Inference using calibration samples

This project refines methods that enable overall conclusions to be reached from the analysis of multiple studies where some of the key measurements have been made in different ways. The methods are developed and compared in a study of general population cohorts, a clinical cohort, and two randomised trials.

It is commonplace for researchers to want to undertake analysis of several studies combined together but are prevented from doing so by the differences in measurement protocols used. We examine various methods (structural equation modelling, multiple imputation and meta-regression) for overcoming this problem by the use of internal or external cross-measure calibration samples. We will highlight differences among the methods and designs in the assumptions made, and in the power and robustness of the conclusions reached. The methods are developed and compared in a study of general population cohorts, a clinical cohort, and two randomised trials.