Huang et al. Respond to “Multigenerational Social Determinants of Health”

We thank Ms. Cohen and Dr. Lê-Scherban for their thoughtful commentary (1) on our paper concerning associations between grandmaternal education and grandchild birth weight among US infants born in the 2000s (2). We appreciate their efforts to set our work within the broader context of multigenerational studies and concur with their assessment of the related challenges and opportunities. We agree with their call for a more comprehensive account of complex social and biological theories when applying novel analytical methods and agree that our approach, among others (3–5), represents an early step. To that end, we highlight opportunities implied by our work to address challenges on which Cohen and Lê-Scherban elaborate, specifically the tenacious issues of complex causal structures and residual confounding. Cohen and Lê-Scherban identified health selection and social status transmission as 2 key features from prevailing social theory that complicate the identification of causal relationships. With suitable data, both issues may be substantively addressed using marginal structural models (MSMs). For example, health selection, wherein status attainment may be hampered by poor health, can be addressed by weighting individuals by their probabilities of low socioeconomic status, as predicted by some earlier health state. In the rare case that the earlier health state is itself outside the pathway of interest and there are sufficient measured predictors to randomize earlier health state (i.e., to satisfy MSM assumptions), earlier health state can also be controlled for in the outcome model. More likely, as we implemented for adolescent (prepregnancy) body mass index, the measure is omitted from the outcome model and therefore included in the estimated “effect.” Additionally, reverse causality may be addressed by incorporating additional longitudinal socioeconomic status and health data at finer time scales. A major challenge identified in past studies of early-life socioeconomic disparities in birth outcomes is appropriately accounting for numerous potential pathways (1, 2), particularly parental behaviors and social status transmission (4). Our study focused on the relative importance of matrilineal early-life socioeconomic status in the context of other life-course variables that may feasibly be intervened upon, such as child maltreatment or prenatal smoking. Nonetheless, MSMs may be used more generally to obtain an estimate of early-life disparity that remains after intervening on mediators (6), including disparities due to patrilineal factors (1). In fact, MSMs provide a platform with which to generate evidence for alternative or competing pathways (5) by allowing for additional mediators to be considered and, as alluded to above, allowing for mediator-outcome associations to be tested under various presumed causal dependencies. Importantly, although the approach assumes no unmeasured confounding of exposures, it does not preclude the presence of other unmeasured mediators, including potentially important paternal contributions (1). In other words, our findings may be consistent with a number of causal pathways, including those suggested by Cohen and Lê-Scherban. Nonetheless, we must remain vigilant for confounding, including confounding due to measurement error in educational quality (1), even in the presence of reassuring bias analyses. Notably, categorization of continuous mediators for the purposes of avoiding positivity violations and facilitating MSM estimation may also introduce residual confounding. While we found point estimates to be similar between variously parameterized regression models, this will not always be true, and we encourage comparisons between alternate parameterizations. Additionally, subgroup analyses may reduce concern about residual confounding. For example, if numbers had allowed, we might have fitted models separately for mothers who lived apart from 1 or both grandparents. Under a hypothesis that maternal early-life development is important independent of status transmission, similar associations with birth weight might also be observed for these mothers. Ultimately, replication and refinement of models will provide the most rigorous support. Future research can build on this work in several ways. First, theory and empirical findings can be employed to improve the set of predictors used to estimate probability weights. For example, certain neighborhood characteristics (7) may be nonignorable, time-dependent confounders of socioeconomic attainment and birth outcomes. Additionally, measurement of biomarkers at discrete time points may help identify mediation through specific biological pathways. Finally, we would be remiss not to mention the potential utility of other causal inference approaches. In particular, difference-in-difference or regression discontinuity approaches (8) may be helpful for settings with identified variations in educational policies (9). A growing body of economics literature on intergenerational health transmission uses sibling fixed effects (10). However, these methods trade off some complexity in understanding pathways in favor of specific causal effects. In contrast, we suggest that the use of MSMs with replication and refinement can help in implementing, rather than avoiding, complex social and biological theories. Author affiliations: Institute for Health and Social Policy, McGill University, Montreal, Quebec, Canada (Jonathan Y. Huang); Department of Epidemiology, School of Public Health, University of Washington, Seattle, Washington (Jonathan Y. Huang, Ali Rowhani-Rahbar, Daniel A. Enquobahrie); School of Social Work, University of Washington, Seattle, Washington (Amelia R. Gavin); Department of Statistics, College of Arts and Sciences, University of Washington, Seattle, Washington (Thomas S. Richardson); and New York Academy of Medicine, New York, New York (David S. Siscovick). Financial support for our study was provided by the US National Institutes of Health through Reproductive, Perinatal, and Pediatric Epidemiology Training Grant 5T32HD052462-08 and Career Development Award K01HL103174. The work was also supported by Canadian Institutes of Health Research Operating Grant 115214. Conflict of interest: none declared.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.