Grading quality of evidence and strength of recommendations in clinical practice guidelines

Journal compilation Ó 2009 Blackwell Munksgaard Grading quality of evidence and strength of recommendationsin clinical practice guidelines Part 1 of 3. An overview of the GRADE approach and grading qualityof evidence about interventions The GRADE (Grades of Recommendation, Assessment, Development, and Evaluation) approach provides guidance to grading the quality of underlying evidence and the strength of recommendations in health care. The GRADE systemÕs conceptual underpinnings allow for a detailed stepwise process that defines what role the quality of the available evidence plays in the development of health care recommendations. The merit of GRADE is not that it eliminates judgments or disagreements about evidence and recommendations, but rather that it makes them transparent. This first article in a three-part series describes 1Department of Epidemiology, Italian National the GRADE framework in relation to grading the quality of evidence about Cancer Institute Regina Elena, Rome, Italy; interventions based on examples from the field of allergy and asthma. In the 2Department of Medicine, Jagiellonian University GRADE system, the quality of evidence reflects the extent to which a guideline School of Medicine, Krakow, Poland; 3Department panelÕs confidence in an estimate of the effect is adequate to support a particular of Medicine and Social and Preventive Medicine, recommendation. The system classifies quality of evidence as high, moderate, School of Medicine and Biomedical Sciences, StateUniversity of New York at Buffalo, Buffalo, NY, USA; low, or very low according to factors that include the study methodology, 4Iberoamerican Cochrane Center, Servicio de consistency and precision of the results, and directness of the evidence.
Epidemiología Clínica y Salud Pfflblica, Hospital deSant Pau, Barcelona, Spain; 5Centro deInvestigación BiomØdica en Red de Epidemiología ySalud Pfflblica (CIBERESP), Spain; 6Allergy/Immunology Section, Respiratory Institute,Cleveland Clinic, Cleveland, OH, USA; 7Departmentof Medicine, McMaster University, Hamilton, ON,Canada; 8Department of Internal Medicine andPsychiatry, Duke University and Durham VA MedicalCenter, Durham, NC, USA; 9Centre for Reviews andDissemination, University of York, York, UK;10HTA-Zentrum, Universität Bremen, Bremen,Germany; 11School of Population Health, Faculty ofMedical and Health Sciences, University of Auckland,New Zealand; 12Service des Maladies Respiratoires,Hôpital Arnaud de Villeneuve, Montpellier, France;13Inserm UMR 780, France; 14Department of ClinicalEpidemiology and Biostatistics, McMasterUniversity, Hamilton, ON, Canada Key words: clinical practice guidelines; evidence basedmedicine; grading.
Holger J. Schünemann MD PhDDepartment of Clinical Epidemiology & BiostatisticsMcMaster University Health Sciences Centre,Room 2C10B1200 Main Street West HamiltonON L8N 3Z5Canada Accepted for publication 16 November 2008 When offered a diagnostic procedure or a treatment Guideline panels develop recommendations on the basis option, patients ask themselves and the clinicians taking of the balance between the desirable and the undesirable care of them about the benefits and downsides of that consequences of the diagnostic or therapeutic options in choice. They ask: What will I gain? Will I feel better question. They will recommend the option that results in (reduced symptoms or morbidity and improved quality greater net benefit and recommend against the option of life)? Will I live longer (reduced mortality)? Patients that results in greater net loss. The strength of their also ask: What will I lose? Is it safe and will I dislike recommendation will depend on the extent to which they some aspects related to the intervention (adverse events, can be confident that desirable effects outweigh undesir- burden – extra time and effort)? How much will it cost able effects, or vice versa. A systematic approach to me? Thus, the decision to choose among the options grading the strength of recommendations can minimize depends on the balance between their desirable and bias and aid interpretation (4, 5). The Grades of undesirable consequences. This balance weighs not only Recommendation, Assessment, Development, and Eval- what patients will gain or lose but also how much they uation (GRADE) working group has conducted a review will gain or lose (one can estimate it based on the of existing grading systems and developed a system for evidence from current research), and how important are grading the quality of evidence and strength of recom- the gains and losses for them (patientsÕ values and mendations that addresses shortcomings of prior systems preferences for the different outcomes and interven- (4, 6–9). The