Increasingly, multiple studies relating biomarkers to cancer and other health outcomes are pooled to obtain an overall risk profile, and a major challenge of pooling biomarker data is potential sources of variability of the biomarker data, including assay and laboratory variability. Currently there are no reliable and well-evaluated statistical methods to conduct the aggregated analysis for pooled biomarker data while taking care of the calibration process that correct for the between-study biomarker variability. In this proposal, we will develop efficient statistical methods for incorporating the calibration process in the aggregated data analysis. User- friendly software implementing the methods will be made publicly available. In addition, analysis results have potential to be substantially different between using the two commonly used methods for analyzing pooled data, the two-stage analysis method and the aggregated data analysis method, and in the two-stage method, between the fixed effect model method and the random effect model method. Investigators conducting consortial research are confronted with the choice between the methods. We will compare these methods such that the choices of analysis methods will be made to exploit the full power of the data available to maximize the information gained, while at the same time only making minimum and realistic assumptions.
Increasingly, multiple studies relating biomarkers to cancer and other health outcomes are pooled to obtain an overall risk profile, and a major challenge of pooling biomarker data is potential sources of variability of the biomarker data, including assay and laboratory variability. In this proposal, we will develop reliable statistical methods for incorporating the calibration process that corrects the between-study biomarker variability in the aggregated data analysis, and the user-friendly software implementing the new methods will be publicly available. Analysis results have potential to be substantially different between the two common practices, the two-stage analysis and the aggregated data analysis, and in the two-stage analysis, between the fixed effect and the random effect meta analysis models, and we will compare these methods so that the choice of analysis method will be made to exploit the full power of the data available to maximize the information gained, while at the same time only making minimum and realistic assumptions.
