Empirical validation of a within-subject pooling approach to improve accuracy of estimation of exposure to biomarkers with strong temporal variations