Moving suicide prevention upstream by understanding the effect of flourishing on suicidal ideation in midlife: an instrumental variable approach

Prior research has examined the association between flourishing and suicidal ideation, but it is unknown whether this association is causal. Understanding the causality between flourishing and suicidal ideation is important for clinicians and policymakers to determine the value of innovative suicide prevention programs by improving flourishing in at-risk groups. Using a linked nationwide longitudinal sample of 1619 middle-aged adults (mean age 53, 53% female, 88% White) from the National Survey of Midlife Development in the United States (MIDUS), this retrospective cohort study aims to assess the causal relationship between flourishing and suicidal ideation among middle-aged adults in the US. Flourishing is a theory-informed 13-scale index covering three domains: emotional, psychological, and social well-being. Suicidal ideation was self-reported in a follow-up interview conducted after measuring flourishing. We estimated instrumental variable models to examine the potential causal relationship between flourishing and suicidal ideation. High-level flourishing (binary) was reported by 486 (30.0%) individuals, and was associated with an 18.6% reduction in any suicidal ideation (binary) (95% CI, − 29.3– − 8.0). Using alternative measures, a one standard deviation increase in flourishing (z-score) was associated with a 0.518 (95% CI, 0.069, 0.968) standard deviation decrease in suicidal ideation (z-score). Our results suggest that prevention programs that increase flourishing in midlife should result in meaningful reductions in suicide risk. Strengthening population-level collaboration between policymakers, clinical practitioners, and non-medical partners to promote flourishing can support our collective ability to reduce suicide risks across social, economic, and other structural circumstances.

www.nature.com/scientificreports/ overlaps with mental well-being and can contribute to a novel approach to reducing suicidal ideation. Recently, human flourishing has been suggested as a conceptual touchstone for prevention-related priorities and objective endpoints in the 2020s 17 . Previous studies have found flourishing associated with a decreased risk of suicidal ideation 18,19 . However, we are unaware of any studies that have assessed the causal effect of flourishing on suicidal ideation. This study is the first to examine the causal effect of flourishing on suicidal ideation. Our design is strengthened by the ability to construct a linked, nationally representative, longitudinal dataset where flourishing was measured temporally before suicidal ideation. To estimate the causal association between flourishing and suicidal ideation 4,20,21 , we conduct instrumental variables (IV) analyses [22][23][24] . IV is a technique used in health economics, psychiatry, and other areas of medicine to remove parameter bias arising from self-selection, random measurement error, and reverse causation (also called simultaneity bias) when analyzing observational data 25,26 .
IV analysis requires the identification of instruments: exogenous variables that influence the factor of interest (primary exposures) but have no direct influence on the primary outcome, conditional on included covariates.

Constructs
Measures Coding  Table 1. Measures of primary exposure (Flourishing Subscales and Domains), primary outcome and instrumental variables. We used two algorithms to measure flourishing based on the theory. (1) Binary Scoring Algorithm: For each of the 13 scales above (6 psychological + 5 social wellbeing + 2 emotional wellbeing), we computed binary measures, where cut-points were applied to code the upper third of the distribution of each summed scale as one and the lower two-thirds as zero. Flourishing was then indicated when six of the 11 psychological and social well-being scales were equal to one, and at least one of the two emotional wellbeing scales was equal to one. A second binary measure was constructed in the same way, but omitted the 2 emotional well-being scales such that the measure was equal to one when six of the 11 psychological and social well-being scales were equal to one and zero otherwise. (2) Z-score Scoring Algorithm: All 13 scales are summed and converted to a z-score. A second measure also summed the measure, but omitted the 2 emotional well-being items, and was then converted to a z-score. Flourishing was measured temporally prior to suicidal ideation. www.nature.com/scientificreports/ We use two valid instruments to determine the causal effect of flourishing on suicidal ideation. These instruments are chosen based on theoretical considerations. The stress-diathesis theory implies that suicidal ideation is due to distal factors (diathesis) from early life and proximal factors from recent stressors 27 . We thus chose two instruments: (1) adverse childhood experiences (ACEs) as a diathesis factor measuring childhood trauma that has been demonstrated to have epigenetic effects [28][29][30][31] ; and (2) daily discrimination as a stressor factor measuring perceptions of discrimination 32 . Figure 1 shows a directed acyclic graph (DAG) and temporal ordering of primary exposures to outcomes. The key assumption is that both ACEs and daily discrimination are exogenous and should only affect suicidal ideation through their inhibitory impacts on flourishing, conditional on covariates.
