Laying the Foundation for New Approaches to Practice

Journal of Clinical Epidemiology 64 (2011) 383e394

GRADE guidelines: 1. IntroductiondGRADE evidence profiles and summary of findings tables

Gordon Guyatt a,b,*, Andrew D. Oxman

c , Elie A. Akl

m , Regina Kunz

d , Gunn Vist

c , Jan Brozek

a ,

Susan Norrise, Yngve Falck-Ytterf, Paul Glaszioug, Hans deBeerh, Roman Jaeschkeb, David Rind

i , Joerg Meerpohl

j,k , Philipp Dahm

l , Holger J. Sch€unemanna,b

a Department of Clinical Epidemiology and Biostatistics, McMaster University, Hamilton, Ontario L8N 3Z5, Canada

b Department of Medicine, McMaster University, Hamilton, Ontario L8N 3Z5, Canada

cNorwegian Knowledge Centre for the Health Services, St. Olavs plass, 0130 Oslo, Norway dAcademy of Swiss Insurance Medicine, University Hospital Basel, Basel, Switzerland

e Department of Medical Informatics and Clinical Epidemiology, Oregon Health and Science University, Portland, OR 97239-3098, USA

fDivision of Gastroenterology, Case and VA Medical Center, Case Western Reserve University, Cleveland, OH 44106, USA gBond University, Gold Coast, Australia

h Dutch Association of Nursing-home Specialists, Mercatorlaan 1200, 3528 BL Utrecht, The Netherlands

i Harvard Medical School, UpToDate, Boston, MA, USA

jGerman Cochrane Center, Institute of Medical Biometry and Medical Informatics, University Medical Center Freiburg, 79104 Freiburg, Germany kDepartment of Pediatric and Adolescent Medicine, Division of Pediatric Hematology and Oncology, University Medical Center Freiburg,

79106 Freiburg, Germany l Department of Urology, University of Florida, College of Medicine, Gainesville, FL 3210, USA

mDepartment of Medicine, State University of New York at Buffalo, Buffalo, NY, USA

Accepted 8 April 2010

Abstract

This article is the first of a series providing guidance for use of the Grading of Recommendations Assessment, Development, and Eval- uation (GRADE) system of rating quality of evidence and grading strength of recommendations in systematic reviews, health technology assessments (HTAs), and clinical practice guidelines addressing alternative management options. The GRADE process begins with asking an explicit question, including specification of all important outcomes. After the evidence is collected and summarized, GRADE provides explicit criteria for rating the quality of evidence that include study design, risk of bias, imprecision, inconsistency, indirectness, and magnitude of effect.

Recommendations are characterized as strong or weak (alternative terms conditional or discretionary) according to the quality of the supporting evidence and the balance between desirable and undesirable consequences of the alternative management options. GRADE sug- gests summarizing evidence in succinct, transparent, and informative summary of findings tables that show the quality of evidence and the magnitude of relative and absolute effects for each important outcome and/or as evidence profiles that provide, in addition, detailed infor- mation about the reason for the quality of evidence rating.

Subsequent articles in this series will address GRADE’s approach to formulating questions, assessing quality of evidence, and devel- oping recommendations. � 2011 Elsevier Inc. All rights reserved.

Keywords: GRADE; systematic reviews; clinical practice guidelines; health technology assessment; quality of evidence; strength of recommendations

The Grading of Recommendations Assessment, Development, and

Evaluation (GRADE) system has been developed by the GRADE Working

Group. The named authors drafted and revised this article. A complete list

of contributors to this series can be found on the Journal of clinical Epi-

demiology website.

* Corresponding author. CLARITY Research Group, Department of

Clinical Epidemiology & Biostatistics, Room 2C12, 1200 Main Street

West Hamilton, Ontario L8N 3Z5, Canada. Tel.: þ905-525-9140; fax: þ905-524-3841.

E-mail address: [email protected] (G. Guyatt).

