Workshop Review by Gary Kramer - Bayesian Statistical Methods for Bioassay, Radiochemistry, and Internal Dosimetry
Excerpts from "43rd Conference on Bioassay, Analytical and Environmental Radiochemistry" by Gary H. Kramer, Ottawa, Canada, published in the Health Physics Society Newsletter, pp. 21-22, January, 1998
"The [43rd Conference on Bioassay, Analytical and Environmental Radiochemistry] concluded with a two-day workshop on Bayesian statistics and the application to bioassay. The first day consisted of an introduction to the subject by Donald Berry, Duke University. The second day was devoted to the application of Bayesian statistics in bioassay measurements. A fundamental part of Bayesian statistics is that the probability of an intake in a worker population must be known (or assumed) before further calculations can continue. One difficulty that was never satisfactorily resolved was the inability of Bayesian statistics to deal with negative numbers (sometimes a background count is higher than a sample count). The determination of the lower limit of detection (LD) was discussed in detail and while Bayesian statistics can reduce the false positives (samples without activity that appear to contain activity), it does so at the expense of increasing the false negatives (samples with activity that are missed). The value of the decision level can vary by a factor of four (at 3 sigma) depending on what assumption one makes about the worker population.
"The workshop concluded with a panel discussion. There was not enough time to discuss all the issues raised in the earlier sessions-another day would have been needed. Consensus was not reached concerning the applicability of the Bayesian statistics technique (despite the fact that Los Alamos National Laboratory has been using for some time now). It was voiced that the concept of prior belief influencing the value of LD is foreign and viewed as circular reasoning. Some participants felt uncomfortable about a mathematical analysis that changes the outcome based on subjective assumptions about intake probabilities; nevertheless, the workshop was an excellent experience."
Return to: PNNL Bayesian homepage
Contact: Dan Strom
Read: Disclaimer
Revised: March 6, 1998