Results. Among 1619 participants, 850 (5.5%) were women, and 769 (47.5%) were men (mean age, 53.4, standard deviation (SD) 12.7 years). Table 2 summarizes the baseline characteristics of our study cohort. 202 participants (12.6%) endorsed suicidal ideation. High-level flourishing (binary) was reported by 486 participants (30%), the mean (SD) of daily discrimination was 12.9 (4.6), and the mean (SD) of ACEs was 3.5 (1.9). Standard overidentification tests were performed using the two-stage limited information maximum likelihood (2SLIML) model. Hansen's J statistic (a test of the joint null hypothesis that the instruments are uncorrelated with the error term and correctly excluded from the estimated equation, assuming at least one instrument is valid) showed that the null hypothesis could not be rejected in any 2SLIML specification.
The strength of the set of instruments could only be tested using the 2SLIML model. The Montiel-Olea-Pflueger weak instrument test 33 , a test that is robust to both heteroskedasticity and serial correlation, yielded an effective F-statistic of 15.52, greater than the relevant critical value of 8.38, indicating less than 5% worst-case bias. Hansen's J (χ 2 = 0.03, p = 0.87) failed to reject the hypothesis that no instrument was correlated with the error term, given that the other instrument is valid. Finally, an endogeneity test (χ 2 = 8.69, p = 0.003) rejected the hypothesis that flourishing is exogenous. See Table S1 in the Supplementary.
The effect of achieving the standard binary definition of flourishing using the bivariate probit model is a 18.6% reduction (95% CI: 8.0, 29.3) in binary suicidal ideation ( Table 3, Table S2 in Supplementary). This is not statistically different from the 44.7 percentage point reduction (95% CI: 11.5, 77.8) in binary suicidal ideation when the 2SLIML model is used.
Our sensitivity analyses yielded additional findings. We find that the hedonic aspect, while important conceptually, does not make a difference in terms of reducing binary suicidal ideation when using a binary flourishing variable. This may be a special case only relevant to suicidal ideation that may not apply in other applications. The relevant statistics for the 2SLIML model that uses a binary flourishing variable that omits the hedonic scales (these statistics are not available for bivariate probit models) are as follows: Montiel-Olea-Pflueger Effective F 15.46 > 7.7 critical value; Hansen's J χ 2 = 0.36, p = 0.55; endogeneity χ 2 = 7.91, p = 0.005). See Table S1 in the Supplementary. As shown in Table 3, there is no statistical difference in the results when this is done.
In addition, as shown in Table 3, the continuous version of flourishing yields a similarly-size parameter, whether or not the hedonic aspects are included and whether or not 2SLIML or instrumental variables probit is used as an estimator. The relevant test statistics for the 2SLIML are as follows (this set of statistics is not available for the instrumental probit model  Table 3, the binary version of the flourishing combined with the z-score of suicidal ideation yields similarly sized parameters, whether or not the hedonic aspects are included, with suicidal ideation being Figure 1. Directed Acyclic Graph.The relationship between flourishing, X, and suicidal ideation, Y, can be estimated using our set of conditional instruments (ACEs, daily discrimination), Z, that are valid conditional on included covariates, W. Y (suicidal ideation) was measured after X (flourishing). W includes severe psychological distress, depression, anxiety, chronic pain, a set of five inflammatory markers, substance use, binge drinking, health status, personality factors, health insurance, and socioeconomic factors (age, sex, race/ethnicity, education, marital status, household income adjusted for household size). The function of Z, conditional on W, in this model is to remove the confounding effects of U (which is a set of unobserved confounders in the error term of the regression analysis) regarding the effect of X on Y. In other words, all these variables in W are included in closing off backdoor paths (non-causal paths) between X and Y, or in econometric terms, to remove any correlation between the set of instruments and the error term.  Table 3, the z-score version of flourishing combined with the z-score of suicidal ideation yields similarly sized parameters, whether or not the hedonic aspects are included, with the reduction being approximately half a standard deviation when flourishing increases by one standard deviation:  Table S6 in the Supplementary. All of the above models correct for measurement error and omitted variable bias [34][35][36] .

Discussion
This is the first study to clarify the nature of the association between flourishing and suicidal ideation. Using the IV approach, we corrected for bias in the estimated parameters of flourishing due to omitted variables or measurement error 35 . Reverse causation was ruled out by the temporal ordering of the data (suicidal ideation was measured after flourishing). Our findings demonstrated the negative associations of ACEs and discrimination with flourishing 37-39 , as well as the effect of flourishing on reducing suicidal ideation 18,19 . Results are larger in magnitude but consistent with previous studies, most of which used cross-sectional designs. Flourishing inhibits suicidal ideation both when defined as a threshold to achieve and as a continuous set of measures to improve. Flourishing also need not include hedonic measures, at least in this application.