0895-4356/$ - see front matter � 2011 Elsevier Inc. All rights reserved. doi: 10.1016/j.jclinepi.2010.04.026

1. Introduction

In this, the first of a series of articles describing the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach to rating quality of evidence and grading strength of recommendations, we will briefly summarize what GRADE is, provide an overview of the GRADE process of developing recommendations, and present the endpoint of the GRADE evidence summary: the evidence profile (EP) and the summary of findings

384 G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

Key Points

� Grading of Recommendations Assessment, Devel- opment, and Evaluation (GRADE) offers a trans- parent and structured process for developing and presenting summaries of evidence, including its quality, for systematic reviews and recommenda- tions in health care.

� GRADE provides guideline developers with a com- prehensive and transparent framework for carrying out the steps involved in developing recommendations.

� GRADE’s use is appropriate and helpful irrespec- tive of the quality of the evidence: whether high or very low.

� Although the GRADE system makes judgments about quality of evidence and strength of recommen- dations in a systematic and transparent manner, it does not eliminate the inevitable need for judgments.

(SoFs) table. We will provide our perspective on GRADE’s limitations and present our plan for this series.

2. What is GRADE?

GRADE offers a system for rating quality of evidence in systematic reviews and guidelines and grading strength of recommendations in guidelines. The system is designed for reviews and guidelines that examine alternative management strategies or interventions, which may include no interven- tion or current best management. In developing GRADE, we have considered a wide range of clinical questions, in- cluding diagnosis, screening, prevention, and therapy. Most of the examples in this series are clinical examples. The GRADE system can, however, also be applied to public health and health systems questions.

GRADE is much more than a rating system. It offers a transparent and structured process for developing and pre- senting evidence summaries for systematic reviews and guidelines in health care and for carrying out the steps in- volved in developing recommendations. GRADE specifies an approach to framing questions, choosing outcomes of in- terest and rating their importance, evaluating the evidence, and incorporating evidence with considerations of values and preferences of patients and society to arrive at recom- mendations. Furthermore, it provides clinicians and pa- tients with a guide to using those recommendations in clinical practice and policy makers with a guide to their use in health policy.

A common definition of guidelines refers to ‘‘systemat- ically developed statements to assist practitioner and

patient decisions about appropriate health care for specific clinical circumstances’’ [1]. This series will describe GRADE’s comprehensive approach to guideline develop- ment and to other similar guidance documents.

The optimal application of the GRADE approach re- quires systematic reviews of the impact of alternative man- agement approaches on all patient-important outcomes. In the future, as specialty societies (e.g., American College of Physicians), national guideline developers and HTA agencies (e.g., National Institute for Health and Clinical Excellence), publishers (e.g., BMJ), publications (e.g., UpToDate), and international organizations (e.g., World Health Organization, Cochrane Collaboration) pool re- sources, high-quality evidence summaries will become increasingly available. As a result, even guideline panels with limited resources charged with generating recommen- dations for local consumption will be able to use GRADE to produce high-quality guidelines [2].

3. Purpose of this series

This series of articles about GRADE is most useful for three groups: authors of systematic reviews, groups con- ducting HTAs, and guideline developers. GRADE suggests somewhat different approaches for rating the quality of ev- idence for systematic reviews and for guidelines. HTA practitioners, depending on their mandate, can decide which approach is more suitable for their goals.

The GRADE approach is applicable irrespective of whether the quality of the relevant evidence is high or very low. Thus, all those who contribute to systematic reviews and HTA, or who participate in guideline panels, are likely to find this series informative. Consumersdand criticsdof reviews and guidelines who desire an in-depth understand- ing of the evidence and recommendations they are using will also find the series of interest.

The series will provide a ‘‘how to’’ guide through the process of producing systematic reviews and guidelines, us- ing examples to illustrate the concepts. We will not start with a broad overview of GRADE but rather assume that readers are familiar with the basics. Those who are not fa- miliar may want to begin by reading a brief summary of the approach [3]. Those who want to start with a more detailed overview should examine all the articles in a previously published series describing the GRADE approach [4e9]. Finally, a computer program (GRADEpro) [10] and associ- ated help file [11] that facilitate the development of EPs and SoFs tables provide a complement to this series.

