Brazzelli Miriam, Lewis Stephanie C, Deeks Jonathan J, Sandercock Peter A G
Division of Clinical Neurosciences, University of Edinburgh, Western General Hospital, Edinburgh, UK.
J Clin Epidemiol. 2009 Apr;62(4):425-30. doi: 10.1016/j.jclinepi.2008.06.018. Epub 2008 Nov 14.
There is little empirical evidence on publication bias in diagnostic test accuracy studies. We evaluated the proportion of abstracts presented at international stroke meetings, which were later published in full, and investigated which study features characterized publication.
We reviewed all diagnostic abstracts presented at two international stroke conferences between 1995 and 2004. We assessed the characteristics and findings of the identified abstracts. We identified full publications through electronic databases and by contacting the authors. Determinants of publication were assessed by Cox regression.
Seventy-six percent (121 out of 160) of identified abstracts were subsequently published in full. Sixty-two percent were published within 24 months of presentation. The median time to publication was 16 months. Assessment of interobserver agreement between test readers was a significant predictor of full publication (P=0.02). No other study characteristic (including clinical utility of results, multicenter status, or Youden's index) was predictive.
We found no clear evidence of bias in the publication process that occurs after abstract acceptance. We were unable to assess bias in abstract submission or acceptance. "Interobserver agreement" was the only characteristic statistically associated with publication. Clinical utility of results and other study characteristics did not predict publication. Diagnostic abstracts often did not report many relevant methodological aspects.
关于诊断试验准确性研究中的发表偏倚，实证证据很少。我们评估了在国际卒中会议上发表的摘要后来全文发表的比例，并调查了哪些研究特征决定了发表情况。
我们回顾了1995年至2004年间在两次国际卒中会议上发表的所有诊断摘要。我们评估了所确定摘要的特征和研究结果。我们通过电子数据库和联系作者来确定全文发表情况。通过Cox回归评估发表的决定因素。
76%（160篇中的121篇）所确定的摘要随后全文发表。62%在发表后24个月内发表。发表的中位时间为16个月。测试读者之间观察者间一致性的评估是全文发表的一个重要预测因素（P = 0.02）。没有其他研究特征（包括结果的临床实用性、多中心状态或尤登指数）具有预测性。
我们没有发现摘要被接受后发表过程中存在明显偏倚的证据。我们无法评估摘要提交或接受过程中的偏倚。“观察者间一致性”是与发表在统计学上唯一相关的特征。结果的临床实用性和其他研究特征并不能预测发表情况。诊断摘要往往没有报告许多相关的方法学方面。
