Guyatt Gordon, Schandelmaier Stefan, Brignardello-Petersen Romina, De Beer Hans, Prasad Manya, Murad M Hassan, Eachempati Prashanti, Chu Derek K, D'Souza Rohan, Iorio Alfonso, Agoritsas Thomas, Yao Liang, Mustafa Reem A, Parpia Sameer, Santaguida Pasqualina, Vandvik Per Olav, Hultcrantz Monica, Montori Victor M
Department of Health Research Methods, Evidence, and Impact, McMaster University, Hamilton, ON, Canada
Department of Medicine, McMaster University, Hamilton, ON, L8S 4L8, Canada.
BMJ. 2025 May 6;389:e081905. doi: 10.1136/bmj-2024-081905.
This third article in a seven part series presents the Core GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach to deciding whether to rate down certainty of evidence due to inconsistency—that is, unexplained variability in results across studies. For binary outcomes in which relative effects are consistent across baseline risks while absolute effects are not, Core Grade users assess consistency in relative effects. For continuous outcomes, they assess consistency in the absolute effects. When planning for the possibility of inconsistent results across studies, systematic review authors using Core GRADE construct a priori hypotheses regarding population or intervention characteristics that may explain inconsistency. They then judge the magnitude of inconsistency by considering the extent to which point estimates differ and the degree to which confidence intervals overlap. Before making a decision on rating down, Core GRADE users will evaluate where individual study estimates lie in relation to the threshold of the certainty rating (minimal important difference or the null). Finally, they will test their subgroup hypothesis and if an effect proves credible will provide separate evidence summaries and rate certainty of evidence separately for each subgroup. When they find no credible subgroup effect, they will provide a single evidence summary, rating down for inconsistency if necessary.
本系列七篇文章中的第三篇介绍了核心GRADE（推荐分级评估、制定与评价）方法，用于判断是否因结果不一致（即不同研究结果中无法解释的变异性）而降低证据的确定性。对于二元结局，若相对效应在不同基线风险水平下一致但绝对效应不一致，核心GRADE的使用者会评估相对效应的一致性。对于连续性结局，他们会评估绝对效应的一致性。在规划研究结果可能出现不一致的情况时，使用核心GRADE的系统评价作者会针对可能解释不一致性的人群或干预特征构建先验假设。然后，他们通过考虑点估计值的差异程度和置信区间的重叠程度来判断不一致性的大小。在做出降低证据等级的决定之前，核心GRADE的使用者会评估各个研究估计值相对于确定性评级阈值（最小重要差异或无效值）的位置。最后，他们会检验其亚组假设，如果效应被证明可信，将为每个亚组分别提供证据总结并分别评定证据的确定性。当他们未发现可信的亚组效应时，将提供单一的证据总结，必要时因不一致性而降低证据等级。
