Dane Van Domelen
[email protected]
2019-02-05
Pooling refers to the process of measuring a biomarker in combined samples (e.g. blood, saliva) from multiple subjects rather than for each subject separately. The main benefit is drastically better statistical efficiency in certain scenarios, particularly when the per-assay cost is high relative to other per-subject costs.
Although not used very frequently, one high-profile application of pooling was the National Health and Nutrition Examination Survey (NHANES), where 8-specimen pools were used in 2005-2006 to reduce the number of assays by nearly 90% (saving $2.78 million) (Caudill 2012).
The purpose of the pooling package is to provide tools for designing pooling studies (e.g. choosing the pool size and number of assays) and for analyzing data from them.
You can install pooling from CRAN:
install.packages("pooling")
Or from GitHub:
library("devtools")
install_github("vandomed/pooling")
Pooling works great in the two-sample t-test scenario, because it reduces the variance of each observation from to
, where
is the pool size.
The poolcost_t function lets you plot total study costs vs. pool size. To illustrate, suppose you want 80% power to detect a difference in group means of 0.25, and the biomarker has variance 1. If it costs $100 per assay and $10 in other per-subject recruitment costs, this is what it looks like:
p <- poolcost_t(d = 0.25, sigsq = 1, assay_cost = 100, other_costs = 10)
p
Caudill, Samuel P. 2012. “Use of Pooled Samples from the National Health and Nutrition Examination Survey.” Statistics in Medicine 31 (27). Wiley Online Library: 3269–77.