use, and prescription drug abuse in young adulthood predict all-cause, suicide, and drug poisoning mortality. Notably, all four domains of despair (cognitive, emotional, biosomatic, and behavioral) and overall despair predict all-cause mortality and mortality from drug poisoning and suicide. This research note provides new empirical evidence regarding the relationship between individual despair and mortality, improving our understanding of the life course contributors to working-age mortality.
Background
Working-age (25–64) mortality in the United States rose from 1980 onward relative to that of its international peers (Masters et al. 2022; Woolf 2023), with particularly steep increases from 2000 up to, and including, the COVID-19 pandemic serving as a troubling indication that population health is trending in the wrong direction (National Academies of Sciences, Engineering, and Medicine [NASEM] 2021). This concern was underscored by declines in life expectancy at birth before the onset of the COVID-19 pandemic (Arias et al. 2022; Woolf and Schoomaker 2019). Demographers have been instrumental in documenting and describing these mortality trends at the population level, with particular emphasis on the key age- and cause-specific contributions to these trends (Dowd et al. 2024; Ho and Preston 2010; Masters et al. 2018; Mehta et al. 2020; Trias-Llimós and Permanyer 2023). However, for all the attention on worsening health and increasing mortality at younger ages, investigation of the life course determinants of working-age mortality at the individual level has been more limited.
Indeed, increases in working-age mortality have been driven partly by increases in specific causes of death often stemming from harmful individual behaviors, including suicide, drug overdose, and alcohol-related liver disease (NASEM 2021). Case and Deaton (2020) adopted the “deaths of despair” label in referring to this cluster of causes of death, in keeping with their hypothesis that macroeconomic and sociocultural changes have eroded the life chances of non-Hispanic White adults with low educational attainment, who suffer from a resulting lack of hope and cope by turning to self-destructive behaviors, such as self-harm and substance use. Although other explanations for rising midlife mortality have been proposed and investigated (Ho 2022; Masters et al. 2018; Mehta et al. 2020) and mortality increases have been observed among other demographic groups (Woolf et al. 2018), the despair narrative remains compelling and continues to garner significant academic and popular attention (Beseran et al. 2022; Harper et al. 2021).
Despite its draw, despair as a construct and a determinant of mortality has received limited formal research attention, partly owing to a lack of conceptual and empirical clarity regarding its definition and measurement. Building on the conceptual model defining and operationalizing despair first outlined by Shanahan et al. (2019), we employ a latent measure of despair developed and validated by Gutin et al. (2023). The conceptual framework describes four plausible domains underlying a broader measure of overall despair, corresponding with emotional, cognitive, biosomatic, and behavioral dimensions and corresponding measures. These measures collectively tap into the experience and embodiment of despair as a negative emotional state, a cognitive state of hopelessness, an accumulation of various physical and somatic stressors, and a propensity for risky and reckless behavior (Shanahan et al. 2019). Gutin et al. (2023) used structural equation modeling with latent variables to identify a well-fitting empirical model consistent with this conceptual definition and found confirmatory evidence for an overarching measure of despair for which these four domains are indicators. Whereas Gutin et al. were limited in using these measures of despair to predict subsequent despair-related behaviors, we extend this analysis by using the factor scores derived from these latent models of despair to examine mortality as a more direct test of the deaths of despair hypothesis.
Evidence regarding the role of despair, specifically, in working-age mortality can be organized into three categories: (1) studies that correlated ecological measures of despair with geographic measures of mortality rates; (2) research that linked individual despair to individual behaviors; (3) and a limited set of studies that directly examined the association between despair and mortality at an individual level. First, studies found evidence that places where reported levels of hopelessness are high (Graham and Pinto 2019) or levels of White status threat are high (Siddiqi et al. 2019) also had higher mortality. Second, at the level of behaviors, cognitive despair predicts self-destructive behaviors, such as illicit drug and opioid use (Copeland et al. 2020); overall despair is predictive of suicidal ideation (Gutin et al. 2023). This line of investigation used behavioral measures, such as substance use, as proximate determinants of the relevant cause-specific mortality but did not test the relationship between individual behaviors and subsequent mortality.
