Methods/Statistics
Within-person pooling of biospecimens, a solution to highly variable expensive biomarkers: practical or idealistic? Yajnaseni Chakraborti* Yajnaseni Chakraborti Enrique F. Schisterman Sunni L. Mumford Aijun Ye Stefanie N. Hinkle
Background: Biomarkers are crucial for measuring exposure in environmental and nutritional sciences but can be costly. Pooling biospecimens across individuals is cost-effective but leads to biased exposure effect estimates (β ) when target biomarkers have high temporal variability. A novel study design combined with a regression calibration approach is needed to address this issue, as repeated within-person samples for the whole study cohort are not always feasible.
Methods: We propose a hybrid pooled-unpooled study design, where biomarkers are repeatedly measured in a small subset of the cohort, while in the rest, biomarkers are measured from biospecimens that are pooled together before assay. To address any remaining measurement error (ME) and resulting pooling error (PE) from this pooling design, a correction factor was derived to obtain the bias-corrected estimate of the exposure effect (βc), along with the bias-corrected 95% CI. A simulation study (1000 runs, n=1088 with 10 repeated samples) was conducted with varying pooling sizes, and three variability scenarios: 1) PE > ME, 2) PE < ME, and 3) PE = ME. We assessed the robustness of the design by comparing the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for uncorrected and bias-corrected estimates of β for a binary outcome, as well as the coverage of the 95% CIs.
Results: The MAE and RMSE for βc’s were lower compared to uncorrected β’s, across all scenarios of biomarker variability. Bias from not calibrating was more severe when PE > ME and PE = ME, compared to when PE < ME (Figure). Coverage of the 95% CIs were acceptable.
Conclusion: The approach of hybrid within-person biospecimen pooling, and bias correction through regression calibration, provides a robust and cost-effective solution for measuring highly variable, expensive biomarkers in research areas where exposure misclassification is a concern (e.g., to study the effect of exposure to environmental toxicant(s) on likelihood of live birth).