4. The GRADE processddefining the question and collecting evidence

Figure 1 presents a schematic view of GRADE’s process for developing recommendations in which unshaded boxes describe steps in the process common to systematic reviews

noitseuQ eraC htlaeH )OCIP( weiver citametsyS

seidutS

semoctuO

tnatropmI semoctuo

emoctuo hcae rof ecnedive fo ytilauq eht etaR seiduts ssorca ,

esbo ,gnitar hgih a htiw trats sTCR gnitar wol a htiw seiduts lanoitavr

R:drawnwod deifidom si gnitaR :drawpu deifidom si gnita tceffe fo edutingam egraL-snoitatimil ydutS-

esnopser esoD-noisicerpmI- tceffe eht eziminim ylekil srednuofnoC-stluser fo ycnetsisnocnI-

ecnedive fo ssentceridnI- ylekil saib noitacilbuP-

:emoctuo hcae rof ytilauq fo gnitar laniF wol yrev ro ,wol ,etaredom ,hgih

)tsniaga/rof( noitcerid eht no ediceD )*kaew/gnorts( htgnerts edarg dna

:gniredisnoc noitadnemmocer eht fo

ecnedive eht fo ytilauQ semoctuo elbarisednu/elbarised fo ecnalaB

secnereferp dna seulaV fo noisiver yna fi ediceD si htgnerts ro noitcerid :gniredisnoc yrassecen esu ecruoseR

*Also labeled “conditional” or “discretionary”

ecnedive fo ytilauq llarevo etaR

)semoctuo lacitirc gnoma ytilauq tsewol(

1S 2S 3S 4S

1CO 2CO 3CO 4CO lacitirC

semoctuo

fo etamitse na etareneG emoctuo hcae rof tceffe

5S

Fig. 1. Schematic view of GRADE’s process for developing recommendations. Abbreviation: RCT, randomized controlled trials.

385G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

and guidelines and the shaded boxes describe steps that are specific to guidelines. One begins by defining the question in terms of the populations, alternative management strate- gies (an intervention, sometimes experimental and a compar- ator, sometimes standard care), and all patient-important outcomes (in this case four) [12]. For guidelines, one clas- sifies those outcomes as either critical (two outcomes in the figure) or important but not critical (two outcomes). A sys- tematic search leads to inclusion of relevant studies (in this schematized presentation, five such studies).

Systematic review or guideline authors then use the data from the individual eligible studies to generate a best esti- mate of the effect on each patient-important outcome and an index (typically a confidence interval [CI]) of the uncer- tainty associated with that estimate.

5. The GRADE processdrating evidence quality

In the GRADE approach, randomized controlled trials (RCTs) start as high-quality evidence and observational stud- ies as low-quality evidence supporting estimates of interven- tion effects. Five factors may lead to rating down the quality of evidence and three factors may lead to rating up (Fig. 2). Ultimately, the quality of evidence for each outcome falls into one of four categories from high to very low.

Systematic review and guideline authors use this ap- proach to rate the quality of evidence for each outcome

across studies (i.e., for a body of evidence). This does not mean rating each study as a single unit. Rather, GRADE is ‘‘outcome centric’’: rating is made for each outcome, and quality may differdindeed, is likely to differdfrom one outcome to another within a single study and across a body of evidence.

For example, in a series of unblinded RCTs measuring both the occurrence of stroke and all-cause mortality, it is possible that strokedmuch more vulnerable to biased judg- mentsdwill be rated down for risk of bias, whereas all- cause mortality will not. Similarly, a series of studies in which very few patients are lost to follow-up for the out- come of death, and very many for the outcome of quality of life, is likely to result in judgments of lower quality for the latter outcome. Problems with indirectness may lead to rating down quality for one outcome and not another within a study or studies if, for example, fracture rates are measured using a surrogate (e.g., bone mineral density) but side effects are measured directly.

6. The GRADE processdgrading recommendations

Guideline developers (but not systematic reviewers) then review all the information to make a final decision about which outcomes are critical and which are important and come to a final decision regarding the rating of overall qual- ity of evidence.

Fig. 2. Quality assessment criteria.

