|
Abstract:
|
A “fusion” study design combines data from different sources to answer a question that could not be answered by subsets of the data sources (e.g., measurement-error correction studies). A “fusion” estimator is an algorithm, here a stacked estimating function, which, on average, produces a correct answer to a question using data from a fusion design. We describe a pair of examples of fusion designs and illustrate fusion estimators. First, we transport an estimate of the proportion, using external auxiliary data on a target population; second, we correct a misclassified estimate of the proportion, using external auxiliary validation data. For each case, we present an example motivated by work in HIV and summarize results from limited simulation studies. In the examples, fusion estimators provided approximately unbiased results, appropriate confidence interval coverage, and are an appealing approach to estimate the parameters of fusion designs. Fusion designs can help health scientists to appropriately combine data in answering important questions that demand multiple sources of information.
|
