Combining estimates from two surveys: an example from monitoring 2009 influenza A (H1N1) pandemic vaccination

Abstract

During the 2009 influenza A (H1N1) pandemic, there was an ongoing need to monitor 2009 H1N1 vaccination coverage at the national and state level to evaluate the vaccination campaign; thus, precise vaccination coverage estimates were needed in a timely fashion. The current objective is to describe and evaluate the methodology used to combine 2009 H1N1 vaccination coverage estimates from the Behavioral Risk Factor Surveillance System (BRFSS) and the National 2009 H1N1 Flu Survey (NHFS). H1N1 state level vaccination coverage estimates were combined by taking weighted averages of the BRFSS and NHFS estimates, with more weight given to the estimate with the larger effective sample size (sample size/design effect). The impact of the choice of weights was evaluated by comparing estimates when the design effect was removed from the weights. Combined vaccination coverage estimates for children generally fell midway between NHFS and BRFSS estimates because of larger NHFS sample sizes but smaller BRFSS design effects. Adult estimates were more closely weighted to BRFSS estimates because of larger BRFSS sample sizes. Combined standard errors were smaller than the survey-specific standard errors. When removing the design effect from the weights, the child combined estimates were more closely weighted to those from NHFS, resulting in larger standard errors. Adult combined estimates were similar regardless of choice of weight because of similar design effects across the two surveys. Combining estimates by weighting by the effective sample size allowed timely release of more precise estimates in all states during the 2009 H1N1 pandemic.