resulting GRADE system has a number of advantages over other grading systems (Table 1). These The role of a clinician is not only to order a diagnostic advantages are reflected in the increasing number of test or prescribe a treatment, but also to advise patients – professional societies and organizations endorsing or sometimes to decide for them – which of the available using the GRADE system – examples include the tests or treatments is likely to be most beneficial and American College of Chest Physicians (ACCP) (10), the American Thoracic Society (ATS) (7), the British Medical As we cannot predict the future, we always have to Journal (11), the Cochrane Collaboration (12), the make these decisions under uncertainty about the Endocrine Society (13), the European Respiratory Society outcomes for a particular patient. Optimal decision- (ERS), Infectious Disease Society of America (IDSA), making requires informing these decisions with the best Surviving Sepsis Campaign (14), UpToDateÒ (15), and the World Health Organization (WHO) (16) among experience of the effect of similar management of others (a comprehensive list of endorsing organizations similar patients). Clinical practice guidelines can help is available at http://www.gradeworkinggroup.org). Most clinicians and patients make these decisions but their recently, the Allergic Rhinitis and its Impact on Asthma application is not always easy as every patient isdifferent.
Table 1. Merits of the GRADE system for grading quality of evidence and strength of recommendations in comparison to other systems Clinical practice guidelines offer recommendations for 1. Clear separation between quality of evidence and strength of recommendations* diagnostic procedures or treatment options for typical 2. Explicit and comprehensive criteria for downgrading or upgrading quality of patients. They are Ôsystematically developed statements to assist practitioner and patient decisions about appro- 3. Explicit consideration of the relative importance of various outcomes to patients priate health care for specific clinical circumstancesÕ (1).
4. Explicit acknowledgement of values and preferences assumed when making The purpose of guidelines is Ôto make explicit recom- 5. Transparent process of moving from evidence to recommendations mendations with a definite intent to influence what 6. Explicit advice to make recommendations about the most appropriate course clinicians doÕ (2). Clinical decisions – and also the of action, even when very little evidence is available related recommendations and their strength – depend on 7. Grading the strength only for recommendations about the diagnostic or both the research evidence and the values and prefer- therapeutic course of action, but not about prognosis or etiology ences of patients. For clinicians and patients to be 8. Clear and pragmatic interpretation of ÔstrongÕ and ÔweakÕ recommendations confident that following these recommendations will do 9. Balance between simplicity and methodological comprehensiveness more good than harm, guidelines need to be evidence- *In the context of clinical practice guidelines: Quality of evidence, the extent to based, transparent, and explicit about whose values and which our confidence in an estimate of the treatment effect is adequate to support preferences were taken into account and how they particular recommendation; Strength of recommendation, the extent to which we influenced the final recommendations. Systematic ap- can, across the range of patients for whom the recommendations are intended, be proach, transparency, and explicitness also facilitate confident that desirable consequences of an intervention outweigh undesirable implementation, adaptation to local circumstances, and consequences (or vice versa, that undesirable consequences outweigh desirableones – in this case one would recommend against this intervention).
Journal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Grading quality of evidence and strength of recommendations (ARIA) guideline panel decided to use the GRADE recommendations would answer. As guidelines include system for the 2009 revision of the guidelines and to recommendations about the most appropriate course of follow the GRADE approach in the future (17).
action, they should answer clinical management questions We suggest conceptualizing GRADE as a system of about diagnosis or treatment of disease, but not about grading quality of evidence and also as a systematic and prognosis or etiology. A clinical management question transparent approach to the process of developing should have four components: patient population, inter- recommendations for clinical practice including indicat- vention (diagnostic or therapeutic), alternative interven- ing the strength of these recommendations (Table 2).
