Introduction

The enigma of human suffering has received considerable attention throughout history. Some have described suffering as useless, signaling that the suffering people experience in life is intrinsically “for nothing” (p. 158)1. Others have suggested that suffering is a necessary part of life, in that it is only through suffering that a person can grow in character and virtue2. While perspectives on the utility of suffering may vary, few would disagree with the observation that suffering is a universal and ineradicable part of the human condition3.

The subjective experience of suffering has been characterized as “an undesired experiential state, of considerable duration or intensity, involving the loss or privation of some perceived good” (p. 2)4,5. Although suffering shares some common ground with other debilitating experiences that are laden with negative affect (e.g., depression, pain, loneliness), suffering can be distinguished from other forms of distress both conceptually and empirically. To illustrate, suffering involves a “felt aversion” toward one’s situation (a desire that the situation not be occurring) that can only be adjudicated as such by the sufferer (p. 24)6, whereas major depressive disorder is a mental health condition that is formally diagnosed by healthcare professionals using standardized diagnostic frameworks. This distinction suggests that suffering may not be co-present with depression symptoms that meet the threshold for clinical diagnosis, as has been documented in previous empirical research7,8. Longitudinal research has also shown that suffering predicts worse subsequent well-being (e.g., life satisfaction, meaning in life) even after adjusting for severity of depression symptoms9,10, indicating that suffering is not simply a descriptive label signifying the severity of an individual’s depression symptoms (for a review related to mental health disorders, see ref. 11). Along similar lines, while pain and suffering frequently co-occur, the two concepts are not equivalent. As one illustration, while all suffering is undesired, some types of pain (as in the case of masochism) may be desired12. Nonequivalence of pain and suffering has also been demonstrated empirically (for a review, see ref. 13). For example, one cross-sectional study with emergency department patients found that 16% of those with a pain score of ≥7/10 did not report any suffering14.

To date, empirical research on suffering has mostly been concentrated within the fields of medicine, nursing, and palliative care9. However, a growing number of studies have expanded beyond clinical populations to focus more on everyday suffering in nonclinical samples (e.g., college-attending young adults, middle-aged employees). This research has shown that even relatively healthy adults who are not dealing with debilitating physical health problems may experience considerable suffering, with potentially disruptive effects on physical, mental, and social dimensions of a person’s life4,7,10. These findings suggest that the public health implications of suffering may be quite substantial15, reinforcing the need to develop an “epidemiology of suffering” that can provide a foundation for a public health agenda to target suffering more explicitly (p. 65)5.

Developing an epidemiology of suffering requires research focused on identifying possible determinants of suffering, which could provide insight into important risk and protective factors that ought to be considered by professionals (e.g., public health practitioners, policymakers) who are working to promote human flourishing. While some empirical research has examined potential predictors of suffering16,17, no prior study has applied a life course approach to explore childhood predictors of suffering in adulthood. Research along these lines would contribute to identifying (1) subpopulations that may be especially likely to benefit from early-life interventions focused on preventing childhood events or conditions known to heighten risk of suffering in adulthood, and (2) suitable targets for nurturing a childhood environment that could have cascading positive effects on the development of internal capacities and external resources a person might draw on to address the suffering they experience during adulthood18,19. Toward these ends, the present study uses the first wave of nationally representative data from 22 countries in the Global Flourishing Study to explore associations of individual characteristics and retrospectively assessed childhood factors with suffering in adulthood.

The concept of social suffering has been used to capture the idea that “subjective components of distress are rooted in social situations and conditioned by cultural circumstance,” suggesting that there are connections between a person’s experience of suffering and the broader socioecological system in which they are embedded (p. 146)20. Applying a socioecological life course perspective, the notion of social suffering can be extended further by recognizing that an individual’s experience of suffering at a particular point in time is shaped by earlier developmentally salient experiences that have unfolded within a multilayered network of influential immediate and extended environments21. The socioecological context of childhood is of particular interest in the present study, given that childhood is the earliest period of postnatal development and exposure to risk (e.g., sexual abuse) and protective (e.g., parental warmth) factors during this phase of life can have powerful downstream effects on well-being in adulthood22,23.

Although there are different ways to carve up the conceptual landscape of a child’s socioecological system, Bronfenbrenner24 presented four theoretical layers (i.e., micro, meso, exo, and macrosystem) and a temporal layer that serves a methodological function in empirical research (i.e., chronosystem). The microsystem is the most proximal and directly influential layer to child development, referring to the child’s immediate environment where they interact with people (e.g., parents, peers, teachers) and engage in activities. The mesosystem recognizes that connections between elements of the microsystem are important (e.g., the interactions between a child’s family and the faith leader of the religious community in which they participate). The exosystem refers to contexts that do not contain the child but may indirectly influence their development via impacts on the microsystem (e.g., a parent’s highly demanding job might limit their availability to play an active role in child rearing). The macrosystem includes broader sociocultural factors (e.g., economic conditions, cultural norms) within society that can shape a child’s developmental experiences (e.g., participation in a religious community may be lower among families living in a highly secular culture). The chronosystem considers the child’s experiences over time, such as pivotal events that take place during their life (e.g., the unexpected death of a parent) or historical events that affect their developmental experiences (e.g., racial differences in educational opportunities among children reared in South Africa during the apartheid regime compared to after the country became a democracy in 1994). While the different layers of a child’s socioecology are interdependent and can mutually influence each other in complex ways, highly proximal layers (e.g., microsystem) are often prioritized by researchers and interventionists because they tend to have more of a direct influence on child development21,25. The present study takes a similar approach by attending to candidate predictors within the microsystem, but we engage with broader layers of the socioecological system to contextualize our findings.

