Immortal time bias in the study of stillbirth risk factors: the example of gestational diabetes.
Epidemiology (Cambridge, Mass.), 24 (6):
787-90 (November 2013)7501<br/>JID: 9009644; ppublish;<br/>Mesures d'associació; Immortal time bias.DOI: 10.1097/EDE.0b013e3182a6d9aa
Abstract
BACKGROUND: Current understanding of the increased risk for stillbirth in gestational diabetes mellitus is often based on large cohort studies in which the risk of stillbirth in women with this disease is compared with the risk in women without. However, such studies could be susceptible to immortal time bias because, although many cohorts begin at 20 weeks' gestation, pregnancies must "survive" until 24-28 weeks in order to be screened and diagnosed with gestational diabetes. METHODS: We describe the theoretical potential for immortal time bias in studies of stillbirth and gestational diabetes and then quantify the magnitude of the bias using 2006 United States vital statistics data. RESULTS: Although gestational diabetes was protective against stillbirth when including all births (relative risk = 0.88 95% confidence interval = 0.79-0.99), restricting analyses to births at >28 weeks' gestation reversed the effect and diabetes became associated with an increased risk of stillbirth (1.25 1.11-1.41). CONCLUSION: Immortal time before diagnosis of gestational diabetes may bias our understanding of the stillbirth risk associated with this condition.