Confidence in polymerase chain reaction diagnosis can be improved by Bayesian estimation of post-test disease...

Polymerase chain reaction (PCR) techniques allow highly sensitive detection of specific DNA for diagnosis of infectious and genetic disease, but uncertainty relating to sensitivity and contamination has frequently resulted in controversy over results. We propose a new design in which the PCR contamination rate is estimated experimentally. The sensitivity of duplicate test results, and hence the post-test disease probabilities, can be derived algebraically, but wide confidence limits around these point estimates reduce their usefulness.

We have developed a Bayesian method which gives better estimates of post-test disease probability and can substantially reduce uncertainty by using the prior belief that sensitivity is not lower than 90%.

With 100 duplicate test samples and 100 control samples, we find that the post-test disease probability for concordant results (both positive or both negative) is generally unequivocal. The post-test disease probability for discordant results (one test positive and one negative) is often sufficiently clear to allow useful interpretation of individual test results, depending on the context.

Using this approach, the performance of a PCR can be evaluated experimentally allowing post-test disease probability to be estimated, giving improved confidence in test results.