Capture-recapture method for estimating misclassification errors: application to the measurement of vaccine efficacy...

The measure of efficacy is optimally performed by randomized controlled trials. However, low specificity of the judgement criteria is known to bias toward lower estimation, while low sensitivity increases the required sample size. A common technique for ensuring good specificity without a drop in sensitivity is to use several diagnostic tests in parallel, with each of them being specific. This approach is similar to the more general situation of case-counting from multiple data sources, and this paper explores the application of the capture-recapture method for the analysis of the estimates of efficacy.

An illustration of this application is derived from a study on the efficacy of pertussis vaccines where the outcome was based on > or =21 days of cough confirmed by at least one of three criteria performed independently for each subject: bacteriology, serology, or epidemiological link. Log-linear methods were applied to these data considered as three sources of information.

The best model considered the three simple effects and an interaction term between bacteriology and epidemiological linkage. Among the 801 children experiencing > or =21 days of cough, it was estimated that 93 cases were missed, leading to a corrected total of 413 confirmed cases. The relative vaccine efficacy estimated from the same model was 1.50 (95% confidence interval: 1.24-1.82), similar to the crude estimate of 1.59 and confirming better protection afforded by one of the two vaccines.

This method allows supporting analysis to interpret primary estimates of vaccine efficacy.