Cautions on the use of medical databases

Abstract

The use of medical databases is tied in with the conduct of epidemiologic and clinical investigations. I expound the thesis that investigative results are frequently ambiguous and subject to a multiplicity of interpretations, some even diametrically opposed, and describe examples of such situations. For instance, the very fact that an agent is particularly safe may make it seem dangerous--the agent becomes the one of choice for high risk patients. Other topics include the use of models, including regression models, Cox's proportional hazard model, the search for epidemiologic interactions, ratio measures, and grouping effects.