Testing and Estimating Model-Adjusted Effect-Measure Modification Using Marginal Structural Models and Complex Survey Data

Abstract

Recently, it has been shown how to estimate model-adjusted risks, risk differences, and risk ratios from complex survey data based on risk averaging and SUDAAN (Research Triangle Institute, Research Triangle Park, North Carolina). The authors present an alternative approach based on marginal structural models (MSMs) and SAS (SAS Institute, Inc., Cary, North Carolina). The authors estimate the parameters of the MSM using inverse weights that are the product of 2 terms. The first term is a survey weight that adjusts the sample to represent the unstandardized population. The second term is an inverse-probability-of-exposure weight that standardizes the population in order to adjust for confounding; it must be estimated using the survey weights. The authors show how to use the MSM parameter estimates and contrasts to test and estimate effect-measure modification; SAS code is provided. They also explain how to program the previous risk-averaging approach in SAS. The 2 methods are applied and compared using data from the 2007 Florida Behavioral Risk Factor Surveillance System Survey to assess effect modification by age of the difference in risk of cost barriers to health care between persons with disability and persons without disability.

Age Group Differences in Healthcare Access for People With Disabilities: Are Young Adults at Increased Risk?

“…Finally, we conducted a test for homogeneity of the risk difference across the three age groups. These analyses used inverse weighting to adjust for confounding and selection bias introduced by the complex survey design [5]; analyses were completed using SAS 9.2 (SAS Institute Inc., Cary, NC). This study was approved by the University of Florida's Institutional Review Board.…”

Investigating the relationship between neighborhood poverty and mortality risk: A marginal structural modeling approach

“…In contrast, the weighting for the naïve models is composed of only the PSID longitudinal survey weights. For both strategies, we use Huber–White sandwich estimators to obtain robust standard errors (Brumback, Bouldin, Zheng, Cannell, & Andresen, 2010). …”

Expanding the Scope of Risk Assessment: Methods of Studying Differential Vulnerability and Susceptibility

“…Recently, marginal structural models and inverse probability-of-treatment weighting have been used to examine effect modification. 19,20 This approach is effective in studying dynamic life course developmental processes, where the value of either exposures or effect modifiers is known to be time varying. 21 Berkey et al demonstrated the use of multilevel random-effects models for estimating effect modification across places, 22 an approach well illustrated in an analysis of EM effect modification in 29 European cities in the APHEA2 project.…”