Influenza vaccination in the elderly: Identifying disparities, predictors of county-level dissemination, and methodologic examination of classification and regression tree analysis.

摘要

This study applied two novel approaches to the assessment of factors associated with influenza vaccination. The first approach, Classification and Regression Tree (C&RT) analysis, was used to identify segments of the elderly population at highest risk for not receiving influenza vaccination. A methodologic review of C&RT and a comparison of its statistical properties to logistic regression were detailed; followed by a C&RT analysis that identified unique, mutually exclusive population segments whose members shared characteristics associated with likelihood of influenza vaccination receipt. Data from 57,553 participants aged 65 and over in the 1997 and 1999 Behavioral Risk Factor Surveillance Systems (BRFSS) were analyzed. The outcome of interest was receipt of an influenza vaccination in the previous year. Independent variables included socio-demographic characteristics, regular checkup, smoking, problem alcohol use and self-rated health. Prevalence of influenza vaccination was 66%. Eight subgroups were identified, with prevalence estimates ranging from 47.5% to 78.9%. Lowest estimates were observed those who did not have a recent checkup and non-Hispanic Blacks and Hispanics who did receive a recent checkup. The second approach involved examining county-level variation in the dissemination of influenza vaccination in the elderly from 1993 to 1999. A retrospective cohort was assembled from the 1993 and 1999 BRFSS, aggregated to the level of the county (n = 353). Independent variables were also obtained from the 1993 Area Resource File and the Public Health Workforce Enumeration project. Dissemination was defined as the difference between the 1993 and 1999 county-level prevalence rates. A two-stage Heckman selection model was generated. Average prevalence increased from 50% in 1993 to 68% in 1999. Positive predictors of dissemination included greater median per capita income, ratio of state public health workforce to population, dissemination of pneumonia vaccination, and percentage of total deaths from non-ischemic heart conditions and chronic lung conditions. Negative predictors were percentage of the population that was Black or other race, and baseline influenza vaccination prevalence. Results from both approaches indicate outreach efforts should increase targeting of under-served populations, and that community-based approaches should be expanded, as persons with less access to the medical care system are receiving vaccinations less.