Summary of the GRADE System Application
GRADE stands for Grading of Recommendations Assessment, Development, and Evaluation.
Different information resources, such as textbooks and clinical practice guidelines, contain statements telling others what to do. These statements are recommendations. In this section, we describe the formulation and presentation of such recommendations in the McMaster Textbook of Internal Medicine. Our goal is to distinguish statements used when indicating the best course of action with complete certainty from statements that convey a lower degree of certainty.
The Gist of GRADE
The body of evidence of 1 or more studies will start as high or low certainty of evidence (CoE), or quality of evidence (QoE), if it is composed of randomized controlled trials (RCTs) or nonrandomized studies (NRS), respectively (A and B). There are 5 GRADE domains that can downgrade the CoE/QoE (C), while 3 will upgrade it when judged to be present (D). Based on these judgements a final CoE/QoE is obtained for the whole body of evidence for each outcome (E).
Note: Upgrading the CoE is only recommended for nonrandomized studies.
Quality of Evidence (QoE) or Certainty of Evidence (CoE) is described as high through moderate to low and reflects our trust (confidence, certainty) that we know true effects of alternative interventions in terms of outcomes that patients consider important. The quality depends on the methodology of the underlying studies and includes:
- Risk of bias.
- Similarity of findings from study to study (homogeneity vs heterogeneity of findings).
- Precision of estimates.
- Directness of studies to the question of interest in terms of:
- Similarity of our patient(s) to the population studied.
- Similarity of interventions used to what we intend to use.
- Similarity of outcomes measured to what we want to accomplish.
- The likelihood of publication bias.
Strong recommendation is made when we are confident that following our recommendation will do more good than harm (all or almost all fully informed patients would choose the recommended action). The language used in such cases involves “we recommend,” “one should,” or imperative forms like “do it,” “use this drug.”
Weak recommendation (suggestion) is made when we are less confident in the given course of action (the majority of fully informed patients would choose the suggested course of action but a minority would not). The preferred language is “we/I suggest,” “one might,” or equivalent. Please note that making strong recommendations in the setting of low quality of evidence is rare.
Comparison of a statement of fact versus an actionable recommendation: Sometimes our goal is to present a fact rather than make a recommendation. As an example consider: “Impaired left ventricular function predicts poor prognosis” versus “Use medication X to improve left ventricular function.” Both types of statements may have QoE assigned to them to indicate how much we trust the underlying evidence.
Confidence in the recommendation/strength of recommendations: We express confidence/certainty through the language used to formulate recommendations. Apart from that, in the case of selected recommendations we also indicate confidence using appropriate symbols.
When deciding on the strength of recommendations, one needs to take into account the quality of the underlying evidence, the likely balance between desirable and undesirable consequences, variability in patients values and preferences (if relevant), and cost.
GRADE Application in the McMaster Textbook
|Table 1. Strength of Recommendation symbols used in the McMaster Textbook of Internal Medicine|
|Strong recommendation for a given action|
|Weak recommendation for a given action|
|Weak recommendation against a given action|
|Strong recommendation against a given action|
|Table 2. Quality of Evidence symbols used in the McMaster Textbook of Internal Medicine|
|High Quality of Evidence|
|Moderate Quality of Evidence|
|Low Quality of Evidence|
Additional details regarding the evidence are provided only in the electronic version of the McMaster Textbook of Internal Medicine. They may include references or descriptions of the best available evidence, preferably summarized in meta-analyses accessible via PubMed.
Observational studies in the context of interventions (drug, device, and surgery) start as providing low-quality evidence. In well-done observational studies with good controls and a very large effect (for instance, antibiotics in septic shock or insulin in ketoacidosis), the quality would increase to moderate or even—although rarely—high. Randomized controlled trials (RCTs)-derived Quality of Evidence starts as high but could be lowered down to moderate or even low by problems with the design or conduct of individual studies (bias), inconsistent results, imprecision, results that are only indirectly applicable to the problem at hand (eg, RCTs in younger patients applied to geriatric patients), and publication bias.
We judge the precision by looking at the confidence intervals. If the clinical decision remains the same through the range of plausible results (confidence intervals), the precision is adequate. If the decision would differ, the precision is likely not adequate and we would rate down the quality.
|Table 3. Phrases used for indicating Strength of Recommendation|
|Strong recommendation, in favor: Strong recommendation (benefits clearly outweigh downsides; right action for all or almost all patients).|
|Strong recommendation, against: Strong recommendation (downsides clearly outweigh benefits; right action for all or almost all patients).|
|Weak recommendation (suggestion), in favor: Weak recommendation (benefits likely outweigh downsides, but the balance is close or uncertain; an alternative course of action may be better for some patients).|
|Weak recommendation (suggestion), against: Weak recommendation (downsides likely outweigh benefits, but the balance is close or uncertain; an alternative course of action may be better for some patients).|
|Table 4. Phrases used for indicating Quality of Evidence (the level of certainty that we know the true patient-important effects of a given intervention)|
|High Quality of Evidence (high confidence that we know true effects of the intervention).|
|Moderate Quality of Evidence (moderate confidence that we know true effects of the intervention). Quality of Evidence lowered due to…|
|Low Quality of Evidence (low confidence that we know true effects of the intervention). Quality of Evidence lowered due to…|
ReferencesThe GRADE Working Group. Accessed October 23, 2018. http://www.gradeworkinggroup.org/
Rochwerg B, Alhazzani W, Jaeschke R. Clinical meaning of the GRADE rules. Intensive Care Med. 2014 Jun;40(6):877-9.
Balshem H, Helfand M, Schünemann HJ, et al. GRADE guidelines: 3. Rating the quality of evidence. J Clin Epidemiol. 2011 Apr;64(4):401-6
Andrews JC, Schünemann HJ, Oxman AD, et al. GRADE guidelines: 15. Going from evidence to recommendation-determinants of a recommendation's direction and strength. J Clin Epidemiol. 2013 Jul;66(7):726-35.