The first major strength of our investigation is the use of instrumental variables analysis to account for unmeasured confounding variables 40,41 . We determined that ACEs and daily discrimination are negatively correlated with flourishing at approximately the same magnitudes (when z-scores are used), suggesting that stress and diathesis factors damage flourishing similarly. Emerging literature, based on the cross-sectional National Survey of Children's Health, has revealed the negative impact of ACEs on childhood flourishing 42,43 . Similarly, others have shown an association between daily discrimination and flourishing 17 . We extended such dose-response relationships in the context of the U.S. midlife population. Existing positive psychology and human flourishing theories can explain our findings. People who are flourishing will thrive amid adversity by maximizing their potential by changing abilities and limitations 14 , which reduces the risks of suicidal ideation. By contrast, those www.nature.com/scientificreports/ who lack purpose in life (psychological aspect of flourishing) are more likely to consider suicide when exposed to childhood trauma or daily discrimination 14 .
The causal effect of flourishing on reducing suicidal ideation suggests flourishing can serve as a target of suicide prevention, shifting the paradigm from the traditional notion of risk reduction towards a more holistic approach-focusing on wellbeing and social determinants of health (SDoH) including flourishing, especially by enhancing social networks and social connectedness 17 . Our second major contribution is to parse the key constructs of flourishing to those that are modifiable and can reduce suicidal ideation via clinical or populationlevel interventions. Using multiple flourishing measures, we found that the hedonic aspect of flourishing does not add additional magnitude to the measured relationship compared with the situation when we only included the eudaimonic aspects of flourishing. While this may not be the case for other outcomes, this is nevertheless an important issue, as the eudaimonic aspects of flourishing are more modifiable and can be taught through direct interventions (aimed at improving psychological and social well-being measured in our study). Improving the eudaimonic aspects of flourishing may improve the hedonic aspects of flourishing 44 . Evidence-based positive psychology interventions (focusing on systematically promoting mindfulness, patient-caregiver dyadic interpersonal interactions, and coping) have been associated with increased positive affect and reduced depression and mortality in medical populations 10,12 .
Flourishing-based suicide prevention can be useful and effectively implemented in both clinical care and population-based settings 12,12 . Within psychiatry, screening for flourishing may be a promising way to detect individuals susceptible to childhood trauma or stress that could end in suicide. Treating patients using the lens of flourishing may help clinicians provide better patient-centered care that is more holistic and more acceptable to patients 42,45 . Flourishing-centered suicide prevention programs may be especially relevant to suicide attempts encountered in the emergency department setting given the high suicide rates of 178 per 100 000 person-years in the first three months after discharge and the lack of care coordination with the warm handoff 46,47 . Family-focused interventions, as promoted by the Institute of Medicine, may be pivotal to promoting flourishing by supporting family connections 37,48 . Equally important is to promote flourishing among physicians and healthcare workers who may encounter patients with suicidal ideation, as improving the personal growth and environmental mastery of physicians and healthcare workers may reduce their risks of burnout 49 .
At the population level, flourishing-based suicide prevention could be achieved effectively with evidence-based programs and policy efforts across the sectors of health care, education, and human services 17 . On the one hand, it is key to implement a collaborative and holistic approach to address structural factors (e.g., policies that improve education, income, reduce racism and other forms of discrimination, etc.), create safe and connected families, and build connected communities as a foundation for intergenerational flourishing. On the other hand, our findings highlight that the broad dissemination of individual-oriented "positive psychology" interventions (e.g., strength-based asset development, emotional regulations, positive interpersonal skills) 12,13 can reduce suicide via a more effective public health approach. These individually-oriented interventions can amplify the positive effects of other policies focused on structural change. Our findings emphasize the importance of flourishing in reducing suicidal ideation as an urgent consideration with potential short-term and longer-term benefits.

Limitations. Our study had limitations. First, studies of attrition and retention in MIDUS have revealed that
White, female, married people with higher education and better health were more likely to stay in subsequent waves 50 . Given the possible disproportionate exposure to ACEs and discrimination among vulnerable populations (e.g., racial/ethnic minorities, males), this attrition bias suggests that our findings may be understated. However, the recently NIA-funded MIDAS Retention Early Warning project reinstated a substantial portion of dropouts and provided valuable opportunities to investigate the extent to which flourishing among those who dropped out would cause reduced suicidal ideation 43 . Second, the measure of suicidal ideation was self-reported, which may underestimate suicide risks. In addition, we only used a single question to measure suicidal ideation. Suicidal ideation is a multifaceted issue, and there are full instruments developed for this single issue, but to our knowledge, such instruments are not available in data that also contains measures of flourishing. Third, our operationalization of flourishing is relatively narrow, whereas alternative measures of flourishing cover a broader array of life domains. Future studies are encouraged to validate the findings in this study using a more diverse sample and other measures of flourishing. Lastly, although we included all common suicide risks in our data, results may be biased if some risk factors for suicide that correlate with our instruments are unmeasured.
