Problems of the use of the trend test as a dose-response measurement in epidemiology.
Gaceta sanitaria / S.E.S.P.A.S, 11 (3):
143-9 (1997)3676<m:linebreak></m:linebreak>Mesures d'associació.Abstract
BACKGROUND: The objective of the dose-response analysis is to reveal the presence of a progressively response to an increased dose. One of the usual ways of doing so is the carrying out of the so-called trend test. However, the fact that this test may show statistical significance does not necessarily indicate that there is a uniform response between dose and response. This study aims to examine the various situations in which the use of the trend test is not always appropriate as well as given an overview of the use of potential alternatives. METHODS: Five theoretical dose-response scenarios have been created using for supposed case-control studies, in which the relationship was not always linear. The results-tables obtained have been analysed, leading to the attainment of the Chi-squared trent test, as well as the calculation of the linear component by means of the logistical regression. Subsequently, other alternative methods of trends analysis, such as categorical analysis, the transformation of the findings within the logistical model and the use of risk incrementals have been employed. RESULTS: The test of trends invariably proved to be significant, despite the fact that in four of the theoretical models there did not exist a clear relationship between dose and response. This significant result could lull its readers into proposing a linear dose-response relationship which, in fact, proves false. The alternative use of incremental risk models, or categorical analysis, permits the identification of situations in which there is no relationship. CONCLUSIONS: The trend test should only be employed in cases where the data demonstrates the existence of monotonous, lineal relationship that can be deduced from the results of categorical analysis, the use of incremental risk or the adjustment of various no linear models, although, in this cases, the significance of the test may bring forward no new information.