Alexander Paul E, Brito Juan P, Neumann Ignacio, Gionfriddo Michael R, Bero Lisa, Djulbegovic Benjamin, Stoltzfus Rebecca, Montori Victor M, Norris Susan L, Schünemann Holger J, Guyatt Gordon H
Health Research Methods (HRM), Department of Clinical Epidemiology and Biostatistics, McMaster University, Health Sciences Building (HSB), 1280 Main Street West, Hamilton, Ontario L8N 3Z5, Canada.
Knowledge and Evaluation Research Unit, Divisions of Endocrinology, Diabetes, Metabolism and Nutrition, Mayo Clinic, Plummer 3-35, 200 First Street SW, Rochester, MN 55905, USA.
J Clin Epidemiol. 2016 Apr;72:98-106. doi: 10.1016/j.jclinepi.2014.10.011. Epub 2014 Dec 19.
In 2007 the World Health Organization (WHO) adopted the GRADE system for development of public health guidelines. Previously we found that many strong recommendations issued by WHO are based on evidence for which there is only low or very low confidence in the estimates of effect (discordant recommendations). GRADE guidance indicates that such discordant recommendations are rarely appropriate but suggests five paradigmatic situations in which discordant recommendations may be warranted. We sought to provide insight into the many discordant recommendations in WHO guidelines.
We examined all guidelines that used the GRADE method and were approved by the WHO Guideline Review Committee between 2007 and 2012. Teams of reviewers independently abstracted data from eligible guidelines and classified recommendations either into one of the five paradigms for appropriately-formulated discordant recommendations or into three additional categories in which discordant recommendations were inconsistent with GRADE guidance: 1) the evidence warranted moderate or high confidence (a misclassification of evidence) rather than low or very low confidence; 2) good practice statements; or 3) uncertainty in the estimates of effect would best lead to a conditional (weak) recommendation.
The 33 eligible guidelines included 160 discordant recommendations, of which 98 (61.3%) addressed drug interventions and 132 (82.5%) provided some rationale (though not entirely explicit at times) for the strong recommendation. Of 160 discordant recommendations, 25 (15.6%) were judged consistent with one of the five paradigms for appropriate recommendations; 33 (21%) were based on evidence warranting moderate or high confidence in the estimates of effect; 29 (18%) were good practice statements; and 73 (46%) warranted a conditional, rather than a strong recommendation.
WHO discordant recommendations are often inconsistent with GRADE guidance, possibly threatening the integrity of the process. Further training in GRADE methods for WHO guideline development group members may be necessary, along with further research on what motivates the formulation of such recommendations.
2007年，世界卫生组织（WHO）采用了GRADE系统来制定公共卫生指南。此前我们发现，WHO发布的许多强烈推荐是基于对效应估计仅有低或极低置信度的证据（不一致的推荐）。GRADE指南指出，此类不一致的推荐很少是合适的，但提出了五种典型情况，在这些情况下不一致的推荐可能是合理的。我们试图深入了解WHO指南中众多不一致的推荐。
我们审查了2007年至2012年间所有采用GRADE方法并经WHO指南审查委员会批准的指南。审查团队独立从符合条件的指南中提取数据，并将推荐分为适当地制定不一致推荐的五种范式之一，或分为不一致推荐与GRADE指南不一致的另外三类：1）证据保证中等或高置信度（证据的错误分类）而非低或极低置信度；2）良好实践声明；或3）效应估计的不确定性最好导致有条件（弱）推荐。
33项符合条件的指南包括160项不一致的推荐，其中98项（61.3%）涉及药物干预，132项（82.5%）为强烈推荐提供了一些理由（尽管有时并不完全明确）。在160项不一致的推荐中，25项（15.6%）被判定与适当推荐的五种范式之一一致；33项（21%）基于对效应估计保证中等或高置信度的证据；29项（18%）是良好实践声明；73项（46%）保证有条件而非强烈推荐。
WHO不一致的推荐往往与GRADE指南不一致，可能威胁到该过程的完整性。可能需要对WHO指南制定小组成员进行GRADE方法的进一步培训，同时需要进一步研究促使制定此类推荐的因素。