In sum, risk-reduction suicide interventions have focused on teaching people how not to die, whereas flourishing-focused suicide prevention teaches us how to live. Our evidence showing the possible causal effect of promoting flourishing on reducing suicidal ideation provides a promising future direction for a more effective and scalable approach.

Methods
Data and participants. We obtained data from four samples of the National Survey of Midlife Development in the United States (MIDUS) study that allowed us to construct two longitudinal cohorts, which we then combined 51 . The first is the MIDUS 2 Project (2004-2006 52 , which contains a 10-year follow-up of participants from the original MIDUS study. The second is the MIDUS 2 Biomarker Project (2004-2009) 53 , a longitudinal follow-up of MIDUS 2, containing biological assessments and information on suicidal ideation. The third is the MIDUS Refresher Study (2011-2014) 54 , designed to replenish the MIDUS cohort. The fourth is from the MIDUS Refresher Biomarker study (2012)(2013)(2014)(2015)(2016) 55 , which paralleled the MIDUS 2 Biomarker Project and was a longitudinal follow-up of the MIDUS Refresher Study, containing biological assessments and information on suicidal ideation. All participants gave informed consent 56 . The response rates for the four samples were 81.0%, 39.3%, 73.0%, and 41.5%, respectively. Following previous research 40, 57 , the current study linked these Measures. The primary outcome, primary exposure, and instrumental variable described below have a clear temporal ordering (Fig. 1). Suicidal ideation is temporally preceded by flourishing, which is temporally preceded by the set of instrumental variables (all of which look back to exogenous events from the past).
Primary outcome: suicidal ideation. Suicidal ideation is a single question from the Mood and Anxiety Symptoms Questionnaire (MASQ) available in both the MIDUS 2 Biomarker and Refresher Biomarker projects 58 , asking the respondent, "during the past week, how much you have felt or experienced thoughts about death or suicide?" Because we want to capture any amount of suicidal ideation, we transformed this question into a binary variable (having "a little bit" up to "extreme" thoughts of suicide) vs. non-suicidal ideation cases (having "no" thoughts of suicide). Binary measures of suicidal ideation are commonly employed when measuring selfreport suicidal ideation using similar measures of MASQ like PHQ-9 59 (e.g., dichotomizing the 4-point scale responses 60,61 ) and have been found to be valid for clinical assessment 41 . For sensitivity analyses, following a recent review regarding the effectiveness of different suicide screening measures 62 , we transformed the original suicidal ideation question into a z-score, which is expressed in standard deviations and avoids any loss of data.
Primary exposure: flourishing. Flourishing is based on a 13-scale index measuring emotional well-being (hedonic aspect), psychological well-being, and social well-being (eudaimonic aspects) ( Table 1)  Psychological well-being includes 6 subscales. Each item is coded on a 7-point scale (1 = strongly disagree to 7 = strongly agree). Items are reverse coded as indicated in Table 1.
Social well-being includes 5 subscales, one with 2 items and the remaining each having 3 items. Each item is scored on a 7-point scale and coded in the same way as the items for psychological well-being. Items are reverse coded as indicated in Table 1.
Keyes' binary measure of flourishing (high-level flourishing) is constructed using the following algorithm: for each of the 13 scales described above, we computed binary measures, where cut-points were applied to code the upper third of the distribution of each scale as one, and the lower two-thirds as zero. Flourishing was then indicated when six of the 11 psychological and social well-being scales were equal to one, and at least one of the two emotional well-being scales was equal to one 16 .
To facilitate sensitivity analyses, we also constructed additional flourishing measures. These alternative measures are designed to determine whether outcomes are sensitive to the inclusion/exclusion of the hedonic aspect of flourishing and whether flourishing may be not only be considered as a state (whether an individual is in the upper third of emotional, psychological, and social well-being), but also as a process (improvements in emotional, psychological, and social well-being are valuable and desirable as one moves toward the upper one-third threshold goal). We thus constructed an alternative binary measure that omitted the two emotional well-being (hedonic) scales. We also constructed two continuous measures of flourishing (one that included both the hedonic and eudaimonic scale sets, and one that only included the eudaimonic scale set) by summing all relevant scales and converting the sum to a z-score [15][16][17] . The distribution of each continuous measure is approximately normal. See Fig. 2. These scales are described in Table 1, and the relevant scoring algorithms are described in the footnotes to the table.
