Spline Analysis of Biomarker Data Pooled From Multiple Matched/Nested Case-Control Studies
share
Pooling biomarker data across multiple studies enables researchers to get more precise estimates of the association between biomarker exposure measurements and disease risks due to increased sample sizes. However, biomarker measurements vary significantly across different assays and laboratories, and therefore calibration of the local laboratory measurements to a reference laboratory is necessary before pooling data. We propose two methods that can estimate a nonlinear relationship between biomarker exposure measurements and disease risks using spline functions with a nested case-control study design: full calibration and internalized calibration. The full calibration method calibrates all observations using a study-specific calibration model while the internalized calibration method only calibrates observations that do not have reference laboratory measurements available. We compare the two methods with a naive method whereby data are pooled without calibration. We find that: (1) Internalized and full calibration methods have substantially better performance than the naive method in terms of average relative bias and coverage rate. (2) Full calibration is more robust than internalized calibration when the size of calibration subsets varies. We apply our methods to a pooling project with nested case-control study design to estimate the association of circulating Vitamin D levels with the risk of colorectal cancer.
READ FULL TEXT