So far, research concerning potential predictors of suffering has mostly focused on individual-level factors among older patient populations (often within clinical settings), such as those with chronic pain, debilitating illnesses, and terminal conditions (for reviews, see refs. 26,27,28), typically using cross-sectional designs in which it is difficult to establish a clear temporal order between the predictors and the outcome of suffering5. For example, in one cross-sectional study with N = 98 Colombian advanced cancer patients receiving palliative care, Krikorian et al.16 used structural equation modeling to explore dimensions of physical, psychological, and spiritual problems, coping strategies, and adjustment problems as predictors of suffering while controlling for a small set of sociodemographic (i.e., age, sex, education level, place of residence), illness- and treatment-related (e.g., type of cancer, time in palliative care treatment), and study-related covariates (e.g., place of research interview). They found evidence suggesting that the physical, psychological, and spiritual dimensions of problems were positively related to suffering (either directly or indirectly through coping strategies or adjustment problems), though none of the covariates showed evidence of association with suffering.

When longitudinal studies have reported on predictors of suffering, the tendency to emphasize individual-level factors has remained dominant. For instance, working with three waves of data from a nonclinical community sample of N = 184 United States adults living with chronic illness, Cowden et al.9 explored covariate-adjusted associations of psychological well-being, anxiety symptoms, and depression symptoms assessed ~1 month into the COVID-19 pandemic (Wave 2) with suffering assessed ~2 months later (Wave 3). The results indicated that worse anxiety and depression symptoms were associated with an increase in subsequent suffering. In the same study, Cowden and colleagues produced a supplementary analysis in which Wave 1 sociodemographic characteristics were examined as predictors of Wave 2 suffering. Those who were unmarried or separated and those living in a household with more than two people were more likely to score in the upper tertile of suffering, though none of the other sociodemographic factors (i.e., age, gender, racial/ethnic status, sexual orientation, religious status, educational attainment, annual household income, geographic region) were associated with suffering. In another three-wave longitudinal study involving a community sample of N = 594 Indonesian adults, Ho et al.10 examined two indices of psychological distress (i.e., anxiety symptoms, depression symptoms) and 10 facets of well-being (e.g., life satisfaction, sense of purpose) in Wave 2 as predictors of subsequent suffering ~1 month later in Wave 3 (controlling for a set of Wave 1 covariates). Higher scores on both indices of psychological distress and lower scores on 7/10 facets of well-being were associated with an increase in subsequent suffering.

While prior research has provided a useful foundation for developing an epidemiology of suffering, there are several gaps in knowledge. First, scholars working within the sphere of social suffering have signaled the importance of considering the social contexts and cultural conditions that intersect with individual experiences of suffering20,29, but empirical research in this area has progressed at a slower pace because individual-level predictors of suffering have typically been prioritized. If we are to develop a more comprehensive picture of the potential determinants of suffering, additional research addressing factors within layers of the broader socioecological system (e.g., microsystem) is needed. Second, prior studies that have reported on predictors of suffering have typically used data situated within a particular phase of human development. Compared to evidence about other forms of distress (e.g., major depressive disorder, physical pain), little is known about the way in which events, experiences, and conditions during one phase of life might influence suffering at a later phase of life. Evidence along these lines could be derived by applying chronosystem designs that attempt to identify the impact of prior life events or experiences on subsequent outcomes24. Such research would resonate with a life course epidemiological approach to suffering, which might be practically useful to healthcare professionals, public health authorities, and policymakers involved in supporting population well-being18. Third, empirical research on suffering has typically focused on population segments (e.g., older adults with medical conditions) using small, nonrepresentative samples from Western, educated, industrialized, rich, and democratic countries10; therefore, the generalizability and transportability of existing research concerning predictors of suffering remains unclear. To build an inclusive and globally representative understanding of risk and protective factors for suffering, nationally representative data from around the world is needed. Research with geographically and culturally diverse populations could also shed light on potential cross-national variation in predictors of suffering, given that the features and dynamics of socioecological systems will be somewhat heterogeneous across countries30.

This preregistered study takes an initial step toward addressing some of these gaps in knowledge by employing a life course epidemiological approach to explore predictors of suffering later in life. Specifically, we use nationally representative cross-sectional data from the first wave of the Global Flourishing Study to examine 13 candidate predictors of suffering in adulthood across 22 countries, ranging from individual characteristics (e.g., birth year, gender) to early-life events or experiences (e.g., child abuse, religious service attendance) and social conditions or influences (e.g., family structure, relationships with parents) within the microsystem of a child’s socioecology. Among the predictors included, we anticipated that some childhood factors would show evidence of meaningful associations with suffering in adulthood across all countries. Due to the diversity of sociocultural contexts that characterize the 22 countries, we expected that there would be some cross-national variation in the pattern of associations between the candidate predictors and suffering in adulthood.

