Multiple imputation for correcting verification bias

Abstract

In the case in which all subjects are screened using a common test and only a subset of these subjects are tested using a golden standard test, it is well documented that there is a risk for bias, called verification bias. When the test has only two levels (e.g. positive and negative) and we are trying to estimate the sensitivity and specificity of the test, we are actually constructing a confidence interval for a binomial proportion. Since it is well documented that this estimation is not trivial even with complete data, we adopt multiple imputation framework for verification bias problem. We propose several imputation procedures for this problem and compare different methods of estimation. We show that our imputation methods are better than the existing methods with regard to nominal coverage and confidence interval length. Copyright © 2006 John Wiley & Sons, Ltd.

Multiple imputation: review of theory, implementation and software

“…When using MI, one can use several complete-data methods to estimate the sensitivities and specificities and use the different results as sensitivity analysis. Harel and Zhou [58] performed a simulation study comparing six different complete-data procedures and showed that one method was superior to all other complete-data methods. This method is:…”

“…In Harel and Zhou [58], we compared the performance of the logit-based MI and the B&G method and found that the logit-based MI outperformed the B&G method. Since the sample size in this example is quite large (N = 34, 874), and we have proper MI, we can assume that the MI results are more appropriate than the B&G results.…”

Clinical decision rule to predict the presence of interstitial lung disease in systemic sclerosis

“…The direction and extent of the bias depend, therefore, on the relationship of the selection mechanism to the true state of the disease. Methods have been developed to adjust for the problem of verification bias, including multiple imputations (13).…”

“…Multiple imputation was used to impute HRCT data for those who were missing these data and thereby correct for a possible verification bias. Indeed, the problem of verification bias is, by definition, a missing data problem (13), and multiple imputation has been used successfully in many clinical contexts for addressing problems due to missing data in the last 25 years (16,17). The basic principles of multiple imputation are quite simple.…”

“…Finally, these separate analyses are combined using a set of combining results to obtain a single estimate and a single confidence interval for each parameter of interest. Multiple imputation has previously been used to assess the diagnostic accuracy of clinical criteria while correcting for verification bias (13).…”