386 G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

Guideline (but not systematic review) authors then con- sider the direction and strength of recommendation. The balance between desirable and undesirable outcomes and the application of patients’ values and preferences deter- mine the direction of the recommendation and these fac- tors, along with the quality of the evidence, determine the strength of the recommendation. Both direction and strength may be modified after taking into account the re- source use implications of the alternative management strategies.

7. The endpoint of the GRADE process

The endpoint for systematic reviews and for HTA re- stricted to evidence reports is a summary of the eviden- cedthe quality rating for each outcome and the estimate of effect. For guideline developers and HTA that provide advice to policymakers, a summary of the evidence repre- sents a key milestone on the path to a recommendation.

The GRADE working group has developed specific ap- proaches to presenting the quality of the available evidence, the judgments that bear on the quality rating, and the effects of alternative management strategies on the outcomes of interest. We will now summarize these approaches, which we call the GRADE EP and the SoFs table. In doing so, we are taking something of a ‘‘flashback’’ approach to this series of articles: we begin by presenting the conclusion of the evidence summary process and will then go back to de- scribe in detail the steps that are required to arrive at that conclusion.

8. What is the difference between an EP and a SoFs table?

An EP (Table 1) includes a detailed quality assessment in addition to a SoFs. That is, the EP includes an explicit judgment of each factor that determines the quality of

evidence for each outcome (Fig. 2), in addition to a SoFs for each outcome. The SoF table (Table 2) includes an as- sessment of the quality of evidence for each outcome but not the detailed judgments on which that assessment is based.

The EP and the SoF table serve different purposes and are intended for different audiences. The EP provides a re- cord of the judgments that were made by review or guide- line authors. It is intended for review authors, those preparing SoF tables and anyone who questions a quality assessment. It helps those preparing SoF tables to ensure that the judgments they make are systematic and transpar- ent and it allows others to inspect those judgments. Guide- line panels should use EPs to ensure that they agree about the judgments underlying the quality assessments and to establish the judgments recorded in the SoF tables.

SoF tables are intended for a broader audience, includ- ing end users of systematic reviews and guidelines. They provide a concise summary of the key information that is needed by someone making a decision and, in the context of a guideline, provide a summary of the key information underlying a recommendation. GRADEpro computer soft- ware facilitates the process of developing both EPs and SoFs tables [10].

9. More than one systematic review may be needed for a single recommendation

Figure 1 illustrates that evidence must be summarizedd the summaries ideally coming from optimally conducted systematic reviewsdfor each patient-important outcome. For each comparison of alternative management strategies, all outcomes should be presented together in one EP or SoFs table. It is likely that all studies relevant to a health care question will not provide evidence regarding every outcome. Figure 1, for example, shows the first study pro- viding evidence for the first and second outcome, the

Table 1

GRADE evidence profile: antibiotics for children with acute otitis media

Quality assessment Summary of findings

Number of patients Absolute risk

No of studies

(Design) Limitations Inconsistency Indirectness Imprecision

Publication

bias Placebo Antibiotics

Relativ risk

(95% C

Control

risk a

Risk

difference

(95% CI) Quality

Pain at 24h

5 (RCT) No serious

limitations

No serious

inconsistency

No serious

indirectness

No serious

imprecision

Undetected 241/605 223/624 RR 0.9

(0.78 1.04)

367/1,000 Not Significant 4444

High

Pain at 2e7 d 10 (RCT) No serious

limitations

No serious

inconsistency

No serious

indirectness

No serious

imprecision

Undetected 303/1,366 228/1,425 RR 0.7

(0.62 0.83)

257/1,000 72 fewer per

1,000 (44e98)

4444

High

Hearing, inferred from the surrogate outcome abnormal tympanometryd1 mo 4 (RCT) No serious

limitations

No serious

inconsistency

Serious

indirectness

(because of

indirectness of

outcome)

No serious

imprecision

Undetected 168/460 153/467 RR 0.8

(0.75 1.07)