Third, to our knowledge, only three recent studies directly tested whether despair or proxies of despair predict mortality at the individual level. Our investigation contributes most directly to this body of literature, which we believe is critical to evaluating the deaths of despair hypothesis. Using data from the Midlife in the United States study, Song and colleagues (2023) used measures of eudaimonic and hedonic well-being as proxies for despair, finding that well-being predicts suicide, drug-related, and alcohol-related mortality more strongly than cardiovascular mortality. Glei et al. (2024) also drew on these data to find that labor force detachment and lower social integration better differentiate drug, alcohol, and suicide mortality from other causes than psychological or economic distress. Using data from the National Health Interview Survey, Zheng and Choi (2024) used psychological distress as a proxy for despair and found that it predicts all-cause mortality. However, all three studies considered large age ranges—some well outside working ages or midlife—thus challenging our understanding of the complex age-specific mechanisms at play.
Our analysis contributes to and advances these findings by using a multidimensional measure of despair to predict all-cause and cause-specific mortality rather than relying on proxies that reflect only psychological well-being and by focusing on an emerging cohort of midlife adults to better assess the predictive power of despair for mortality at younger ages. These proxy measures are valuable, and extant research has identified key elements of despair at play. However, a unified way to measure despair is lacking. Our holistic, multidimensional approach underscores the many and complementary—rather than competing—ways individuals experience despair across biological, psychological, and social domains that prove harmful to their health.
Methods
Data
Data for this analysis come from the National Longitudinal Study of Adolescent to Adult Health (Add Health), a nationally representative sample of U.S. adolescents (ages 12–19) first interviewed in 1994–1995 (Wave I) and followed through four additional waves of data collection (Harris et al. 2019). Our measures come from in-home survey data collected at Waves I and IV (ages 24–32; 2008–2009) and mortality surveillance through 2022. Data on mortality come from Mortality Outcomes Surveillance, which provides an accounting of the vital status of all Add Health sample members from Wave I (Lawrence et al. 2024). Add Health mortality surveillance uses multiple data sources and tracing methods to ascertain respondents’ vital status, ranging from simple address tracing to intensive field tracing with multiple visits to examine possible leads (Trani et al. 2024). Sample members identified as deceased or unlocatable were subject to screening by an internal Medical Outcomes Coordinator, after which possibly deceased sample members were matched to the National Death Index data. Using an internal scoring algorithm, only respondents matched 1:1 were formally identified as decedents. Detailed cause-of-death data are available for all decedents through December 31, 2022 (Trani et al. 2024). We restrict our analytic sample to Wave IV participants and their subsequent survival/mortality through 2022. After accounting for the complex survey weights (Chen and Harris 2020), we obtain final analytic sample sizes of 12,271 for models including sociodemographic and contextual factors and 12,120 for models including behavioral risk factors.
Measures
Our primary measures of interest include sociodemographic, contextual, and behavioral risk factor measures that we anticipate being important predictors of despair mortality on the basis of extant research. Specifically, our sociodemographic indicators include sex at birth (female or male, based on Wave I); race and ethnicity (non-Hispanic White; non-Hispanic Black; Hispanic; non-Hispanic Asian; non-Hispanic American Indian or Pacific Islander; or non-Hispanic other; based on Wave I); and educational attainment, defined as having received a bachelors’ degree by Wave IV. Our contextual variables include respondents’ region of residence at Wave IV (Northeast, Midwest, West, or South) and a categorized measure of rural–urban commuting area (RUCA) codes, including metropolitan, micropolitan, and rural or small-town communities. Finally, our behavioral risk factors at Wave IV include suicidal ideation in the past year, suicide attempts in the past year, ever misusing prescription drugs, and ever using illegal drugs, all dichotomized according to respondents’ indicating any such activity. The reported frequency of consuming five or more alcoholic beverages in one sitting over the past year was included as a continuous measure.
The remaining key predictor of interest is latent despair, drawing on the multidimensional measures proposed and validated by Gutin et al. (2023). In brief, this measure uses 40 variables at Wave IV reflecting emotional (e.g., depressive symptoms, mental health conditions), cognitive (e.g., pessimism, mastery), biosomatic (e.g., biological dysfunction, sleep issues), and behavioral (e.g., isolation, antisocial behaviors) dimensions of despair. The behavioral dimension is distinct from the behavioral risk factors noted previously (i.e., suicidal ideation and substance use); it captures behavioral attributes that might precede the onset of such harmful behaviors and causes of death (see Gutin et al. 2023; Shanahan et al. 2019). We include these measures both separately by domain and as an overall latent measure of despair.
