Ng Shu-Kay, Scott Rani, Scuffham Paul A
School of Medicine and Menzies Health Institute Queensland, Griffith University, University Drive, Meadowbrook, QLD, 4131, Australia.
Matern Child Health J. 2016 Jul;20(7):1472-84. doi: 10.1007/s10995-016-1946-8.
Objective This research aims to identify predictors of attrition in a longitudinal birth cohort study in Australia and assess differences in baseline characteristics and responses in subsequent follow-up phases between contactable non-responders and uncontactable non-responders deemed "lost to follow-up (LTF)". Methods 3368 women recruited from three public hospitals in Southeast Queensland and Northern New South Wales during antenatal visits in 2006-2011 completed a baseline questionnaire to elicit information on multiple domains of exposures. A follow-up questionnaire was posted to each participant at 1 year after birth to obtain mother's and child's health and development information. Multivariate logistic regression was used to model the association between exposures and respondents' status at 1 year. The effect of an inverse-probability-weighting method to adjust for non-response was studied. Results Overall attrition at 1-year was 35.4 %; major types of attrition were "contactable non-response" (27.6 %) and "LTF" (6.7 %). These two attrition types showed different responses at the 3-year follow-up and involved different predictors. Besides shared predictors (first language not English, higher risk of psychological distress, had smoked during pregnancy, higher levels of family conflict), distinguishable predictors of contactable non-responders were younger age, having moved home in the past year and having children under 16 in the household. Attrition rates increased substantially from 20 % in 2006 to 54 % in 2011. Conclusions This observed trend of increased attrition rates raises concern about the use of traditional techniques, such as "paper-based" questionnaires, in longitudinal cohort studies. The supplementary use of electronic communications, such as online survey tools and smart-device applications, could provide a better alternative.
目的 本研究旨在确定澳大利亚一项纵向出生队列研究中失访的预测因素，并评估在后续随访阶段中，可联系上但未应答者与被视为“失访（LTF）”的无法联系上的未应答者在基线特征和应答方面的差异。方法 2006年至2011年期间，从昆士兰州东南部和新南威尔士州北部的三家公立医院招募了3368名女性，她们在产前检查时完成了一份基线问卷，以获取多领域暴露信息。产后1年向每位参与者发放一份随访问卷，以获取母亲和孩子的健康与发育信息。采用多变量逻辑回归模型来模拟暴露因素与1年后应答者状态之间的关联。研究了采用逆概率加权法调整无应答情况的效果。结果 1年时的总体失访率为35.4%；主要失访类型为“可联系上但未应答”（27.6%）和“失访”（6.7%）。这两种失访类型在3年随访时表现出不同的应答情况，且涉及不同的预测因素。除了共同的预测因素（母语非英语、心理困扰风险较高、孕期吸烟、家庭冲突程度较高）外，可联系上但未应答者的显著预测因素为年龄较小、过去一年搬过家以及家中有16岁以下子女。失访率从2006年的20%大幅上升至2011年的54%。结论 观察到的失访率上升趋势引发了对纵向队列研究中使用传统技术（如“纸质”问卷）的担忧。补充使用电子通信手段，如在线调查工具和智能设备应用程序，可能会提供更好的选择。