We find evidence suggesting a combination of risk and protective factors during childhood may be associated with suffering in adulthood. Pooled meta-analytic results provide evidence of association with suffering in adulthood for eight of the candidate predictors that were considered, although country-specific results indicate that the pattern of associations is somewhat varied across the countries.

Methods

Methodological details described below have been adapted from VanderWeele et al.31. Further methodological information is available elsewhere32,33,34,35,36,37,38. The Global Flourishing Study was ruled exempt by the Baylor University Institutional Review Board (#1841317-2) because it met the criteria for exemption according to Baylor’s Institutional Review Board guidelines (e.g., minimal risk to participants, adherence to specific federal regulations). Ethical approval for all data collection activities was also obtained from the Institutional Review Board at Gallup Inc. Data collection activities were performed in accordance with relevant ethical regulations, and informed consent was obtained from all participants. All direct common identifiers were removed from the data used in the present study by Gallup Inc.

Study sample

Wave 1 of the Global Flourishing Study included nationally representative samples from 22 geographically and culturally diverse countries: Argentina, Australia, Brazil, Egypt, Germany, Hong Kong (Special Administrative Region of China), India, Indonesia, Israel, Japan, Kenya, Mexico, Nigeria, the Philippines, Poland, South Africa, Spain, Sweden, Tanzania, Türkiye, the United Kingdom, and the United States (N = 202,898). The countries were selected to (1) maximize coverage of the world’s population, (2) ensure geographic, cultural, and religious diversity, and (3) prioritize feasibility in line with existing data collection infrastructure. Data collection was conducted by Gallup Inc. Data for Wave 1 was collected primarily during 2023, although some countries began data collection in 2022 and exact dates of data collection varied to some extent by country38. Plans are in place to collect four additional waves of annual panel data on the participants from 2024–2027. The Global Flourishing Study survey centers on salient aspects of well-being, such as happiness, health, meaning, character, relationships, and financial stability39, along with other sociodemographic, social, economic, political, religious/spiritual, personality, childhood, and community variables. Gallup Inc. translated the Global Flourishing Study survey into multiple languages following the TRAPD (translation, review, adjudication, pretesting, and documentation) model for cross-cultural survey research37. Extensive details about the translation, cognitive interviewing, and piloting testing phases of the Global Flourishing Study can be found elsewhere32,34,37,40.

Sampling design

The precise sampling design that was used to ensure samples were nationally representative varied by country37,38. In most countries, local field partners were guided in implementing a probability-based face-to-face or telephone methodology to recruit panel members. Recruitment involved an intake survey that mainly gathered basic sociodemographics and information for recontacting participants. Shortly following recruitment, participants received invitations to participate in the annual survey via phone or online. Three major sampling frames were used for recruitment in the Global Flourishing Study, namely a probability-based sample, a nonprobability-based sample, or a combination of the two. Post-stratification and nonresponse adjustments were carried out within each country separately using census data or a reliable secondary source. Additional information about the sampling design used in Wave 1 of the Global Flourishing Study is available in Padgett et al.37 and Ritter et al.38.

Measures

Predictor variables

Quality of relationship with mother when growing up was assessed with the question: “Please think about your relationship with your mother when you were growing up. In general, would you say that relationship was very good, somewhat good, somewhat bad, or very bad?” An analogous variable was used for the quality of a person’s relationship with their father when growing up. Both maternal and paternal relationship quality were dichotomized (i.e., very bad/somewhat bad versus very good/somewhat good) in the regression analyses to help address multicollinearity issues. “Does not apply” was treated as a dichotomous control variable for participants who did not have a mother or father due to death or absence. Family structure around age 12 was assessed with: “Were your parents married to each other when you were around 12 years old?” Responses included married, divorced, never married, and one or both had died. Subjective financial status of one’s family around age 12 was measured with: “Which one of these phrases comes closest to your own feelings about your family’s household income when you were growing up, such as when you were around 12 years old?” Responses were lived comfortably, got by, found it difficult, and found it very difficult. Childhood abuse when growing up was assessed with yes/no responses to: “Were you ever physically or sexually abused when you were growing up?” Similarly, participants provided a yes/no response to whether they felt like an outsider in their family when growing up: “When you were growing up, did you feel like an outsider in your family?” Self-rated health when growing up was assessed by: “In general, how was your health when you were growing up? Was it excellent, very good, good, fair, or poor?” Immigration status was assessed with: “Were you born in this country, or not?” Frequency of religious service attendance around age 12 was assessed with: “How often did you attend religious services or worship at a temple, mosque, shrine, church, or other religious building when you were around 12 years old?” Responses were at least once/week, 1–3 times/month, less than once/month, or never. Gender was assessed as male, female, or other. Year of birth (age) was classified as 1998–2005 (18–24 years), 1993–1998 (25–29 years), 1983–1993 (30–39 years), 1973–1983 (40–49 years), 1963–1973 (50–59 years), 1953–1963 (60–69 years), 1943–1953 (70–79 years), and 1943 or earlier (80 years or older).

