Modern Decision Methodology in Radiochemical Testing

Oral Presentation

Prepared by T. Semkow, C. Bradt
New York State Department of Health, Wadsworth Center, Empire State Plaza, Albany, NY, 12201, United States

Contact Information:; 518-474-6071


Radiochemical testing is concerned with the detection of radioactive contaminants in the environment. A question facing an analyst is: is it better to report the result with the uncertainty as is or is it better to censor it? The answer depends on the intended use of the data. From the point of view of risk assessment it is desirable to censor the result by determination with a given and high statistical significance whether the radioactive contaminant is detected or not. This is particularly important with respect to disposing of long-lived artificial radionuclides into environment. Traditional decision methodology in radiochemical testing is based on the decision level and detection limit. If a radionuclide is found not to be detected, an upper limit is set. We extend this concept to three regions: not detected, no decision, and detected. This can be explained by first assuming that statistical fluctuations are absent, and then adding the fluctuations. The complexity of mathematical treatment depends on the choice of model distribution. If Gaussian statistics is assumed, then the formulas contain simple algebra, however complicated. With the Gaussian statistics, this methodology can be applied to any contaminant testing in addition to radiochemical. For radiochemical testing, however, the statistics of radiation counting is that of a Poisson, which can be approximated by a Gaussian at high counting rates only. Working with the Poisson distribution requires calculation of infinite sums or integrals of special functions. A satisfactory solution to this problem is a goal of continuing research, the progress of which will be reported. The basic decision methodology in radiochemical testing obtains net signal by subtracting a background from the gross signal (sample plus background). Extensions are described when a gamma peak is measured in the gamma-ray spectrum of the sample, whereas the background contains an interfering peak. Advanced topics are discussed such as propagating of calibration coefficient uncertainty into decision methodology and handling of overdispersed statistics.