Statistical Analysis Of Biomarker Data Combined From Multiple Cohort Studies

Combining biomarker data from multiple cohort studies could promote the estimation of biomarker-disease association by increasing sample size.However,the measurements of the biomarker can be different among laboratories,and we usually demand a calibration to a reference labtoratory before aggregating.Former research articles treated the biomarker measurements from the reference laboratory as the gold standard,whereas the biomarker measurements from the reference laboratory are not certainly equal to the true value.We develop two statistical methods for combined biomarker data from multiple cohort studies,the cut-off calibration and exact calibration methods,which no longer consider the biomarker measurements from any laboratory as the underlying truth.Our findings include the following items: Firstly,the cut-off calibration method provides estimators with minimal bias and valid confidence intervals under weak biomarker-disease associations or small measurement errors.Secondly,comparing with the cut-off calibration method,the exact calibration method presents more precise confidence intervals and lesser biased estimators.Thirdly,Controls-only calibration design may introduce additional bias,however,the bias is relative small for rare disease prevalences or weak biomarker-disease associations.