Bias from missing values: sex differences in implication of failed venepuncture for the Scottish Heart Health Study

Abstract

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 10,359 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. Missing blood samples are shown to be associated with increased body mass index and difficult peripheral veins. The bias caused by the missing female blood samples is considerable, with estimated errors of between 4% and 12% in the mean values of total cholesterol, HDL-cholesterol, triglycerides, creatinine, uric acid and fibrinogen.