Religious affiliation at age 12 was also assessed in all countries, but there were considerable cross-country differences in the response categories endorsed by participants because some religious affiliations are only applicable in certain countries and not others. Religious affiliation at age 12 response category options included Christianity, Islam, Hinduism, Buddhism, Judaism, Sikhism, Baha’i, Jainism, Shinto, Taoism, Confucianism, primal/animist/folk religion, Spiritism, Umbanda, Candomblé, and other African-derived religions, Chinese folk/traditional religion, some other religion, or no religion/atheist/agnostic. When more than 5% of a within-country sample endorsed the no religion/atheist/agnostic category, this was used as the reference category in the country-specific analyses; otherwise, the most prominent religious group was used. Additionally, all religious affiliation categories endorsed by less than 3% of a within-country sample were collapsed into a single religious affiliation category. Racial/ethnic identity was assessed in most (18/22) countries, and response categories varied across countries. Country-specific analyses used a binary racial/ethnic identity variable based on whether an individual was in the most prominent group in the sample versus a minority group. Additional details about the measurement of the predictors can be found in the Global Flourishing Study codebook (https://osf.io/cg76b).

Outcome variable

Various measures of suffering are available and have been employed in the empirical literature (for reviews, see refs. 5,11,41,42); they vary in modality (e.g., standardized scales versus pictorial measures) and scope of assessment (e.g., generalized versus domain-specific). Because many existing measures of suffering are not well suited for research outside of clinical contexts (e.g., those that make reference to illness), the Personal Suffering Assessment was developed to provide a generalized measure of suffering that could be used more broadly beyond the clinical context5. In this study, suffering was assessed using an adapted version of the extent of suffering item from the Personal Suffering Assessment5: “To what extent are you suffering? This can be any type of physical or mental suffering” (response options: “Not at all,” “Not very much,” “Some,” and “A lot”).

Variations of this single item have been used over the years, dating back at least two decades8,43,44. The original Personal Suffering Assessment item has been used in prior research, including with samples from several countries that are part of the Global Flourishing Study (e.g., Indonesia, the United States)4,7,9,10. Slight modifications were made to the phrasing and response categories of the original Personal Suffering Assessment item in the Global Flourishing Study survey based on the results of cognitive interviews that were performed during the survey development process to strengthen its cross-cultural equivalence (further details are available elsewhere32,34,37,40). For example, the clause (i.e., “This can be any type of physical or mental suffering”) was added to provide clarification about the meaning of the term suffering. Some previous work has reported large to very large correlations between responses to variations of the original Personal Suffering Assessment item and scores on alternative brief measures of suffering that have been widely used, such as versions of the visual-based Pictorial Representation of Illness and Self Measure45,46. Following previous research, we dichotomized responses to the suffering measure in our primary analysis into (0) not at all/not very much and (1) some/a lot47.

Statistics and reproducibility

The research questions, variables, and analyses for the current study were preregistered with the Center for Open Science prior to accessing data, with only slight subsequent modification to the initial preregistered regression analyses due to multicollinearity (https://osf.io/z3tgr). All analyses were performed using R 4.2.248.

Descriptive statistics for the observed sample, weighted to be nationally representative within country, were estimated for each candidate predictor category. A weighted modified Poisson regression model (with complex survey-adjusted standard errors) was fit within each of the 22 countries separately, with the outcome of suffering regressed on all candidate predictors simultaneously. When available, religious affiliation at age 12 and racial/ethnic identity were included as control variables. We conducted a global test of association between each candidate predictor and the outcome in each country. For each candidate predictor, we calculated E-values to evaluate the sensitivity of results to unmeasured confounding. An E-value is the minimum strength of association that an unmeasured confounder must have with both the predictor and the outcome, above and beyond all measured covariates, for it to explain away their association49. The lowest E-value is 1, with higher values indicating that stronger unmeasured confounding would be needed to explain away the observed association between the predictor and the outcome.

The metafor package50 was used to perform random effects meta-analyses of the regression coefficients51,52 along with 95% confidence intervals (CIs), estimated proportions of effects across countries with effect sizes smaller than 0.90 and larger than 1.10, and I2 for evidence concerning variation within a given candidate predictor category across countries53 (further details about the meta-analytic methodology that was used can be found in Padgett et al.35). Coefficients for religious affiliation and racial/ethnic identity were not included in the meta-analyses because these variables were not measured consistently across all countries. A pooled p value54 was used across countries to report whether there is evidence of an association in at least one country. Bonferroni-corrected p value thresholds are provided based on the number of candidate predictors that were included in the meta-analyses: p = 0.05/11 = 0.004555. E-values were estimated for the regression coefficients included in the meta-analyses.

As a supplementary analysis, we conducted population-weighted meta-analyses of the same regression coefficients that were included in the random effects meta-analyses. We also performed a post-hoc sensitivity analysis in which we repeated the analyses after dichotomizing the outcome of suffering into categories of (0) not at all/not very much/some and (1) a lot.

Missing data

Missing data on all variables were imputed using multivariate imputation by chained equations, with five imputed datasets produced56,57. To account for variation in the assessment of certain variables across countries (e.g., religious affiliation, racial/ethnic identity), the imputation process was conducted separately in each country. This within-country imputation approach ensured that the imputation models accurately reflected country-specific contexts and assessment methods. Sampling weights were included in the imputation models to account for specific-variable missingness that may have been related to the probability of inclusion in the study.