tion (comparison), and the outcomes of interest (23). For In this series of three articles, we will present the instance, consider the following: in patients with persis- GRADE approach to transparent development of evi- tent allergic rhinitis (patient population) should oral dence-based recommendations. In this article, we will H1-antihistamines (intervention) vs no oral H1-antihista- start with a brief overview of GRADE approach and we mines (alternative intervention) be used to improve will discuss grading the quality of available evidence quality of life, reduce symptoms, and minimize the supporting the recommendations about therapeutic inter- ventions. In a second article, we will present the approach There are potential problems arising at the stage of to grading the quality of available evidence about asking a clinical question. One is the failure to consider all diagnostic strategies. In a third article, we will present relevant alternatives. This may be particularly important the GRADE approach to formulating the recommenda- in international guidelines where treatment options vary tions, deciding on their strength, and suggested interpre- for patients in many diverse jurisdictions. Two other closely related mistakes in formulating questions are the As many guidelines are adopting GRADE, including failure to include all patient-important outcomes, e.g.
the 2009 revision of ARIA, it is important that allergists disregarding quality of life or adverse effects, and placing understand the underlying concepts. Therefore, this series excessive emphasis on surrogate outcomes with question- is intended for clinicians especially interested in allergy, able importance to patients such as pulmonary function who want to be able to fully interpret the recommenda- or nasal airway resistance rather than objectively tions in guidelines developed following the GRADE measured quality of life or symptoms.
approach. Whenever possible, we will use examplesspecific to the field of allergy and asthma.
Decide on the relative importance of outcomes For clinicians interested in more in depth review of the GRADE approach and system, we recommend a series of The GRADE approach asks guideline developers to articles recently published in the British Medical Journal make explicit judgments about the importance of each (18–22) or an even more detailed series for guideline outcome for making a recommendation. GRADE developers that will be published in the Journal of demands that those making recommendations classify Clinical Epidemiology, and the earlier papers (4, 7).
each of the outcomes of interest as either critical formaking a recommendation, important but not critical, ornot important (24). Because experts, clinicians, andpatients differ in their preferences and how they value particular outcomes (25), input from those affected by the recommendation (i.e. patients, their families, or membersof the public) should be sought if possible. For example, Following the GRADE approach, one begins with outcomes such as mortality, quality of life, or exacerba- formulating appropriate clinical questions that the tions of asthma might be considered critical, nasalsymptoms judged by a physician or use of rescue Table 2. An overview of steps followed during the development of an evidence- medications – important but not critical, and peak expiratory flow or nasal eosinophilia – not important, but perhaps informative, for making a recommendation.
Define the scope of the guidelinesPrioritize the problems (60) Identify the existing evidence for every clinical question Ask precise clinical questionsDecide on the relative importance of outcomes Every clinical question should then be answered based on Identify the existing evidence for every clinical question a systematic review of the relevant evidence (26). Guide- line developers can either conduct the systematic review Grade the quality of existing evidence for each outcome separatelyDetermine the overall quality of available evidence across outcomes themselves or identify an existing high quality systematic Decide on the balance between desirable and undesirable consequences review. This systematic review will serve to create a summary of available evidence that a guideline panel can Formulate the recommendation reflecting its strength use to inform judgments about the balance of desir- able and undesirable effects in order to develop a Ó 2009 The AuthorsJournal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 recommendation. GRADE suggests that this summary of Table 3. Quality of evidence and the explanation of the categories evidence is prepared in a structured format of evidence profile – a table including detailed judgments aboutquality of evidence and estimates of the effect for each The recommendation and its strength depend not only on the best estimates of the expected benefits and downsides, but also on the confidence in these estimates. If we know the best estimates of the magnitude of the effects, but we have no confidence in these estimates (i.e. we do not ÔbelieveÕ in them), it is very difficult to determine the balance of desirable and undesirable consequences.
One of the factors that influence our confidence in the estimates of treatment effects is the quality of supporting evidence – the higher it is, the more confidence we have in these estimates. Formal grading of the quality of evidence and its explicit consideration are essential to the process of developing recommendations. The examples of errors arising from disregarding the quality of supporting evidence are abundant in the modern history of medicine.
