The use of bivariable selection (BVS) for selecting variables to be used in multivariable analysis is inappropriate despite its common usage in medical sciences.
In BVS, if the statistical p value of a risk factor in bivariable analysis is greater than an arbitrary value (often p=0.05), then this factor will not be allowed to compete for inclusion in multivariable analysis.
This type of variable selection is inappropriate because the BVS method wrongly rejects potentially important variables when the relationship between an outcome and a risk factor is confounded by any confounder and when this confounder is not properly controlled.
This article uses both hypothetical and actual data to show how a nonsignificant risk factor in bivariable analysis may actually be a significant risk factor in multivariable analysis if confounding is properly controlled.
Furthermore, problems resulting from the automated forward and stepwise modeling with or without the presence of confounding are also addressed.
To avoid these improper procedures and deficiencies, alternatives in performing multivariable analysis, including advantages and disadvantages of the BVS method and automated stepwise modeling, are reviewed and discussed.
Mots-clés Pascal : Analyse multivariable, Facteur risque, Critère sélection, Variable, Analyse statistique, Méthodologie, Pronostic, Analyse régression, Analyse 2 variables, Biais
Mots-clés Pascal anglais : Multivariate analysis, Risk factor, Selection criterion, Variable, Statistical analysis, Methodology, Prognosis, Regression analysis, Bias
Notice produite par :
Inist-CNRS - Institut de l'Information Scientifique et Technique
Cote : 96-0461686
Code Inist : 002B30A01A1. Création : 10/04/1997.