Enhancing Exposure Assessment: A Hybrid Within-Person Biospecimen Pooling Approach to Reduce Biomarker Variability and Cost

ABSTRACT Objective: Biomarkers are crucial for exposure assessment in environmental research, but single biospecimen measurements can be unreliable due to high intra-individual variability (IIV). Repeated within-person sampling improves reliability but is costly and impractical. The hybrid pooled-unpooled design, which pools biospecimens from individuals with varying pooled sizes, offers a cost-effective alternative but fails when pooling error (PE) exceeds IIV. We developed a regression calibration method to correct exposure effect estimates (βs) if post-experiment variance components reveal excessive PE. Material and Methods: This study, motivated on phenol measurements in the EAGeR trial where PE > IIV. A calibration factor was developed to correct for both IIV and PE, ensuring accurate β estimation and its 95% confidence intervals (CI). A simulation (1000 runs, n=1000, 10 repeated samples) was conducted with varying pooling sizes and three error scenarios (PE>IIV, PE<IIV, PE=IIV), assuming the normally distributed errors are independent with constant variance. We assessed the design’s robustness by comparing bias and other metrics for uncalibrated versus calibrated β estimates in a binary outcome model, with 95% CI coverage. Results: Uncalibrated estimates of β, were consistently underestimated across all scenarios, with percent bias ranging from 51.3-66.8%. Bias was most severe when PE > IIV or when both were similar (PE ≈ IIV). After applying calibration, bias improved substantially, reducing to a range of 7.8%-20.7%, with the highest residual bias observed when PE ≈ IIV. CI coverage was within acceptable limits, close to 95%. Conclusion: Hybrid within-person biospecimen pooling with regression calibration is a cost-effective method for estimating β when biomarkers are variable and costly. Its accuracy depends on the relationship between PE and IIV—unknown before pooling. If PE is excessive, bias can occur, but our method provides a post-experiment correction, making it ideal for studies with high biomarker variability, like environmental phenols.