Modelling the effect of treatment adherence on the health stock of Chronic Obstructive Pulmonary Disease patients in the US
Background: Smoking is the most important risk factor for COPD and smoking cessation is known to slow disease progression and improve symptoms. Annual influenza vaccination helps to prevent chest infections while medications, which reduce or abolish symptoms, complement these effective preventive treatments. Smoking cessation and regular annual influenza vaccination are known to be efficacious but their effectiveness depends on the patient’s adherence to treatment. Currently, about 5-40% of COPD hospital admissions are associated with non-adherence. Therefore, improving adherence offers another important benefit: it can significantly reduce potentially avoidable hospitalizations and the associated costs.
Data and Method: The study utilizes pooled cross-sections of the most recent individual level data of US adults aged 18 years and above with a diagnosis of COPD prior to consecutive BRFSS survey rounds in 2011-2013. The BRFSS database contains information on individual level behavioural risk factors associated with the leading causes of morbidity and premature mortality among adults in the U.S. and its territories. In addition to demographic and socioeconomic information, the database has data on risk factors including cigarette smoking, alcohol use, diet, hypertension, physical activity and safety belt use. The key outcome variable of our study is the count of days of good health in the 30 days preceding the survey and we model this as a function of adherence to smoking cessation and flu vaccination, controlling for biological and economic factors. We assume that different processes generate the positive and zero values of counts of healthy days and employ the double hurdle model to test the hypothesis that adherence to smoking cessation and influenza vaccination advice has no effect on the count of healthy days. Although count data usually follow a Poisson distribution, the different mean and variance of the dependent variable in this study suggests a Poisson regression will be inappropriate. The double hurdle regression model is preferred to OLS regression in this context due to the usually skewed nature of count data.
Potential Contributions: Unlike previous studies, this paper adds to the treatment adherence literature in three significant ways. First, we focus on the impact of treatment adherence on an economic outcome. Second, we model adherence to treatment as a two-stage decision process by fitting a double hurdle regression model to 2011-2013 data of the Behavioral Risk Factor Surveillance System (BRFSS) of the Centers for Disease Control and Prevention (CDC). Third, we analyze a large and nationally representative data set to make the research finding broadly applicable.