Holistic Models of Health: A Structural Equation Modeling Framework

OBJECTIVE: Structural equation modeling (SEM) has undergone major advances in the last 10 to 15 years including a merger with generalized linear modeling, mortality hazard modeling, and the incorporation of estimation for complex survey data. Traditional modeling capabilities available in the SEM framework include streamline mediation and moderation testing, the incorporation of latent (unobserved) variables, item-missing data analysis methods, and longitudinal (random effects) modeling.

METHODS: SEM combines simultaneous equation estimation with factor analysis. A major advantage of SEM is the ability to extract random measurement error from observed clinical and survey measures, for example, self-reported health statuses. Multiple pathways of health may also be analyzed allowing for an evaluation of complex health processes and holistic models of health and health behavior. Several applications of specific SEM techniques are demonstrated using the National Health Interview Survey and the Health and Retirement Survey.

RESULTS: SEM is used to 1) evaluate mediated pathways to mortality correcting for the effects of health conditions that operate through multiple pathways, 2) measure disability and depression among other health constructs with multiple indicators, and 3) estimate random effects trajectories in visual impairment among an aging population. These methodological examples also incorporate probability weighting and corrections for nesting as well as accommodate item missingness, which maximizing sample size and power.

CONCLUSION: SEM has traditionally been a methodology used by social scientists. Advances over the last 15 years have made SEM more useful to health researchers especially the ability to model non-normal outcomes and time-to-event outcomes such as mortality.