|
Abstract:
|
Human biomonitoring aims to monitor human health conditions by measuring the accumulation of harmful chemicals from specimens such as blood samples. To save costs, researchers may mix the specimens physically and then measure the concentration level of toxic substances from the pool. Despite the advantages, the pooling strategy reduces the number of observations since it measures the pools rather than each specimen individually. Thus, when fixing the number of specimens, statistical models designed for the pooled data have smaller sample sizes and are expected to be less efficient than individual-level data. Many existing methods have suffered from this loss of efficiency. This work will introduce a new method to reconstruct the individual-level information from the pooled observations iteratively. We focus on an additive partially linear model and show that the proposed model achieves comparable performance when individual data is available, particularly for small pool sizes. We use simulations to assess our method and illustrate the proposed models through a human monitoring study of brominated flame retardant from the National Health and Nutrition Examination Survey (NHANES).
|
