Selection or Causation? Using Marginal Structural Models to Estimate the Causal Effect of Neighborhood Disadvantage on Alcohol Use and Misuse

Magdalena Cerda, The New York Academy of Medicine
Ana Diez Roux, University of Michigan
Eric Tchetgen, Harvard School of Public Health
Penny Gordon-Larsen, University of North Carolina at Chapel Hill
Catarina Kiefe, University of Alabama at Birmingham

Previous research has reported conflicting findings about the relationship between neighborhood disadvantage and alcohol use/misuse. We used data from the Coronary Artery Risk Development in Young Adults study of 5,115 adults interviewed seven times between 1985 and 2000. We estimated inverse-probability-of-treatment-and-censoring (IPTC)-weighted models to assess the effect of cumulative neighborhood poverty on alcohol use after accounting for time-dependent confounders and mediators. The IPTC-weighted estimate of the number of drinks-per-week associated with a one-unit increase in poverty was OR: 1.23 (95% CI:0.98,1.56), while the estimate from an unweighted model controlling for baseline and time-varying covariates was OR: 1.28 (95% CI:1.05,1.57). Under certain conditions, traditional regression methods provide estimates comparable to techniques proposed to adjust for simultaneous time-varying mediators and confounders in longitudinal studies. Neighborhood poverty was not associated with the odds of binging: cumulative exposure to neighborhood poverty may influence level of alcohol use, but not extreme manifestations of use such as binging.