Missing values are common in epidemiological data, and yet their possible effect on the results of the investigation is seldom quantified despite the fact that they are a likely source of bias.
This paper describes a simple method, based on odds ratios, for assessing whether missing values are likely to cause bias, and for quantifying the magnitude of any such bias.
The method is applied to the Scottish Heart Health Study, a study of risk factors for coronary heart disease amongst 10359 men and women.
It is found that, although no bias is apparent in the male subjects females with missing blood samples are twice as likely to have a history of myocardial infarction than other women.
Mots-clés Pascal : Grande Bretagne, Royaume Uni, Europe, Epidémiologie, Biais méthodologique, Méthodologie, Donnée manquante, Prélèvement, Homme, Facteur risque, Infarctus, Myocarde, Sexe, Cardiopathie coronaire, Appareil circulatoire pathologie, Ecosse
Mots-clés Pascal anglais : Great britain, United Kingdom, Europe, Epidemiology, Methodological bias, Methodology, Missing data, Sampling, Human, Risk factor, Infarct, Myocardium, Sex, Coronary heart disease, Cardiovascular disease, Scotland
Notice produite par :
Inist-CNRS - Institut de l'Information Scientifique et Technique
Cote : 91-0580784
Code Inist : 002B30A. Création : 199406.