Beware of registries for their biases.

Abstract

Patient registries are very popular. On the other hand, scientific data collections in registries are commonly observational and retrospective and, in many instances, are prone to biases. Same thing is true of administrative data bases. The selection of the control group(s) is probably the Achilles heel of scientific data collection in observational studies, and there are historical examples of how a properly chosen control group can help or its absence deceive us. Somewhat more recently recognized biases are the wandering comparisons of risk, confounding by disease severity, channeling bias, depletion of the susceptible, and the immortal time bias. The last bias can especially be deceiving and give us false hopes of new remedies. A particularly important selection bias we have come across is what we call the "mortality bias." This is where the mortality in a mother population lessens the mortality in the registry that stems from this mother population simply because deaths in the former cannot be represented in the latter.