Maria Gultzow

Mental health and ill-health are not equally distributed across the population. However, even though there is a vast amount of literature that describes these differences across subgroups and suggests points for intervention, these differences persist. The overarching aim of this dissertation was to gain a deeper understanding of the underlying mechanisms that contribute to subgroup differences in mental health and ill-health. This thesis used the potential outcomes framework, the g-formula, marginal structural models and genetically informed designs with the aim to add knowledge about the impact of determinants that are potential targets for interventions (i.e. health behavior, obesity, or socioeconomic factors), and determinants that are not intended as potential intervention targets but might substantially explain subgroup differences in mental health and ill-health (i.e. genetics). The focus was on investigating two mechanisms that might explain why some subgroups are better off than others, namely differential exposure and differential impact. Differential exposure indicates that the variation in mental health is partly driven by the unequal distribution of intermediary determinants across subgroups, whereas differential impact indicates that subgroup variations are driven by a stronger effect of intermediary determinants on mental health among certain subgroups.

The first part of this dissertation focused on assessing whether intervening on modifiable determinants of mental health would reduce subgroup differences in mental health and ill-health in the population. In detail, Chapter 2 assessed the contribution of health behavior and health behavior-related determinants (i.e. alcohol consumption, smoking, physical activity, and obesity) to the worsening of mental health among more recently born cohorts. We analysed panel data from US adults born 1916-1966 enrolled in the Health and Retirement Study and performed a counterfactual decomposition analysis in which we combined age-period-cohort models with g-computation. This chapter introduced a hypothetical intervention that assigns every birth cohort the health behavior of the cohort with the lowest depression risk, namely cohort 1945. Holding age and time period constant, cohorts born before 1920 and after 1950 had a higher depression risk compared to birth cohorts born between those years. The hypothetical intervention on alcohol consumption increased depression risk of the 1916-1949 and 1950-1966 birth cohorts, whereas the hypothetical intervention on obesity increased the depression risk for the 1916-1940 cohorts and decreased depression risk for the 1950-1966 cohorts. The contribution of alcohol was more pronounced for White than for other racial/ethnic groups, and the contribution of obesity was more pronounced for women than for men. There was no evidence for contributions of smoking and physical activity.

Chapter 3 focused on understanding to what extent gender inequality at the labour market explains the higher depression risk for women compared to men aged 50 onwards. We analysed data from 35,699 US adults aged 50-80 years that participated in the Health and Retirement Study. This chapter employed a dynamic causal decomposition and simulated the life course of a synthetic cohort from ages 50–80 with the longitudinal g-formula. We introduced four nested hypothetical interventions and assigned women the same probabilities of being in an employment category, occupation class, current income and prior income group as men. Women’s depression risk was higher than that of men across ages 50-70. Equalizing opportunities at the labour market across gender would reduce this gap. The reduction was largest for Hispanics and low educated groups.

Chapter 4 investigated how childhood obesity contributes to the larger burden of mental ill-health among young Dutch adolescents from low education or income households, compared to higher education and income households. The sample consisted of children residing in the Netherlands that participated in the Generation R study. This chapter employed a four-way decomposition and used marginal structural models with inverse probability of treatment weighting. The implied hypothetical interventions were: 1) assign the low and medium socioeconomic position (SEP) groups the obesity distribution of the high SEP group (remove differential exposure) and 2) assign the low and medium SEP group the same impact of obesity on mental health as in the high SEP groups (remove differential impact). Children who grew up in low maternal education or low-income households had more emotional and behavioral problems than children who grew up in high SEP settings. Equalizing the differential exposure to obesity across SEP would lead to a reduction in emotional problems which was larger for girls than for boys. Obesity did not contribute to behavioral problems and there was no evidence that differential impact of obesity contributes to socioeconomic differences in mental health.

Chapters 2-4 suggest that intervening on modifiable determinants of mental health generally reduced (and in some cases increased) subgroup differences in mental health and ill-health. The contributions of modifiable determinants to subgroup differences in mental health and ill-health is driven by differential exposure with no evidence for differential impact driving these subgroup differences.

The second part of this dissertation investigated whether determinants of mental health that do not present potential intervention targets contribute to subgroup differences in mental health and ill-health in the population. In detail, Chapter 5 and 6 focused on the contribution of genetic factors to subgroup differences in mental health and ill-health by partnership status. Both chapters studied Finnish individuals that participated in the FINRISK and Health 2000 and 2011 surveys. Chapter 5 examined whether partnership status moderated the association between the genetic predisposition to depression and time to antidepressant purchasing through an accelerated failure time model. Widowed had the largest predicted cumulative hazard of antidepressant purchasing, followed by divorced, single, married and cohabiting. The highly genetically predisposed had a higher predicted cumulative hazard of antidepressant purchasing than the medium and low genetically predisposed. This chapter found no evidence that genetic factors had a differential impact on incident antidepressant purchasing across partnership status groups. Chapter 6 examined whether antidepressant purchasing surrounding partnership transitions are driven by the genetic predisposition to depression. This study reported that the high genetically predisposed group is most adversely affected in regards to antidepressant purchasing leading up to a union dissolution, but not union formation. This indicates that differential impact of genetic factors may depend on the time surrounding a change in partnership status. Chapter 5-6 found inconsistent evidence for a contribution of genetic factors to subgroup differences in mental health. Whereas Chapter 5 found no evidence for a differential impact of partnership status groups on incident antidepressant purchasing driven by genetic predisposition to depression, Chapter 6 reported that the trajectories in antidepressant purchasing before a divorce depend on the genetic predisposition to depression.

This dissertation identified that differential exposure to modifiable determinants explained at least part of the subgroup differences in mental health and ill-health, but the size of the contribution depends on the type of determinants that were examined, the underlying hypothetical intervention, and the intersection with other subgroups. The contributions of differential impact are less clear and may be time dependent and harder to quantify as it may be a consequence of differential exposure. Furthermore, the methodological considerations underscore the need to balance simplicity and complexity in our models, evaluate the trade-off between conservative (but well-defined) or radical (but vague) interventions in regards to the consistency assumption in causal inference, distinguish between modifiable and non-modifiable determinants, and clarify the advantages and disadvantages of subjective and objective measures of mental health. To conclude, in order to address the persistence of subgroup variation in mental health and ill-health, future policy needs to not only address the vulnerability of certain subgroups at the individual level, but also shift structures and the overall socioeconomic and political context in which these subgroups act in.