Statistical Methods for Pooled and Calibrated Biomarker Data

Abstract

Pooling biomarker exposure data across multiple studies in a participant level meta-analysis allows for examination of a wider exposure range than generally possible in individual studies, evaluation of population subgroups and disease subtypes with more statistical power, and more precise estimation of biomarker-disease associations. However, biomarker measurements often require calibration to a reference assay prior to pooling due to assay and laboratory variability across studies. In the following chapters, I develop and compare methods for pooled and calibrated biomarker data across multiple data settings. In Chapter 1, I develop two participant level meta-analysis methods for biomarker data pooled from multiple cohort studies and compare their performance to existing methods in regression calibration. Special consideration is given to the performance of these methods when the calibration data is derived from controls only or a random sample of cases and controls from each underlying cohort study. In Chapter 2, I extend these statistical methods to biomarker data combined from nested case-control studies when the reference assay data are obtained from a subset of controls in each contributing study. I also derive methodology to support inference for a biomarker-covariate interaction term under an aggregated approach or regression calibration approach. In Chapter 3, I propose a repeated measures method to calibrate biomarker measurements pooled from multiple cohort studies and compare this approach to the methods described in Chapter 1. The methods of each chapter are illustrated in applications evaluating the relationship between blood circulating vitamin D levels and various health outcomes in a pooling project of cohort studies.