Matok Ilan, Azoulay Laurent, Yin Hui, Suissa Samy
Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montreal; Department of Epidemiology, Biostatistics and Occupational Health McGill University, Montreal.
Birth Defects Res A Clin Mol Teratol. 2014 Sep;100(9):658-62. doi: 10.1002/bdra.23271. Epub 2014 Jul 1.
The use of decongestants during the second or third trimesters of pregnancy has been associated with a decreased risk of preterm delivery in two observational studies. This effect may have been subject to immortal time bias, a bias arising from the improper classification of exposure during follow-up. We illustrate this bias by repeating the studies using a different data source.
The United Kingdom Hospital Episodes Statistics and the Clinical Practice Research Datalink databases were linked to identify all live singleton pregnancies among women aged 15 to 45 years between 1997 and 2012. Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals of preterm delivery (before 37 weeks of gestation) by considering the use of decongestants during the third trimester as a time-fixed (biased analysis which misclassifies unexposed person-time as exposed person-time) and time-varying exposure (unbiased analysis with proper classification of unexposed person-time). All models were adjusted for maternal age, smoking status, maternal diabetes, maternal hypertension, preeclampsia, and parity.
Of the 195,582 singleton deliveries, 10,248 (5.2%) were born preterm. In the time-fixed analysis, the HR of preterm delivery for the use of decongestants was below the null and suggestive of a 46% decreased risk (adjusted HR = 0.54; 95% confidence interval, 0.24-1.20). In contrast, the HR was closer to null (adjusted HR = 0.93 95% confidence interval, 0.42-2.06) when the use of decongestants was treated as a time-varying variable.
Studies of drug safety in pregnancy should use the appropriate statistical techniques to avoid immortal time bias, particularly when the exposure occurs at later stages of pregnancy.
两项观察性研究表明，在妊娠中期或晚期使用减充血剂与早产风险降低有关。这种效应可能存在不朽时间偏倚，这是一种因随访期间暴露分类不当而产生的偏倚。我们通过使用不同的数据源重复这些研究来说明这种偏倚。
将英国医院事件统计数据和临床实践研究数据链数据库相链接，以识别1997年至2012年期间15至45岁女性中的所有单胎活产妊娠。通过将妊娠晚期使用减充血剂视为固定时间暴露（将未暴露的人时错误分类为暴露人时的有偏分析）和随时间变化的暴露（正确分类未暴露人时的无偏分析），使用Cox比例风险模型估计早产（妊娠37周前）的风险比（HR）和95%置信区间。所有模型均针对产妇年龄、吸烟状况、产妇糖尿病、产妇高血压、先兆子痫和产次进行了调整。
在195,582例单胎分娩中，10,248例（5.2%）为早产。在固定时间分析中，使用减充血剂导致早产的HR低于无效值，提示风险降低46%（调整后HR = 0.54；95%置信区间，0.24 - 1.20）。相比之下，当将使用减充血剂视为随时间变化的变量时，HR更接近无效值（调整后HR = 0.93；95%置信区间，0.42 - 2.06）。
妊娠药物安全性研究应使用适当的统计技术以避免不朽时间偏倚，尤其是当暴露发生在妊娠后期时。
