系统综述和叙述性综述引导专家对癌症试验做出不同的预测：一项随机试验。

Systematic review and narrative review lead experts to different cancer trial predictions: a randomized trial.

机构信息

Biomedical Ethics Unit, McGill University, 3647 Peel Street, Montréal, Quebec H3A 1X1, Canada.

Clinical Epidemiology Program, Ottawa Hospital Research Institute, 501 Smyth Road, Ottawa, Ontario K1Y 4E9, Canada.

出版信息

J Clin Epidemiol. 2021 Apr;132:116-124. doi: 10.1016/j.jclinepi.2020.12.006. Epub 2020 Dec 17.

DOI:10.1016/j.jclinepi.2020.12.006

PMID:33346013

Abstract

OBJECTIVES

The objective of the study was to assess the impact of narrative review (NR) vs. systematic review (SR) on expert assessments of a clinical trial.

STUDY DESIGN AND SETTING

Experts in colon and rectal surgery were randomized to read an NR or SR for an ongoing clinical trial involving surgery for colorectal cancer. Experts from the United States and Canada completed online or paper surveys between December 2018 and June 2019. After reading the NR or SR, experts predicted the trial's outcome and evaluated the trial under a hypothetical ethical review.

RESULTS

Experts who read the NR (n = 55) compared with those who read the SR (n = 56) were more likely to predict a higher absolute risk reduction, 58% vs. 33%, P = 0.018, mean predictions 10.6% vs. 6.6%, mean difference 4.0% [95% confidence interval: 0.3%, 7.8%]. Experts who read the NR were more likely to evaluate the trial more favorably under a hypothetical ethical review, 48% vs. 26%, P = 0.039, 20.0% vs. 8.9% "strongly in favor" of trial being pursued.

CONCLUSION

An NR and an SR for the same trial led to different judgments of likely outcomes and ethical appropriateness. These differences point to a potential source of unaddressed bias in ethical review.

摘要

目的

本研究旨在评估叙述性综述（NR）与系统性综述（SR）对临床试验专家评估的影响。

研究设计和设置

结直肠外科专家被随机分配阅读正在进行的涉及结直肠癌手术的临床试验的 NR 或 SR。美国和加拿大的专家在 2018 年 12 月至 2019 年 6 月期间在线或通过纸质问卷完成调查。在阅读 NR 或 SR 后，专家预测试验结果，并在假设的伦理审查下评估试验。

结果

与阅读 SR 的专家（n=56）相比，阅读 NR 的专家（n=55）更有可能预测更高的绝对风险降低，分别为 58%和 33%，P=0.018，平均预测值分别为 10.6%和 6.6%，平均差值为 4.0%[95%置信区间：0.3%，7.8%]。阅读 NR 的专家在假设的伦理审查下更有可能评估试验更有利，分别为 48%和 26%，P=0.039，20.0%和 8.9%“强烈支持”继续进行试验。