350/1,000 Not Significant 444B

Moderate

Hearing, inferred from the surrogate outcome abnormal tympanometryd3 mo

3 (RCT) No serious

limitations

No serious

inconsistency

Serious

indirectness

(because of

indirectness of

outcome)

No serious

imprecision

Undetected 96/398 96/410 RR 0.9

(0.76 1.24)

234/1,000 Not Significant 444B Moderate

Vomiting, diarrhea, or rash

5 (RCT) No serious

limitations

Serious

inconsistency

(because of

inconsistency in

absolute

effects)

No serious

indirectness

No serious

imprecision

Undetected 83/711 110/690 RR 1.3

(1.09 1.76)

113/1,000 43 more per

1,000 (10e86)

444B

Moderate

Abbreviations: GRADE, Grading of Recommendations Assessment, Development, and Evaluation; RCT, randomized controlled trials; CI, confid ce interval; RR, risk ratio. a The control rate is based on the median control group risk across studies.

3 8 7

G . G u y a tt e t a l. / Jo u rn a l o f C lin

ic a l E p id e m io lo g y 6 4 (2 0 1 1 ) 3 8 3 e 3 9 4

e

I)

e

2

e

9

e

7

e

8

e

en

Table 2

Summary of finding: antibiotics for acute otitis media in children

Antibiotics compared with placebo for acute otitis media in children

Patient or population: Children with acute otitis media

Setting: High- and middle-income countries

Intervention: Antibiotics

Comparison: Placebo

Outcomes

Estimated risks (95% CI)

Relative effect (95% CI)

No. of Participants

(studies)

Quality of the

evidence (GRADE) Comments

Control risk a

Intervention risk

Placebo Antibiotics

Pain at 24h 367 per 1,000 330 per 1,000 (286e382) RR 0.9 (0.78e1.04) 1229 (5) 4444

High

Pain at 2e7 d 257 per 1,000 185 per 1,000 (159e213) RR 0.72 (0.62e0.83) 2791 (10) 4444

High

Hearing, inferred from the surrogate

outcome abnormal

tympanometryd1 mo

350 per 1,000 311 per 1,000 (262e375) RR 0.89 (0.75e1.07) 927 (4) 444B Moderateb

Hearing, inferred from the surrogate

outcome abnormal

tympanometryd3 mo

234 per 1,000 227 per 1,000 (178e290) RR 0.97 (0.76e1.24) 808 (3) 444B

Moderate b

Vomiting, diarrhea, or rash 113 per 1,000 156 per 1,000 (123e199) RR 1.38 (1.09e1.76) 1,401 (5) 444B

Moderate c

Ideally, evidence from nonotitis trials

with similar ages and doses (not

obtained) might improve the

quality of the evidence.

Abbreviations: CI, confidence interval; RR, risk ratio; GRADE, Grading of Recommendations Assessment, Development, and Evaluation. a The basis for the control risk is the median control group risk across studies. The intervention risk (and its 95% CI) is based on the control risk in the comparison group and the relative effect of the

intervention (and its 95% CI). b Because of indirectness of outcome. c Generally, GRADE rates down for inconsistency in relative effects (which are not inconsistent in this case). Inconsistency here is in absolute effects, which range from 1% to 56%. Contributing factors to

the decision to rate down in quality include the likely variation between antibiotics and the fact that most of the adverse events come from a single study. Consideration of indirect evidence from other trials of

antibiotics in children (not undertaken) would likely further inform this issue.

3 8 8

G . G u y a tt e t a l. / Jo u rn a l o f C lin

ic a l E p id e m io lo g y 6 4 (2 0 1 1 ) 3 8 3 e 3 9 4

389G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

second study for the first three outcomes, and so on. Indeed, there may be no overlap between studies providing evi- dence for one outcome and those providing evidence for another. For instance, RCTs may provide the relevant evi- dence for benefits and observational studies for rare, serious adverse effects.

Because most existing systematic reviews do not ade- quately address all relevant outcomes (many, for instance, are restricted to RCTs), the GRADE process may require relying on more than one systematic review. Ideally, future systematic reviews will comprehensively summarize evi- dence on all important outcomes for a relevant question.

