Sunday, December 1, 2013

Longitudinally Evaluated Disease Screening Technologies

I've recently had another journal article published. It is focused on the evaluation of disease screening technologies in longitudinal trials (ie. long running research trials which rely on a new technology for monitoring a population). As a technology researcher, I am particularly interested in this subject as I endeavor to develop new disease detection technologies which will be evaluated in longitudinal trials. The article was published in the Journal of Clinical and Diagnostic Research and you can access it here.

The article includes an interesting observation: that it is possible for two disease screening technologies to produce similar disease detection rates even when their respective sensitivities to disease are quite different from each other. This occurs because more sensitive technologies remove cases of disease from the longitudinally monitored population and thus have fewer cases to find in subsequent rounds of screening. There is an upper bound on the longitudinally measured disease detection rate that is well illustrated by a perfect test which catches every instance of disease in the population it monitors. In such a scenario the perfect test's disease detection rate is equal to the rate at which the disease enters the population being monitored. The results in the study demonstrate that even at sensitivities that are much lower than perfect, the test can produce a longitudinally measured disease detection rate that is still reasonably close to the rate at which the disease enters the population. This is due to the fact that the poorer test's monitored population contains disease not found in previous rounds of screening, thus even though the poorer test has a much weaker sensitivity to the disease, it has a larger amount of disease in the population it monitors and thus produces a higher disease detection rate than might have been expected.

If you are interested in reading about this effect further, please read my article: