The use of GRADE methods in World Health Organization (WHO) public health guidelines: distributions of strength of recommendations and confidence in estimates of effect

Article type
Authors
Alexander PE1, Norris S2, Brito JP3, Stolzfus RJ4, Bero L5, Djulbegovic B6, Neumann I7, Montori VM3, Guyatt GH1
1Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Canada
2World Health Association, Geneva, Switzerland
3Mayo Clinic, Rochester, Minnesota, USA
4Division of Nutritional Sciences, Cornell University, USA
5University of California, San Francisco, USA
6University of South Florida & H Lee Moffitt Cancer Center
7Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Canada AND Department of Internal Medicine, School of Medicine, Pontificia Universidad Catolica de Chile
Abstract
Background: A perception exists that expert guideline panelists are sometimes reluctant to offer weak/conditional/contingent recommendations. GRADE guidance warns against strong recommendations when confidence in estimates of effect (quality of evidence) is low or very low, suggesting that such recommendations may seldom be justified.

Objectives: To characterize the distribution of strength of recommendations and confidence in estimates of effect in WHO guidelines that have used the GRADE approach and graded strength of recommendations and confidence in effect estimates.

Methods: 436 WHO documents were initially reviewed (October/November 2012). We identified 116 (26.0%) guidelines of which 48 (41.3%) referred to GRADE methods, and 43 (37%) utilized GRADE and provided both a strength of recommendation and confidence in estimates grading. We describe the distributions of strong and weak recommendations and associated rating of confidence in estimates of effect.

Results: The 43 guidelines offered 456 recommendations: 290 were strong (63.6%) and 166 (36.4%) were weak. Of the 290 strong recommendations 50 (17.2%) were based on high confidence in estimates, 80 (27.6%) were based on moderate confidence, 97 (33.4%) were based on low confidence, and 63 (21.7%) on very low confidence (a total 55.2% on low or very low confidence estimates). Of the 166 weak recommendations, 4 (2.4%) were based on high confidence in estimates, 24 (14.5%) based on moderate confidence, 59 (35.5%) on low confidence, and 79 (47.6%) based on very low confidence.

Discussion: Strong recommendations based on low or very low confidence in effect estimates are very frequently made in WHO guidelines. Further study to determine the reasons for such recommendations is warranted. Implications: Guideline developers/authors should provide a clear, compelling rationale for strong recommendations based on low confidence estimates.