10. A single systematic review may need more than one SoFs table

Systematic reviews often address more than one compar- ison. They may evaluate an intervention in two disparate populations or examine the effects of a number of interven- tions. Such reviews are likely to require more than one SoFs table. For example, a review of influenza vaccines may evaluate the effectiveness of vaccination for different populations, such as community dwelling and institutional- ized elderly patients or for different types of vaccines.

11. An example of an EP

Table 1 presents an example of a GRADE EP addressing the desirable and undesirable consequences of use of antibi- otics for children with otitis media living in high- and middle- income countries. The most difficult judgment in this table relates to the quality of evidence regarding adverse effects of antibiotics. In relative terms, the increases in adverse ef- fects were reasonably consistent across trials. The trials, however, had very different rates of adverse effects (from 1% to 56%). Furthermore, from evidence external to the tri- als, we know that adverse effects differ across drugs (amox- icillin causes more adverse effects than penicillin). In addition, most of the events driving the increase come from a single trial which, of those included, had the highest risk of bias. The investigators recognized that ideally they would generate a summary of adverse effects from nonotitis trials with similar drug doses and patient age. Ultimately, they chose to rate down quality from high (starting high because the evidence comes from randomized trials) to moderate quality on the basis of inconsistency in absolute effects.

This dilemma faced by the investigators in making their rating of quality of evidence for adverse effects highlights two themes that will recur throughout this series. First, for many close-call judgments that are required in evaluating ev- idence, disagreement between reasonable individuals will be common. GRADE allows the pinpointing of the nature of the disagreement. Decision makers are then in a position to make their own judgments about the relevant issues.

Second, GRADE asks systematic review authors and guideline developers to consider quality of evidence under a number of discrete categories and to either rate down or not on the basis of each category (Fig. 2). Rigid adherence to this approach, however, ignores the fact that quality is actually a continuum and that an accumulation of limita- tions across categories can ultimately provide the impetus for rating down in quality. Ultimately, GRADE asks authors who decide to rate down quality by a single level to specify the one category most responsible for their decision (in this case, inconsistency of absolute effects) while documenting (as in the previous paragraph and in the footnotes in Tables 1 and 2), all factors that contributed to the final decision to rate down quality.

This presentation and the EP (Table 1) and SoF table (Table 2) illustrate another point: although we suggest stan- dard formats based on pilot testing, user testing, and evalu- ations [13e16], alternative formats may be desirable for different audiences. Indeed, the order of the columns and the presentation of the absolute risks differs in the EP and SoF we present in this article.

In subsequent articles, we will continue to present exam- ples of different formats for these tables. For both EPs and SoF tables, there is a trade-off between consistency, which facilitates their use and adaptation to address specific audi- ences or characteristics of the evidence, for example, by leaving out columns for some elements of the quality assessment or presenting the findings in a different way. Furthermore, EPs and SoF tables focusing on continuous variables and those addressing diagnostic questions may re- quire a different format. Finally, the user testing conducted thus far is limited, and further testing may generate differ- ing findings.

We suggest, however, that a few items should be in- cluded in all evidence summaries. For example, all EPs should include a row for each patient-important outcome. Typically, each row should include columns for the number of studies and the number of participants, the study design (randomized trials or observational studies), relevant fac- tors that determine the evidence quality (Fig. 2), the overall judgment of quality (high, moderate, low, or very low) for that outcome, and estimates for the relative and absolute effects of the intervention.

12. An example of a SoFs table

Table 2 presents a SoF table in the format we recom- mend on the basis of pilot testing, user testing, and evalua- tions [10,12,13]. The Appendix presents an explanation of the terms found in the SoF table and the EP.

A SoF table presents the same information as the full EP, omitting the details of the quality assessment and adding a column for comments. The logic of the order of the columns is their importancedmore important in the first columns and less important in the later. Aside from

Table 3

Examples of best practice statements and statements that could be confused with motherhood statements

Recommendations that

are not helpful Explanation

Recommendations that

may be helpful but do

not need grading Explanation

Recommendations that

need grading Explanation

In patients presenting with

chronic heart failure, take

a careful and detailed history

and perform a clinical

examination.