Accounting for complex sampling design

Wave 1 of the Global Flourishing Study used different sampling schemes across countries based on the availability of existing panels and recruitment needs. All analyses accounted for the complex survey design components by including weights, primary sampling units, and strata. Additional methodological details, including the approach used to account for the complex sampling design, can be found elsewhere35,37.

Reporting summary

Further information on the research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Descriptive statistics

The distribution of descriptive statistics in the total sample is presented in Table 1. Countries with the largest samples included the United States (19%), Japan (10%), and Sweden (7.4%), while the smallest samples were from Türkiye (0.7%), South Africa (1.3%), and Hong Kong (1.5%). The gender distribution was roughly balanced between females (51%) and males (49%). The entire adult lifespan (18 to ≥80 years old) is represented in the sample, with the cohort that had the largest percentage of participants (20%) born between 1983–1993 (current age of 30–39 years). Most participants were born in the country where the data were collected (94%). A majority of participants reported that their parents were married when they were about 12 years of age (75%), they had a very good relationship with their mother (63%) and father (53%) when growing up, their family either got by financially or lived comfortably when they were around age 12 (76%), they had very good or excellent health while growing up (64%), and they attended religious services at least once a month when they were around age 12 (57%). A minority of participants indicated that they experienced abuse (14%) or felt like an outsider in their family when growing up (14%).

Table 1 Nationally representative descriptive statistics in the observed sample (N = 202,898)
Full size table

Nationally representative descriptive statistics for each country are reported in Supplementary Data 1–22, with some variability observed across the countries. For example, the percentage of individuals whose parents were married around age 12 was considerably higher in Hong Kong and Türkiye (Supplementary Data 6 and 20) than in India and South Africa (Supplementary Data 7 and 16), suggesting that early-life experiences, conditions, and influences are at least somewhat idiosyncratic across the countries.

Cross-country predictors of suffering in adulthood

Results of the random effects meta-analyses are reported in Table 2, while country-specific results are presented in Supplementary Data 23–44. Pooling across the countries, 95% CIs for the effect estimates provided evidence of an association with suffering in adulthood for eight of the candidate predictors. There was a higher likelihood of suffering in adulthood among individuals whose parents were divorced relative to those with married parents at 12 years of age (RR = 1.06, 95% CI: 1.02, 1.11), those who experienced abuse when growing up relative to those who did not (RR = 1.27, 95% CI: 1.21, 1.33), individuals who reported feeling like an outsider in their family when growing up relative to those who did not (RR = 1.20, 95% CI: 1.14, 1.26), and those who reported attending religious services 1–3 times a month around age 12 relative to those who never attended (RR = 1.05, 95% CI: 1.00, 1.10).

Table 2 Random effects meta-analyses for associations of candidate predictors with suffering in adulthood
Full size table

Individuals who reported a very good/somewhat good relationship with their father during childhood had a lower likelihood of suffering in adulthood relative to those who reported a very bad/somewhat bad relationship (RR = 0.96, 95% CI: 0.94, 0.98). Relative to individuals who reported having good health while growing up, the likelihood of suffering in adulthood was lower among those who reported very good (RR = 0.90, 95% CI: 0.87, 0.93) or excellent health (RR = 0.85, 95% CI: 0.79, 0.91) and higher among those with fair (RR = 1.09, 95% CI: 1.05, 1.13) or poor health (RR = 1.14, 95% CI: 1.07, 1.22). Relative to individuals who characterized their family as one that got by financially during their childhood, the likelihood of suffering in adulthood was lower among those who reported that they lived comfortably (RR = 0.94, 95% CI: 0.92, 0.97) and higher among those who reported that their family found it difficult (RR = 1.05, 95% CI: 1.02, 1.09) or very difficult financially (RR = 1.06, 95% CI: 1.01, 1.10). A higher likelihood of suffering in adulthood was also reported by females relative to males (RR = 1.08, 95% CI: 1.04, 1.12). There was little evidence suggesting that the quality of relationship with mother when growing up, immigration status, or year of birth were associated with suffering in adulthood when pooled across the countries, though statistical uncertainty in some of the estimates is notable.

Results of the E-value sensitivity analysis for the random effects meta-analyses can be found in Table 3. E-values for the effect estimates ranged from 1.09 to 4.50, with somewhat smaller E-values for the CI limit (1.00–1.72). For associations that excluded the null, E-values for the effect estimate (1.25–1.86) and CI limit (1.06–1.72) suggested that some of these associations might be moderately robust to potential unmeasured confounding. For example, to explain away the observed association between experiencing abuse when growing up and adult suffering, an unmeasured confounder that was associated with both childhood abuse and adult suffering by risk ratios of 1.86-fold each (above and beyond the other predictors adjusted for in the model) could do so, but weaker joint confounder associations could not. For the limit of the CI, unmeasured confounder risk ratio associations of 1.72 for both childhood abuse and suffering in adulthood could shift the CI to include the null, but weaker joint confounder associations could not.

