Public Health Application Comparing Multilevel Analysis with Logistic Regression: Immunization Coverage among Long-Term Care Facility Residents

Purpose: Public health studies often sample populations using nested sampling plans. When the variance of the residual errors is correlated between individual observations as a result of these nested structures, traditional logistic regression is inappropriate. We used nested nursing home patient data to show that one-level logistic regression and hierarchical multilevel regression can yield different results. Methods: We performed logistic and multilevel regression to determine nursing home resident characteristics associated with receiving pneumococcal immunizations. Nursing home characteristics such as type of ownership, immunization program type, and certification were collected from a sample of 249 nursing homes in 14 selected states. Nursing home resident data including demographics, receipt of immunizations, cognitive patterns, and physical functioning were collected on 100 randomly selected residents from each facility. Results: Factors associated with receipt of pneumococcal vaccination using logistic regression were similar to those found using multilevel regression model with some exceptions. Predictors using logistic regression that were not significant using multilevel regression included race, speech problems, infections, renal failure, legal responsibility for oneself, and affiliation with a chain. Unstable health conditions were significant only in the multilevel model. Conclusions: When correlation of resident outcomes within nursing home facilities was not considered, statistically significant associations were likely due to residual correlation effects. To control the probability of type I error, epidemiologists evaluating public health data on nested populations should use methods that account for correlation among observations. [Copyright &y& Elsevier]