Consider the treatment of patients with myocardialinfarction. For about a decade, experts made recommen- *The examples are not comprehensive. See text for criteria to downgrade or up-grade the quality of evidence.
dations ignoring the high quality evidence about thebenefit from thrombolysis or the lack of benefit, and in fact a continuum and any categorization involves possibly even harm, from routine administration of arbitrariness and the possibility of oversimplification, we antiarrhythmic agents in the early postmyocardial infarc- think that clarity, transparency, and intuitive understand- tion period (27). They based their judgments on patho- ing of the four categories outweigh these limitations.
physiological considerations, such as reduction in thefrequency of arrhythmia that failed to recognize higher Study design. Earlier systems of grading the quality of quality evidence focusing on patient important outcomes evidence relied almost exclusively on overall study design including mortality. In the field of allergy and asthma, (e.g. randomized trials vs observational studies). In the there are less dramatic examples. As an illustration of GRADE system, study design remains a critical, but not misleading conclusions from relying on lower quality a sole factor in judging the quality of evidence (Table 4).
evidence, one might consider a systematic review of observational studies assessing the effect of inhaled or options, randomized trials provide, in general, far oral corticosteroids on height in children with asthma stronger evidence than observational studies, yet rigorous concluding that the use of inhaled beclomethasone observational studies provide far stronger evidence than dipropionate was not associated with diminished stature uncontrolled case series. Therefore, in the GRADE (28). However, a recent systematic review restricted to system, a body of evidence obtained from randomized randomized trials found a statistically significant decrease trials is initially rated as high quality, and that obtained in linear growth velocity in children with mild tomoderate asthma treated with moderate doses of beclo- Table 4. Factors influencing the quality of evidence methasone (29). A formal system of grading the quality ofevidence provides a strategy to clarify how reliable is the Study design (experimental vs observational) decreasing the risk of repeating the types of mistakes Limitations in study design and/or execution Inconsistency of resultsIndirectness of evidence In the context of clinical practice guidelines, the GRADE system defines quality of evidence as the extent to which our confidence in an estimate of the treatment Factors that can increase the quality of evidence effect is adequate to support a particular recommenda- tion. The GRADE system specifies four grades of All plausible confounding may be working to reduce evidence: high, moderate, low, and very low quality the demonstrated effect or increase the effect if no effect was observed (Table 3). Acknowledging that the quality of evidence is Journal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Grading quality of evidence and strength of recommendations from observational studies – as low quality. For example, Inconsistency of results. Widely differing estimates of the a systematic review of the effect of using feather bedding treatment effect across individual studies (variability or in the control of asthma symptoms identified no heterogeneity of results) suggest true differences in randomized trial addressing this clinical question (30).
underlying treatment effects (39). Authors of a systematic The only available evidence indicating that more frequent review should try to identify plausible explanations for wheezing is associated with nonfeather pillows comes inconsistent results but, if they do not succeed, the quality from two case–control studies that found a 20% rise in of evidence decreases. Variability in individual study the population prevalence odds of wheezing from 1978 to results may arise from clinical differences in populations 1991, and identified an increase (from 44% to 67%) in the (e.g. drugs may have larger relative effects in sicker use of nonfeather pillows as the only domestic indoor patients), interventions (e.g. larger effects or larger side- exposure that appeared to explain this (31). The initial effects with higher drug doses), and outcome measures rating for this evidence would be low quality.
(e.g. differences in the definition of Ôresponse to treat- In the GRADE system, Ôexpert opinionÕ is not a mentÕ), or from methodological limitations such as category of quality of evidence, but an interpretation of problems with randomization, early termination of trials, existing evidence. Therefore, expert opinion is nearly or publication bias (40, 41). For example, a systematic always necessary to integrate and contextualize evidence, review of subcutaneous allergen-specific immunotherapy either from a clinical or from a methodological viewpoint.
in adults with allergic rhinitis found inconsistency in the A well-designed and executed randomized trial or effect of treatment on nasal symptoms that was suggested observational study provides different quality evidence both by visual examination of forest plot and statistical than the one that was poorly conducted. Therefore, tests. Despite the effort to find a reason for this relying on study design alone has apparent limitations.
heterogeneity, authors of the systematic review were not GRADE provides additional quality criteria that serve to able to explain it (42). In another example, a systematic overcome this shortcoming. We have identified five review showed that ketotifen reduced the use of bron- factors that can reduce the quality of evidence for each chodilators in children with mild to moderate asthma.
study design and three that can increase it (Table 4).
However, there was significant heterogeneity among theresults of individual trials (I2 = 76.1%) (43). Subgroup Limitations in study design and/or execution (risk of and sensitivity analyses explained this heterogeneity – the bias). Quality of evidence initially rated based on study effect was stronger in school children than in infants or design decreases when studies suffer from major meth- preschool children (differences in populations) and it odological limitations that can bias their estimates of the disappeared in trials with adequate blinding (differences treatment effect. These limitations include lack of alloca- tion concealment, lack of blinding – particularly ifoutcomes are subjective and their assessment is highly Indirectness of evidence. GRADE distinguishes two types susceptible to bias, lack of accounting for a large of indirectness – indirect comparisons and differences in proportion of patients who started the study (large loss populations, interventions, and outcomes of interest to follow-up or outcome not measured in a large between the studies (existing evidence) and the scope of proportion of patients), failure to adhere to the inten- the recommendation (clinical question).
tion-to-treat principle during the analysis, stopping early Indirect comparison arises when, for instance, the for benefit, or selectively reporting outcomes that show recommendation addresses the choice between two active an apparent treatment effect and failing to report other drugs: A or B, but the available studies compared A vs outcomes that show no evident effect (32–36). For placebo and B vs placebo. Such trials allow indirect example, the evidence for the effect of sublingual comparison of the magnitude of the effect of A vs B. Such immunotherapy in children with allergic rhinitis on the an indirect comparison provides lower quality evidence development of asthma, comes from a single randomized than a head-to-head comparison of A vs B would provide.
trial with no description of randomization, concealment This type of indirectness is common when choosing of allocation, type of analysis, no blinding, and 21% of between the drugs within the same class (e.g. long acting children lost to follow-up (37). These very serious b-agonists, oral or topical H1-antihistamines, allergen limitations would warrant downgrading the quality of extracts for immunotherapy, etc.). As an illustration, one evidence by two levels (i.e. from high to low). In another example, a systematic review showed that the family patients with severe allergic rhinitis. Systematic review therapy for children with asthma improved outcomes of the studies in seasonal allergic rhinitis showed a such as daytime wheeze and the number of functionally consistent small to large effect of subcutaneous allergen- impaired days. However, allocation was clearly not specific immunotherapy (SCIT) compared with placebo concealed in one of the two included trials and unclear on symptoms of allergic rhinitis, ocular symptoms, and whether it was concealed in the second trial (38). This quality of life (42). Another systematic review showed limitation might warrant downgrading the quality of that sublingual allergen-specific immunotherapy (SLIT) is also effective in reducing symptoms and medication Ó 2009 The AuthorsJournal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Table 5. Sources of likely indirectness of evidence Early administration of systemic corticosteroids in the Both oral and intravenous routes are effective but there emergency department to treat acute exacerbations is no direct comparison of these two routes of Oral H1-antihistamines for improving quality of life in In the only study that measured quality of life, 60% of adults with asthma and concomitant allergic rhinitis patients had a past history of asthma but no symptoms Ketotifen for long-term control of symptoms and wheeze Inhaled corticosteroids, the mainstay of therapy of asthma nowadays, were allowed as an additional intervention in 60% of trials assessing ketotifen.
There was no enough information to assess the effect of ketotifen as an add-on therapy in children with Anti-leukotrienes plus inhaled glucocorticosteroids Trials that measured asthma exacerbations and nighttime vs inhaled glucocorticosteroids alone to prevent asthma symptoms did not include patients with allergic rhinitis exacerbations and nighttime symptoms in patients with Avoidance of pet allergens in nonallergic infants or Available studies used multifaceted interventions preschool children to prevent development of allergy directed at multiple potential risk factors in addition Oral decongestant as a rescue medication in patients Available studies used oral decongestants administered regularly, but none investigated their use as a rescue medication for quick alleviation of the symptoms Intranasal glucocorticosteroids vs oral H1-antihistamines In the available study, parents were rating the symptoms in children with seasonal allergic rhinitis and quality of life of their teenage children, instead requirements in these patients (44); however, the magni- seasonal allergic conjunctivitis showed that patients using tude of the benefit achieved with SLIT compared with topical sodium cromoglicate were more likely to perceive that of SCIT is not clear, because they have been benefit than those using placebo. However, only small compared directly in only very few studies (45, 46).
trials reported clinically and statistically significant Evidence supporting the recommendation is also indi- benefits of active treatment, while a larger trial showed rect when it comes from studies in which population, a much smaller and a statistically not significant effect intervention, alternative intervention, or outcomes of (51). These findings suggest that smaller studies demon- interest were different from those that the recommenda- strating smaller effects might not have been published.
Evidence supporting a particular recommendation can suffer from more than one of the above limitations, and Imprecision of results. When studies include relatively few the more serious they are, the lower the quality of the patients and few events occur, estimates of the effect evidence is. For example, randomized trials of high usually have wide confidence intervals that include both efficiency particulate air filters in patients with perennial important benefits or no important effects (or even allergic rhinitis suffered very serious limitations in design important harm). With such indeterminate results, one (warranting downgrading by 2 levels) and the results were can judge the quality of the evidence lower than one otherwise would, because of resulting uncertainty in the The GRADE system offers three criteria that, when effect. For instance, observational studies examining fulfilled, can increase the quality of evidence. They are the impact of exclusive breast feeding on the development infrequently applicable, but are the most common reason of allergic rhinitis in high-risk infants showed a relative for upgrading the quality of evidence from well-performed risk of 0.87 (95% CI: 0.48–1.58) that rules out neither observational studies that without these additional merits important benefit nor important harm (47).
would provide only low quality evidence.
Publication bias. The quality of evidence will be reduced Large magnitude of effect. On rare occasions, when when investigators fail to report (publish) studies they studies yield large or very large estimates of the magni- have undertaken – typically those that showed no effect.
tude of the effect, one may be more confident about the Unfortunately, one must often guess about the likelihood results. Based on modeling studies that provide estimates of publication bias. The risk of publication bias is higher of the magnitude of effect that is very unlikely to be when only few small studies are available (48–50). For explained by bias (53, 54), the GRADE system defines a example, a systematic review of topical treatments for large effect as a relative risk (RR) of >2.0 or <0.5 (based Journal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Grading quality of evidence and strength of recommendations on consistent evidence from at least two studies, with no illnesses in early childhood when exposed to second-hand plausible confounders) and a very large effect as a RR of smoke from parental smoking. Moreover, the greater the >5.0 or <0.2 (based on direct evidence with no major exposure, the higher was the risk. While grading the threats to validity). For example, the extremely large and quality of available evidence supporting the recommen- consistent effect of epinephrine injections in anaphylactic dation to reduce second-hand smoke in children, one shock leaves us convinced of the benefits of the interven- might consider these results as indirect evidence of benefit from reducing the second-hand tobacco smoke exposureand initially rate it as low quality evidence from All plausible confounding would reduce the demonstrated observational studies that is downgraded to very low effect or increase it if no effect was observed. On rare because of indirectness (evidence of increased risk with occasions, all plausible biases may be working to increased exposure rather than of benefit with reduced underestimate the true treatment effect. For instance, if exposure). The observed dose-response gradient would only sicker patients receive an experimental intervention, justify upgrading the quality of evidence back to low.
yet they still fare better than patients not receiving it do, itis likely that the actual effect may be larger than the data Determine the overall quality of evidence across outcomes suggest. There are few examples so far in the literatureand we were not able to identify one in the field of asthma Each recommendation depends on the evidence about and allergy. However, consider a systematic review of outcomes identified when asking clinical questions and observational studies that included 38 million patients, regarded as important to patients. Following the which demonstrated higher death rates in private for- GRADE process, those making recommendations first profit vs private not-for-profit hospitals (55). Biases grade the quality of available evidence supporting each related to different disease severity in patients in the outcome separately. Subsequently, they specify the over- two hospital types, and the spill-over effect from well- all quality of evidence across these multiple outcomes, insured patients would both lead to estimates in favor of because guidelines provide a single grade of quality of for-profit hospitals (56). One might therefore consider the evidence for each recommendation. For any recommen- evidence from these observational studies higher than low dation, when the quality of evidence differs across quality. Because the plausible biases would all diminish outcomes, the GRADE system demands that the lowest the demonstrated intervention effect, one might consider grade of quality of available evidence for any of the the evidence from these observational studies as moderate outcomes deemed critical determines the overall quality rather than low quality. A parallel situation exists when of evidence supporting this recommendation. For exam- observational studies have failed to demonstrate an ple, based on a systematic review of monoclonal anti-IgE association but all plausible biases would have increased for chronic asthma in adults and children (58), one might an intervention effect. This situation will usually arise in grade the quality of evidence about asthma symptoms, the exploration of apparent harmful effects. For example, exacerbations, and quality of life as high, but the quality because the hypoglycemic drug phenformin causes lactic of evidence about adverse effects as moderate. Conse- acidosis, the related agent metformin is under suspicion quently, the overall quality of evidence supporting the for the same toxicity. Nevertheless, very large observa- recommendation about the use of this treatment would be tional studies have failed to demonstrate an association (57). Given the likelihood that clinicians would be morealert to lactic acidosis in the presence of the agent andover-report its occurrence, one might consider this moderate or even high quality evidence refuting a causalrelationship between typical therapeutic doses of metfor- The GRADE approach provides a comprehensive, explicit, and transparent methodology for grading thequality of evidence and strength of recommendations Dose-response gradient. The presence of a dose-response about the management of patients. GRADE classifies the gradient may also increase oneÕs confidence in the findings quality of available evidence as high, moderate, low or and thereby increase the quality of evidence. Most very low. Although judgments are required at every step evidence for dose-response gradient in the treatment of of guideline development, a systematic and explicit allergic diseases comes from well-performed randomized approach to grading the quality of evidence facilitates trials that do not require upgrading. However, consider scrutiny and transparency of these judgments.
the following example: there are no studies of interven- In the next article in this series, we will discuss the tions aimed at reduction of second-hand tobacco smoke GRADE approach to making recommendations about exposure that examined development of asthma or diagnostic methods in more detail and we will highlight wheeze in children. On the other hand, observational the differences in grading the quality of available evidence studies found an increased risk of developing wheeze between therapeutic and diagnostic interventions.
Ó 2009 The AuthorsJournal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Glasziou P, Jaeschke R, Vist GE et al.
11]; Available at: http://resources.
Higgins JPT, Deeks JJ, Glasziou P et al.
Vist GE, Bellamy R, Stockman L et al.
tation. Health Res Policy Syst 2006;4:25.
atrial fibrillation: observational study.
Bion J, Parker MM, Jaeschke R et al.
severe sepsis and septic shock: 2008. Crit 15. UpToDate. Editorial Policy. 2006 [cited 6. Atkins D, Eccles M, Flottorp S, Guyatt 16. World Health Organization. Guidelines 29. Sharek PJ, Bergman DA, Ducharme F.
mendations. BMJ 2008;336:1049–1051.
Jaeschke R, Helfand M, Liberati A et al.
What is ‘‘quality of evidence’’ and why R, Falck-Ytter Y, Alonso-Coello P et al.
dence in clinical guidelines: report from Journal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677 Grading quality of evidence and strength of recommendations seasonal allergic conjunctivitis: system- atic review and meta-analysis of efficacy 41. Fletcher J. What is heterogeneity and is 53. Bross ID. Pertinency of an extraneous 44. Wilson DR, Torres LI, Durham SR.
45. Mungan D, Misirligil Z, Gurbuz L.
mite-sensitive patients with rhinitis and asthma – a placebo controlled study.
Canadian Institutes of Health Research.
systematic review and meta-analysis.
ysis of prospective studies. Acta Paediatr chronic asthma in adults and children.
and dissemination of clinical research. J Advanced topics in systematic reviews.
Priority setting. Health Res Policy Syst2006;4:14.
Ó 2009 The AuthorsJournal compilation Ó 2009 Blackwell Munksgaard Allergy 2009: 64: 669–677

Source: http://www.ebcp.com.br/simple/upfiles/pdfs/2009BrozekPart1.pdf