Table 3 Sensitivity of meta-analyzed candidate predictors to unmeasured confounding
Full size table

When we repeated the meta-analyses using a population-weighted approach in which each country’s results were weighted by population size in 2023 (see Supplementary Data 45), results were comparable to those observed for the random effects meta-analyses and there were generally negligible differences in the magnitude of effect sizes (e.g., a risk ratio of 1.26 versus 1.27 for experiencing childhood abuse). The pattern of results was also similar when the random effects meta-analyses were replicated after dichotomizing the outcome of suffering into categories of (0) not at all/not very much/some and (1) a lot (see Supplementary Data 46), although effect sizes were generally somewhat larger (e.g., a risk ratio of 1.66 versus 1.27 for experiencing childhood abuse).

Similarities and differences across countries

In addition to the meta-analytic pooled estimates of the associations, Table 2 also provides evidence of the cross-national variation in the associations between the candidate predictors and suffering in adulthood. The metrics reported on (i.e., proportion of effects by thresholds, I2, global p values) each provide evidence of heterogeneity, and a richer description of the heterogeneity of these effects can be found in the forest plots displaying country-specific effect estimates and 95% CIs for each category of the candidate predictors (see Supplementary Figs. 1–27). There was some evidence of cross-national heterogeneity in associations for all candidate predictors, although heterogeneity was greater for some predictors compared to others. For example, associations for categories of self-rated health when growing up with suffering in adulthood were generally more heterogeneous across countries (I² = 57.1–92.3) than associations for categories of the subjective financial status of one’s family around age 12 (I² = 36.3–56.3). Cross-national heterogeneity in associations also varied across categories of each predictor. To illustrate, for frequency of religious service attendance at age 12 (reference category of never), the association between attending religious services at least once/week and suffering in adulthood was somewhat more heterogeneous across countries compared to the association for attending less than once/month (proportions below RR = 0.90 and above RR = 1.10 were 5% and 23% versus 0% and 9%, respectively). These findings suggest that the potential effects of the candidate predictors on suffering in adulthood may vary to some extent across sociocultural contexts.

All global p values in Table 2 were below the Bonferroni-corrected significance threshold of p = 0.0045, indicating that each candidate predictor was associated with suffering in adulthood in at least one of the countries. The country-specific results for candidate predictors of suffering in adulthood are presented in Supplementary Data 23–44; country-specific effect estimates for each candidate predictor category are also displayed visually using forest plots, with countries ordered on the y-axis by magnitude of association (see Supplementary Figs. 1–27). We did not find evidence of a completely universal pattern of associations between the candidate predictors and suffering in adulthood. However, there were some predictors that demonstrated a somewhat consistent pattern of associations in one direction or the other across a number of countries. The predictors that were associated with a higher likelihood of suffering in adulthood (95% CIs excluded the null) in the widest range of countries were experiencing abuse (see Supplementary Fig. 9) and feeling like an outsider (see Supplementary Fig. 10) when growing up (20/21 and 14/22 countries, respectively), whereas excellent health (see Supplementary Fig. 11) and very good health (see Supplementary Fig. 12) when growing up consistently predicted a lower likelihood of suffering across the widest range of countries (13/22 and 9/22 countries, respectively). Even in cases where the pattern of associations was more consistent across countries, effect estimates were often quite varied.

While most predictors tended to be consistent in their direction of associations with suffering in adulthood across countries, the pattern of associations for many of the birth year categories was quite mixed. For example, 95% CIs for effect estimates indicated that being in the 1943–1953 birth cohort (70–79 years) was associated with a higher likelihood of suffering in 6/22 countries and a lower likelihood of suffering in 6/22 countries (see Supplementary Fig. 24). In addition to birth year, there were other country-specific instances in which we found evidence of an association between a predictor and suffering in adulthood that was not observed in the pooled random effects meta-analyses. To illustrate, while the 95% CI for the association between immigration status and suffering in adulthood included the null when estimates were pooled across countries, there was evidence that being born in another country was associated with a lower likelihood of suffering in 4/22 countries and a higher likelihood of suffering in 1/22 countries (see Supplementary Fig. 15). There were also some cases where evidence of association documented in the pooled random effects meta-analyses was not observed for certain countries. One example of this is feeling like an outsider in the family when growing up, as there were 8/22 countries where the 95% CI for its association with suffering in adulthood included the null (see Supplementary Fig. 10).

Country-specific models also estimated the associations of religious affiliation at age 12 and racial/ethnic identity (when available), which were not meta-analyzed because these variables were assessed differently across countries (see Supplementary Data 23–44). The Philippines, Tanzania, and the United Kingdom were the only countries in which the 95% CIs for effect estimates provided evidence of association between one or more categories of childhood religious affiliation and suffering in adulthood. For example, relative to the no religion/atheist/agnostic category in Tanzania, Christian and Islam affiliations during childhood were both associated with a lower likelihood of suffering in adulthood, while affiliation with any other religion was associated with a higher likelihood of suffering. There were six countries in which 95% CIs for the effect estimates supported associations between racial/ethnic identity and suffering in adulthood. In four of those countries (i.e., India, Nigeria, Poland, South Africa), racial/ethnic minority status was associated with a higher likelihood of suffering; it was associated with a lower likelihood of suffering in adulthood in the other two countries (i.e., Argentina, Kenya).

