Propensity Weighting to Investigate Health Disparities and Assess Indirect Effects.
Health disparities are a public health concern as the term implies that certain groups are not getting the same quality health care as other groups. Two such populations in the United States are African Americans and Latinos. These disparities typically result in the minority population receiving inferior care. We examine this issue by looking at care received in a group of breast cancer patients who are enrolled in medicare. We use the marginal odds ratio to compare adequacy of care received comparing African Americans to Whites and Latinas to Whites. We argue that that the marginal odds ratio is the measure that should be used to compare care received in different racial or ethnic groups and discuss the term ”causality” in this context. We use propensity weighting to calculate the marginal odds ratio. We discuss different methods of calculating marginal odds ratios and show that propensity score matching leads to a biased estimator of the marginal odds ratio.
We also investigate propensity weighting in assessing indirect effects mediated by another risk factor. We demonstrate the ideas via simulation studies and apply the method to a study of Cancer of Unknown Primary (CUP). The exposure is confirmed vs unconfirmed CUP and the mediator is treatment. A CUP is confirmed after a set of recommended tests are performed but a primary cancer is not found. A CUP is unconfirmed if the tests are not done.
Professor Christiana Drake, University of California, Davis