Finally, we use the detailed cause-of-death data (based on ICD-10 codes) to define broader cause-of-death categories consistent with the deaths of despair framework. In addition to all-cause mortality (n = 298 deaths since Wave IV with valid weights), we focus on drug poisoning (n = 55) and suicide (n = 27) mortality. The former is based on combining deaths from “drug poisoning” (X42) and “drug poisoning, unspecified” (X44); the latter combines deaths from “self-harm” (X60–X69/X71/X73/X75–X84), “self-harm, strangulation” (X70), “self-harm, handgun” (X72), and “self-harm, unspecified firearm” (X74). Although alcohol-related mortality is considered despair-related, it was not available in Add Health. In additional analyses, we also examine cardiovascular, cancer, and transport-related mortality.
Analysis
We use Cox proportional hazards models to estimate the relative mortality risks (i.e., hazard ratios) associated with the predictors in our models, with attained age as the underlying metric of time based on individuals’ age at Wave IV interview and age at censoring, defined here as the date of death or the end of surveillance in 2022 (Thiébaut and Bénichou 2004). The Breslow method is used to handle any ties among survival times. We first look at all-cause and cause-specific mortality models that include only sociodemographic and contextual measures and subsequently add behavioral risk factors. We then run models examining the individual dimensions of despair and overall despair, controlling for the sociodemographic and contextual measures. Finally, we estimate models focusing on overall despair, net of the sociodemographic, contextual, and behavioral indicators. The resulting models cover 177,628 person-years of analysis time at risk.
In addition to checking that the proportional hazards assumption in our models is upheld, we conduct several sensitivity analyses: (1) using a measure of polydrug use that accounts for individuals’ propensities to use multiple illegal drugs; (2) examining other causes of death as a counterfactual test of (or an addendum to) the despair framework; and (3) running competing risks models that explicitly account for censoring due to death from other causes.
Results
The distribution of sociodemographic variables in Table 1 demonstrates the national representativeness of this sample: 51% of respondents are female, 67% non-Hispanic White, 15% non-Hispanic Black, 12% Hispanic, 4% non-Hispanic Asian, and 1% are best represented by a different racial and ethnic category. Roughly one third (32%) of respondents have a four-year college degree. With respect to contextual measures, most respondents are from the South (41%), followed by the Midwest (29%), West (18%), and Northeast (12%). Most respondents reside in metropolitan areas (81%), with approximately 1 in 10 in micropolitan or rural/small-town areas. Among the behavioral risk factors of interest, 7% of respondents reported suicidal ideation in the past year, and 2.5% reported having attempted suicide. Approximately 1 in 5 and 3 in 10 respondents reported misusing prescription drugs or using illegal drugs, respectively. The mean number of days of heavy drinking in the past year is 1.2, with a standard deviation of 1.5. The derived latent despair variables have standardized factor scores with a mean of 0 and standard deviation of 1.
Table 2 presents the results from a survival analysis with all-cause, drug poisoning, and suicide mortality as the key outcomes, focusing on the sociodemographic, contextual, and behavioral risk predictors. Consistent with extant research, female respondents have a ∼ 25% lower risk of all-cause mortality than males, as well as lower drug poisoning and suicide mortality risks, although the results are not statistically significant. Relative to White adults, Black adults have a ∼ 40% higher mortality risk when we account for behavioral risk factors; Hispanic adults have a ∼ 25% lower mortality risk; American Indian or Pacific Islander adults have nearly double the mortality risk; and Asian adults have less than half the mortality risk. By contrast, suicide mortality is significantly lower among Black adults; drug poisoning mortality is significant at p < .10. Also consistent with extant research, a college degree is associated with significantly lower risks of all-cause mortality (hazard ratio [HR] = ∼0.3) and drug poisoning mortality (HR = ∼0.16). Mortality risk varies little across contextual measures, although some evidence suggests a lower drug poisoning risk in rural areas.
Mortality risk varies considerably across the behavioral risk factors included in the analysis. Suicide attempts are associated with more than a twofold increase in all-cause mortality risk and nearly a fivefold increase in drug poisoning mortality. Suicidal ideation is a significant predictor of suicide mortality, with a fourfold increase in risk (HR = 4.37 [1.15, 16.59]). By contrast, suicide attempts are associated with a lower risk of suicide mortality (HR = 0.07 [0.01, 0.65]). Although this finding contradicts expectations, we provide a possible explanation in the Discussion. Drug use is also a significant predictor of premature mortality. Misuse of prescription drugs is associated with a twofold increase in all-cause mortality risk, a fourfold increase in drug poisoning mortality risk, and a threefold increase in suicide mortality risk. Illegal drug use is associated with a ∼ 40% higher risk of all-cause mortality and approximately a twofold increase in drug poisoning mortality risk. Drinking does not appear to be associated with increased mortality risk across models and outcomes.