Discussion

Leveraging cross-sectional data from a set of geographically and culturally diverse countries, this study explored associations of 13 individual characteristics and retrospectively assessed childhood factors with suffering in adulthood. Pooled effect estimates across countries suggested that a combination of individual characteristics (e.g., year of birth, gender), early-life events or experiences (e.g., child abuse, feeling like an outsider in one’s family), and social conditions or influences (e.g., family structure, quality of relationship with one’s father) within a child’s system of socioecology may be associated with suffering in adulthood, although the magnitude of effect estimates varied across predictors and there was some cross-national variation in the pattern of associations observed. By considering a range of potential childhood determinants of suffering in adulthood, the present study expands existing knowledge of how events, experiences, and conditions in early life might influence everyday suffering in a later phase of life. It also takes an important step toward developing an inclusive and globally representative epidemiology of suffering that is needed to strengthen a population health agenda focused more explicitly on addressing suffering.

Among the 11 candidate predictors that were considered in the random effects meta-analyses, there was some evidence of an association with suffering in adulthood for eight of them (effect sizes were generally very small). Consistent with a socioecological life course approach to human well-being21, these findings suggest that a combination of early-life risk and protective factors within a child’s socioecological system may have the potential to influence suffering in adulthood. Although the set of childhood predictors that were considered in this study focused principally on the microsystem, it is one of the first studies to apply a socioecological life course approach to the subjective experience of suffering.

Of the candidate predictors for which there was evidence of an association with suffering in adulthood in the random effects meta-analyses, the strongest associations (though still very small in magnitude) were found for childhood abuse and feeling like an outsider in the family predicting higher suffering in adulthood. These findings align with an abundance of research that has documented the potential long-term effects of relational trauma in childhood on health and well-being in adulthood58,59. Relational trauma during childhood is thought to influence a range of early maladaptive schemas (e.g., a tendency to exaggerate or dwell on negative aspects of life) that can distort how a person perceives, makes sense of, and responds to current events or experiences60. Research has shown that early maladaptive schemas are related to worse health and well-being in adulthood61, suggesting that early maladaptive schemas might also play a role in shaping experiences of suffering in adulthood. Additional empirical research is needed to explore this theorizing further.

There was also evidence linking better health when growing up with lower suffering in adulthood, although effect sizes were slightly weaker than those observed for experiencing childhood abuse and feeling like an outsider in one’s family when growing up. These findings correspond to evidence that suggests the quality of a person’s health during childhood can have long-term effects on their health and well-being in adulthood62,63, although a unique contribution of this study is that it extends this evidence to the experience of suffering. Childhood health could affect suffering in adulthood directly, such as through unrelenting chronic health problems, as well as indirectly, such as by constraining opportunities or experiences (e.g., ability to be physically active) that could protect against suffering later in life. Against the backdrop of increasing concerns about the growth in preventable chronic disease and disabilities (e.g., obesity, mental health conditions) among children in recent decades64,65, the findings of this study suggest that the suffering of future adult generations might be reduced through initiatives aimed at promoting their health during childhood.

Effect sizes for associations of having parents who were married around age 12, a more comfortable family financial situation when growing up, and a good/somewhat good relationship with one’s father when growing up with lower suffering in adulthood were more negligible. Compared to the pernicious effects that were observed for the risk factors of childhood abuse and feeling like an outsider in one’s family when growing up, these positive socially-connected factors may be less powerful in their protective influence on suffering in adulthood. This pattern of findings resonates with research on the asymmetrical effect of positive versus negative events or experiences, with the former generally found to have stronger and more enduring consequences than the latter (for a review, see ref. 66). While the role of positive family-related protective factors in shaping suffering in adulthood might be less substantial when salient risk factors are present, enhancing these protective factors may still offer some defense against suffering in adulthood, especially if multiple protective factors can be promoted together. Although the causal interpretation of our findings is limited, intervening on these protective factors could have implications for reducing suffering in adulthood if the population-level reach of those initiatives is extensive67. For example, because most children have a living father, interventions that have the potential to produce modest improvements in the quality of paternal relationships during childhood could have positive effects on population-level suffering as those children become adults.

Applying a systems perspective to the future well-being of the human population30, our findings point to the possibility of multipronged child- and family-centered interventions and policy initiatives aimed simultaneously at reducing risk factors (e.g., poor health) and strengthening protective factors (e.g., quality of the paternal relationship) during childhood having potential downstream implications for experiencing suffering later in life. Even investments that affect a small number of modifiable risk and protective factors in childhood could have an impact on reducing suffering in adulthood. When resource limitations are unable to support a multipronged intervention approach, our findings could be used alongside other evidence to guide decisions about which factors to prioritize, taking into account the magnitude of associations that were observed, how each factor might impact other facets of health and well-being over the life course, and the resources that may be needed to appropriately intervene on each factor.

Closer inspection of the country-specific results revealed some similarities and differences in the pattern of associations across countries. Most notably, 95% CIs for effect estimates supported an association between experiencing abuse during childhood and higher suffering in adulthood in almost all (20/21) countries in which it was assessed (see Supplementary Fig. 9). While effect sizes varied to some extent across countries, this near-universal association further underscores the destructive longer-term effects that childhood abuse can have on suffering among people from a range of countries that vary in socioecological system dynamics. These findings emphasize the importance of prioritizing childhood abuse prevention in early-life interventions aimed at reducing the burden of suffering later in life.

