Practice of Epidemiology Adjusting for Confounding by Neighborhood Using a Proportional Odds Model and Complex Survey Data

In social epidemiology, an individual’s neighborhood is considered to be an important determinant of health behaviors, mediators, and outcomes. Consequently, when investigating health disparities, researchers may wish to adjust for confounding byunmeasured neighborhood factors, such as local availability of health facilities or cultural predispositions. Withasimple randomsampleandabinaryoutcome,aconditional logistic regressionanalysis that treats individualswithin a neighborhood as a matched set is a natural method to use. The authors present a generalization of this method for ordinaloutcomesandcomplexsamplingdesigns.Themethod isbasedonaproportionaloddsmodeland isverysimple to programusingstandardsoftwaresuchasSASPROCSURVEYLOGISTIC (SAS Institute Inc.,Cary,NorthCarolina).The authorsapplied themethod toanalyze racial/ethnic differences indental preventative care, using2008FloridaBehavioral Risk Factor Surveillance System survey data. The ordinal outcome represented time since last dental cleaning, and the authorsadjusted for individual-level confoundingbygender, age, education, andhealth insurancecoverage. Theauthors compared results with and without additional adjustment for confounding by neighborhood, operationalized as zip code. The authors found that adjustment for confounding by neighborhood greatly affected the results in this example.