Understanding the determinants WHO guideline panels consider in making strong recommendations in the face of low or very low confidence (study quality) in effect estimates

Tags: Oral
Alexander P1, Norris S2, Bero L3, Brito JP4, Neumann I1, Djulbegovic B5, Montori VM4, Rave S6, Gionfriddo M4, Guyatt G1
1McMaster University, Canada, 2Guidelines Review Committee Secretariat, WHO, Switzerland, 3University of California San Francisco, USA, 4Mayo Clinic, USA, 5University of South Florida, USA, 6Health Canada, Toronto, Canada

Background:

In 2007, the World Health Organization (WHO) adopted the GRADE approach to guideline development which characterizes evidence confidence in estimates from high to very low, and recommendations as strong or weak. Because low or very low confidence in estimates (l/vl) is associated with recommendations sensitive to values and preferences, GRADE suggests that guideline panels make strong recommendations based on l/vl sparingly. We have previously developed a taxonomy of strong recommendations based on l/vl comprising: 1) 5 identified paradigmatic situations in which strong recommendations are warranted despite l/vl; 2) best practice recommendations; 3) misclassification of confidence in estimates; 4) unwarranted strong recommendations.

Objectives:

To characterize WHO strong recommendations based on l/vl confidence in estimates.

Methods:

We previously reported on 456 WHO GRADEd recommendations published 2012 or earlier, of which 63% were strong and, of these, 55% based on l/vl confidence. We are in the process of characterizing these recommendations according to our taxonomy. We have re-examined WHO guidelines for 2013 and found that 70% of recommendations were strong and of these, 76% based on low or very low confidence in effect estimates.

Results:

Thus far we have noted that all four categories are well represented. Full results will be available by September 2014.

Conclusions:

Our examination of the guideline documents may reveal that: 1) WHO recommendations are based on reasonable judgements and consistent with GRADE guidance; 2) recommendations may not have been formulated with correct applications of the GRADE approach and panellists judgements were incorrect; or 3) even if the determination is accurate, the reasoning of the panel may highlight limitations in GRADE guidance. Understanding when and why strong recommendations are being formulated at WHO with l/vl is an important methodological issue that has implications not just for WHO, but for a wide variety of guideline development groups internationally. This project will help identify potential educational needs of panellists regarding optimal use of GRADE.