Instrumental variables: ACEs and daily discrimination. Exposure to ACEs is a summed score of 8 binary (yes/ no) categories of adverse experiences (physical abuse, physical neglect, emotional abuse, emotional neglect, sexual abuse, parental divorce, parental depression, and alcohol or drug abuse of any parent) occurring before age 18 (range 0-8, Table 1). This measure has been validated and is consistent with empirical applications 29,37,63 and theory 28 . We transformed this scale into a z-score.
Daily discrimination is measured using the validated daily perceived discrimination scale (Table 1) 32,64 . Responses to each question are coded 0-3 (never, rarely, sometimes, often) and then summed. We transformed this scale into a z-score.
Covariates. We controlled for potential pathways between the instrumental variables and suicidal ideation to ensure that the instruments only affect suicidal ideation through their impact on flourishing. We thus include the following known risk factors for suicidal ideation: (1) socioeconomic characteristics, including participants' age, sex, race/ethnicity, marital status, educational level, household income (adjusted for household size), and health insurance status; (2) physical health, including diminished health status (poor health or fair health), binge drinking (number of days with more than five drinks per day in the past 30 days), substance use (ever used the following in the past 12 months either without a doctor's prescription, in larger amounts than prescribed, or for a longer period than prescribed: sedatives, tranquilizers, stimulants, painkillers, antidepressants, inhalants, marijuana/hashish, cocaine/crack, LSD/hallucinogens, heroin), any chronic pain (yes/no), and a set of 5 inflammatory markers, which have been found to be affected by ACEs and could impact suicidal ideation 40

Statistical analysis. Main analysis.
Our main analysis includes three instrumental variables models: a two-stage limited information maximum likelihood (2SLIML) model, a bivariate probit model that is a recursive probit model [71][72][73] and an IV probit model that is a control function 35 . All instrumental variables models must satisfy three criteria: instruments must be exogenous, strongly correlate with the endogenous variable of interest, and not correlate with the error term 24 . Both instruments were theoretically exogenous: ACEs occurred in childhood, and daily discrimination occurred due to the actions of third parties. We evaluated whether or not the correlation between the endogenous variable of interests (flourishing) and instruments (ACEs and daily discrimination) is sufficiently strong using the Montiel-Olea-Pfluegar weak instrument test 33 . To avoid any potential correlations between the instruments and the error term, we included the known risk factors for suicidal ideation, including socioeconomic characteristics (e.g., marital status, household income), physical health (e.g., chronic pain, inflammation), psychological factors, and social relationships, which are listed in detail above 6,47,74,75 . We additionally evaluated whether the instruments were exogenous using an overidentification test and whether the flourishing parameter was significantly different after being corrected (endogeneity test). Figure 1 presents a directed acyclic graph (DAG) of the model 76,77 . All models are estimated using Stata 16 33,78 .
Sensitivity analysis. We performed three sets of sensitivity analyses. We used binary measures of flourishing that omitted emotional well-being (measures included eudaimonic aspects only). We used continuous versions of flourishing (z-scores) for both the eudaimonic-hedonic measure of flourishing and the eudaimonic-only measure of flourishing. Finally, we estimated every model substituting the z-score of suicidal ideation for the binary version. Thus, we estimated all combinations of measures of suicidal outcomes (binary, z-score) and measures of flourishing (binary eudaimonic-hedonic, binary eudaimonic only, continuous eudaimonic-hedonic, continuous eudaimonic only). Binary suicidal ideation-binary flourishing models were estimated using 2SLIML and bivariate probit; binary suicidal ideation-continuous flourishing models were estimated using 2SLIML and IV probit (control function probit); continuous suicidal ideation-binary flourishing models were estimated using 2SLIML, and continuous suicidal ideation-continuous flourishing models were estimated using 2SLIML.
Ethical approval. This study was not required to obtain ethics approval since it uses publicly available data that contains no identifiable private information and is therefore not considered human subjects research by the Committee for the Protection of Human Subjects at the University of California. The authors did not have access to any personally identifiable information or information that would link the data to individuals' identities. All data are reported in aggregate to eliminate the possibility of deductive identification of individuals.

Data availability
Data sharing: Data are accessible through Inter-university Consortium for Political and Social Research (ICPSR). The public-use data files in this collection are available for access by the general public through https:// www. icpsr. umich. edu/ web/ ICPSR/ series/ 203. Access does not require affiliation with an ICPSR member institution. All Stata code is accessible from the corresponding author.