Statistical Methods for Adjusting the Confounding Effects of Multiple Variables

-A Cross-Sectional Study of Respiratory Diseases-

Abstract

This paper describes methods of calculating adjusted rates by using regression coefficients estimated from binary variable multiple regression analysis and multiple logistical regression analysis. The methods discussed are being applied to data collected in a cross-sectional study of respiratory diseases in a rural area of Beijing. The effects of social, biological and environmental factors on pulmonary functions have been examined. Other adjustment methods were also used to analyse the data for the purpose of comparison with the above methods. Of the potential risk factors, associated with impairment of pulmonary function before adjustment, only age and smoking were consistently after adjustment. These results suggest that the effects of sex and social economic status on impairment of pulmonary function were confounded by the other factors. In a further analysis of confounding factors, age and smoking were found to have distorted the risk estimates of social economic status and sex separately. The relative merits of each method are discussed. It is emphasized that when the sample size is relatively small or/and the number of influencing factors is large, regression analysis methods should be used, Mantel-Haenzszel method and logistic regression method are most appropriate for relative odds, and linear regression method is most appropriate for differences of rate in the evaluation of potential risk factors. Finally, regression models were developed to assess the relative risk on the basis of information available in this cross-sectional study, and the overall prevalent risk of impaired pulmonary function.