Open AccessNotes on the Assessment of Trend in the Presence of Nondifferential Exposure Misclassification
Author(s) -
Hermann Brenner
Publication year - 1992
Publication title -
epidemiology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.901
H-Index - 173
eISSN - 1531-5487
pISSN - 1044-3983
DOI - 10.1097/00001648-199209000-00007
Subject(s) - assertion , statistics , econometrics , disease , association (psychology) , medicine , demography , environmental health , psychology , mathematics , pathology , computer science , programming language , sociology , psychotherapist
Estimation of overall linear trend and statistical tests for trend are commonly applied in epidemiologic studies to evaluate the association between ordinal exposure variables and dichotomous health outcomes. In the discussion of these studies, the assertion is commonly made that the observed trends can only be conservative and reported P-values can only be too high owing to the occurrence of exposure misclassification that is supposed to be nondifferential, that is, unrelated to the health outcome of interest. This paper illustrates that this assertion is not generally justified without additional sensitivity analyses. Overestimation of trends and an erroneous reduction of P-values due to nondifferential exposure misclassification are unlikely, but not impossible, for monotonic exposure-disease associations. These may be common occurrences, however, if the observed exposure-disease association is nonmonotonic.