In Table 3, we examine the mortality risks associated with both individual domains of despair and overall despair in separate models, accounting for sociodemographic and contextual covariates (not shown for parsimony). The results clearly show a strong and significant association between all measures of despair and all-cause, drug poisoning, and suicide mortality. Although we are not formally comparing the estimated hazard ratios and some of the confidence intervals are quite large, we generally see the largest mortality risks associated with cognitive despair or feelings of hopelessness, pessimism, and a lack of control, especially for drug poisoning (HR = 11.89 [5.30, 26.66]) and suicide (HR = 8.51 [3.10, 23.41]) mortality. By contrast, mortality risks associated with increased behavioral despair are generally lower, although still quite large and significant.
Finally, in Table 4, we estimate models that include all covariates, primarily to examine the robustness of the association between latent despair and mortality risk when we account for potentially proximate or intermediary behavioral risk factors. The association between increased despair and increased mortality risk remains strong and significant even when we account for suicidal behavior and substance use. The associations between despair and all-cause and suicide mortality are attenuated, but there is still a threefold increase in risk for both all-cause and suicide mortality associated with a 1-standard-deviation increase in latent despair. The association between despair and drug poisoning mortality risk is instead larger and is now associated with a 12-fold greater risk. Drug use and attempting suicide remain significant predictors of overall and cause-specific mortality risks, even when we account for despair.
In assessing the robustness of our results, we checked that the proportional hazards assumption underlying these Cox survival models is upheld. Given the number of covariates in our models, the overall test of proportionality was significant. However, we examined plots of the Schoenfield residuals across multiple models and found no clear evidence of a strong temporal effect, suggesting that the models provide valid estimates (although some caution should be taken in interpreting the estimates for suicide mortality because of the small number of deaths). Plots for overall despair from the models that include all covariates are included in Figure A1 (this figure and all tables designated with an “A” appear in the online appendix). We also ran multiple sensitivity analyses, including other causes of death, respecifying drug use, using competing risks models, and accounting for potential health and socioeconomic selection at Wave I. As seen in Table A1, behavioral risk factors are generally not associated with cardiovascular, cancer, or transport deaths, as would be expected given the etiology of these causes of death. The one exception is an elevated risk of cancer mortality associated with suicide attempts, although we hesitate to provide much interpretation because it is a very rare outcome for an already rare predictor. Despair is associated with greater cardiovascular mortality risk, which might reflect the biological and somatic measures included in the despair variable. Despair is also associated with greater transport risk, which might reflect the propensity for risky or reckless behavior that the variable captures. Critically, we find no association between despair and cancer risk, as would be expected given the nature of cancer deaths at these ages.
In the online appendix, we provide estimates from robustness checks respecifying “any” illegal drug use as a count measure (Table A2); reestimates of the results for drug poisoning and suicide mortality using a competing risks model in which respondents are censored if they died from other causes (Table A3); and reestimates of the results for all-cause, drug poisoning, and suicide mortality when accounting for parental education, self-rated health, and depressive symptoms (based on the CESD-5; Perreira et al. 2005) to account for any potential selection in greater despair and higher mortality risk at later ages (Table A4). We find no change in our results.
Discussion
Using nationally representative longitudinal data, we investigated the life course predictors of all-cause and cause-specific mortality—specifically, the role of behavioral risk factors and measures of despair in young adulthood. We demonstrated that suicidal ideation and suicide attempts in the last year and any previous illegal drug use and prescription drug misuse reported in young adulthood were associated with increased risk of all-cause mortality. Suicide attempts and both types of drug use were also associated with an increased risk of drug poisoning mortality, whereas suicidal ideation and prescription drug misuse were associated with an increased risk of suicide mortality. It is also possible that the cause-of-death attribution miscategorized suicide by intentional poisonings as drug poisonings (Rockett et al. 2018), and both might be indicative of an underlying mental health crisis (Powell 2023). This possibility could be reflected in the strong association between previous suicide attempts and drug poisoning mortality. Finally, suicide attempts in the last year reported in young adulthood was associated with a lower risk of suicide mortality in the following decade. We hypothesize that this finding might result from services obtained or intervention following a nonfatal attempt. However, it is difficult to fully understand the nature of this association because of the rarity of both the predictor and the outcome.