Other findings suggest that the influence of some candidate predictors on suffering in adulthood may be more limited to specific countries. For example, those who reported one or both of their parents had died by age 12 had a higher likelihood of experiencing suffering in adulthood (versus those whose parents were married) in India, Nigeria, and South Africa (see Supplementary Fig. 5), pointing to the possible unique impact that growing up without one or both parents may have on later-life suffering within certain socioecological contexts. Although there may be several reasons for this pattern of findings (e.g., differences in population demographics, cultural variation in how suffering is understood and experienced), one possible explanation is that in some contexts the influence of proximal factors within the child’s socioecological system on suffering in adulthood might themselves be affected by factors situated within layers that are more distal24. In South Africa, for instance, pervasive social-structural disadvantages within the macrosystem (e.g., poverty, unemployment) may exacerbate the strains experienced by parentally bereaved families (e.g., income loss) in ways that impinge upon developmental opportunities which might ordinarily reduce sources of suffering later in life68.

As a further indication of the unique socioecological context of each country, the pattern of associations observed for the full set of candidate predictors varied across countries. In some countries (e.g., Australia, Mexico), 95% CIs for effect estimates indicated that fewer than half of the 13 candidate predictors were associated with suffering in adulthood, whereas in other countries (e.g., Japan, the United Kingdom) there was evidence of association for most candidate predictors. Such cross-country variation suggests it may be important to develop policies and interventions that are responsive to the country-specific profile of childhood risk and protective factors for suffering in adulthood, while also being sensitive to socioecological system dynamics that could impact the effectiveness of infrastructure and resources put in place.

This study has selected limitations. First, childhood predictors were assessed retrospectively in the same wave that the outcome of suffering was measured. Therefore, our findings may be influenced by recall and common method bias. For example, it is possible that some of the recalled childhood experiences might have been perceived more negatively by those who endorsed higher levels of suffering. Recall bias can also vary based on participant characteristics (e.g., age differences in memories about childhood due to the passage of time) and across countries (e.g., cultural differences in memory processing), which should be considered when interpreting the findings of this study. However, for recall bias to completely explain away the observed associations would require that the effect of suffering in adulthood on biasing retrospective assessments of the childhood predictors is at least as strong as the observed associations themselves69. To mitigate sources of bias, future research might employ prospective longitudinal designs that follow individuals from childhood to adulthood and complement self-report survey responses with data derived from other sources (e.g., parents).

Second, all psychosocial constructs (including the outcome of suffering) were assessed using a single item. While the use of single items to assess a construct is somewhat common in large-scale epidemiologic studies, conceptual coverage is typically narrower compared to multi-item measures. Research could build on this study’s findings by using measures that provide broader conceptual coverage of the psychosocial constructs that were assessed.

Third, suffering is a contextualized experience; it unfolds under a set of personal circumstances embedded within a particular sociocultural environment that may impinge upon a person’s experience of suffering20,29. For example, experiences of suffering may vary based on differences in the principal cause of suffering (e.g., posttraumatic stress disorder versus cancer), the object of suffering (e.g., disrupted sense of peace versus loss of identity), or cultural influences (e.g., norms about whether suffering ought to be expressed versus repressed)5,70. Our single-item measure of suffering does not capture the broader context of suffering, which limits our understanding of the suffering experienced by participants. Future large-scale studies might consider attending more closely to the contextualized nature of suffering, which could provide opportunities for exploring finer-grained questions about suffering (e.g., how do childhood predictors of suffering in adulthood vary based on the precipitating cause of suffering?).

Fourth, we cannot rule out the possibility that some of the observed associations might be explained away by unmeasured confounding. While E-values suggested that some of the observed associations which did not include the null might be somewhat robust to unmeasured confounding, the set of predictors may not have provided adequate confounding control. Although decisions about the inclusion of predictors were constrained by the variables that were measured in the Global Flourishing Study survey, future studies could strengthen causal claims by including a richer set of relevant covariates (e.g., parental health during childhood, neighborhood quality while growing up).

Fifth, it is possible that certain predictors might be on the pathway (i.e., a mediator) between another predictor and the outcome. For example, parental divorce around age 12 might be associated with suffering in adulthood through feeling like an outsider in one’s family while growing up that arises following parental divorce. If there were such instances of mediation in the present study, those observed associations might be conservative estimates of actual overall effects.

Sixth, while Wave 1 of the Global Flourishing Study includes a substantially diverse set of 22 countries that collectively represent approximately half of the world’s population, some cultures and contexts are not represented in the dataset. Attempts to generalize the findings beyond the countries that were included in our analytic sample should be made with some caution, and replication studies are needed in countries that are not represented in Wave 1 of the Global Flourishing Study.

In summary, the present multinational study employed a life course epidemiological approach to explore candidate predictors of suffering in adulthood. The findings provided some evidence suggesting that a combination of individual characteristics, early-life events or experiences, and social conditions or influences within a child’s system of socioecology may be associated with suffering in adulthood (both within and across the countries), laying some of the groundwork for a globally inclusive epidemiology of suffering. Although additional research is needed to build on the findings of this study, the evidence reported herein could have practical utility both in terms of supporting adults who are currently suffering and developing multipronged early-life interventions to reduce later-life suffering among children.