We also demonstrated that individual despair in young adulthood across each of the four measured domains and overall was associated with all-cause, drug poisoning, and suicide mortality. Young adults who were experiencing despair were more likely to die in the following decade, often by a very large margin. Notably, this association remains even net of behavioral risk factors in young adulthood, which are likely the more proximate determinants of mortality. Furthermore, the increased mortality risk is the largest for cognitive despair and the smallest for behavioral despair, potentially suggesting that thoughts of hopelessness and pessimism play a particularly salient role in mortality risk. Overall despair is also associated with all-cause, drug poisoning, and suicide mortality, although the magnitude is the largest for drug poisoning. Interestingly, we also observe strong associations between despair and causes of death not traditionally included in the deaths of despair framework, including cardiovascular disease and transport deaths. These findings are similar to those of Glei et al. (2024). Aside from noting how the constituent elements of our despair measures might be associated with these outcomes, we believe that these findings help demonstrate the extent to which self-destructive behaviors can manifest across a variety of outcomes, especially at these younger ages. The harmful coping behaviors hypothesized as stemming from experiences and feelings of despair can easily be implicated in the etiologies of causes of death other than suicide and substance use. For instance, the strong association observed with transport deaths might capture some degree of recklessness in driving or a propensity for impaired driving that accounts for a large proportion of U.S. traffic-related deaths (Centers for Disease Control and Prevention 2024), although more detailed data are needed to examine this association. Likewise, there are valid reasons to link cardiovascular disease mortality to despair, at least from the perspective of chronic stress and its biophysiological consequences, as well as its psychosocial and behavioral correlates (Nudy et al. 2023). However, chronic disease is highly multifactorial, and a better understanding of the etiology of chronic disease mortality at these ages is necessary. A core challenge of using mortality as an outcome is the inability to perfectly capture the sequence of events that resulted in death from a specific cause. However, the availability of individual-level data on precursors, as in this case, provides insight into plausible mechanisms. Results from these additional causes of death suggest that, if anything, our conceptualization of deaths of despair is potentially too limited rather than too broad and nonspecific.
In sum, although the deaths of despair hypothesis has been rightfully challenged in recent demographic and population health research, our analyses suggest that it continues to have merit in explaining suicide and substance use–related mortality among recent cohorts of U.S. adults entering midlife. However, we acknowledge the unresolved challenge of reaching a consensus on the conceptualization, definition, measurement, and operationalization of despair across the many different data sources and samples used to study these population health patterns and trends in recent years. We use a broader, multidimensional measure of despair than past studies and limit the analysis to a specific U.S. cohort of younger adults. However, our results are generally in line with prior findings from individual-level analyses in other samples using one or more proxy measures of despair (Glei et al. 2024; Song et al. 2023; Zheng and Choi 2024). This consistency is reassuring, but repeated findings across multiple studies using similar operationalizations of despair are necessary to fully evaluate the despair hypothesis as it pertains specifically to mortality and how the constituent elements of despair are associated with one another.
Finally, although our findings suggest a strong association between despair and mortality, we cannot interrogate heterogeneity in this relationship by important sociodemographic characteristics given the (thankfully) limited number of observed deaths. Future research would benefit from investigating whether despair's relationship with mortality is limited to those with low educational attainment, non-Hispanic White adults, or those in more rural or geographically isolated areas, as suggested in previous work. Further, this analysis does not identify the underlying causes of individual despair or whether it is related to structural, institutional, and economic conditions. Our analysis provides a rich prospective view of despair and mortality risk among one cohort over the period 2009–2022, but it cannot interrogate period trends over time or whether despair is responsible for the observed increases in population working-age mortality. Indeed, the more cohort-specific experiences of deindustrialization and working-class decline hypothesized to underlie deaths of despair might be less relevant to this younger, post–Great Recession cohort. Consequently, it is important to understand whether cohort differences in the factors underlying despair influence not only how despair is observed and operationalized but also how it is associated with mortality and which aspects or dimensions of despair are most salient. Nevertheless, our research contributes to the understanding of the life course determinants of premature mortality and provides evidence of the role of despair in shaping mortality risk.
Acknowledgments
This research uses data from Add Health, funded by grant P01 HD31921 (Harris) from the Eunice Kennedy Shriver National Institute of Child Health and Human Development, with cooperative funding from 23 other federal agencies and foundations. Add Health is currently directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01 AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Add Health was designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill. This research received support from a Population Research Infrastructure Program award to the Carolina Population Center (P2C HD050924) at The University of North Carolina at Chapel Hill by the Population Dynamics Branch of the Eunice Kennedy Shriver National Institute of Child Health and Human Development.