‘‘Careful and detailed history’’

is neither specific nor

actionable.

In patients presenting with heart

failure, initial assessment

should be made of the

patient’s ability to perform

routine/desired activities of

daily living (LOE: C).

The alternative: initial

assessment excluding

ascertainment of ability to

perform routine activities is

not credible.

In patients with hypertension,

the PE should include

auscultation for carotid,

abdominal, and femoral

bruits.

This recommendation is

specific, but may be a waste

of time, or lead to positive

results that lead to fruitless,

resource-consuming

investigation.

In patients with hypertension,

the PE should include an

appropriate measurement of

BP, with verification in the

contralateral arm.

It is not clear what exactly the

authors mean by ‘‘appropriate

measurement of BP.’’

Pregnant women should be

offered evidence-based

information and support to

enable them to make

informed decisions regarding

their care, including details of

where they will be seen and

who will undertake their care

(LOE: C).

Most would consider a

recommendation to not offer

such information a violation

of basic standards of care.

In patients with diabetes,

monofilaments should not be

used to test more than 10

patients in one session and

should be left for at least 24h

to ‘‘recover’’ (buckling

strength) between sessions

(LOE: C).

If there is only very low-quality

evidence to support such a

recommendation, clinicians

should be aware of this, and

the recommendation should

be weak.

All patients should undergo PE

to define the severity of the

hospital-acquired pneumonia,

to exclude other potential

sources of infection and to

reveal specific conditions that

can influence the likely

etiologic pathogens (level II).

The elements of a PE that are

necessary to reveal conditions

that can influence the likely

pathogens is uncertain.

Routinely record the daytime

activities of people with

schizophrenia in their care

plans, including occupational

outcomes.

A recommendation to omit

recording such activities is

not credible.

Monitoring for the development

of diabetes in those with

prediabetes should be

performed every year

(LOE: E).

The alternative should be

specified (is it more

frequently, less frequently, or

not at all?). Specifying the

alternative would make it

evident that formal grading is

desirable.

In patients presenting with

a seizure, a PE (including

cardiac, neurological, and

mental state) and

developmental assessment,

where appropriate, should be

carried out (LOE: C).

It is unclear what makes the

particular aspects of PE or

developmental assessment

appropriate.

When working with caregivers

of people with schizophrenia:

provide written/verbal

information on schizophrenia

and its management,

including how families/

caregivers can help through

all phases of treatment.

Although randomized trials of

specific educational programs

may be warranted, a trial in

which the basic information

described here is withheld

would be unacceptable.

Perform the A1C test at least

two times a year in patients

who are meeting treatment

goals (and who have stable

glycemic control) (LOE: E).

The alternative should be

specified (is it more

frequently, less frequently, or

not at all?). Specifying the

alternative would make it

evident that formal grading is

desirable.

Health care professionals should

facilitate access as soon as

possible to assessment/

treatment and promote early

access throughout all phases

of care.

The specific actions required to

facilitate access are not

specified and thus obscure.

Abbreviations: BP, blood pressure; PE, physical examination; LOE, level of evidence.

3 9 0

G . G u y a tt e t a l. / Jo u rn a l o f C lin

ic a l E p id e m io lo g y 6 4 (2 0 1 1 ) 3 8 3 e 3 9 4

391G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

a different order of columns, the SoF table (Table 2) pres- ents the absolute risks in intervention and control groups with a CI around the intervention group rate, while the EP (Table 1) presents the risk difference with an associated CI. In addition, for nonsignificant outcomes (e.g., hearing, inferred from the surrogate outcome tympanometry) for the absolute risk difference, the EP notes only that results are nonsignificant, whereas the SoF table provides a CI around the intervention event rate.

The suggested format for SoF tables represents a com- promise between simplicity (to make the information as easily accessible as possible to a wide audience) and com- pleteness (to make the information and the underlying judg- ments as transparent as possible). When this format is used, judgments must still be made about what information to present (e.g., which outcomes and what levels of risk) and how to present that information (e.g., how to present continuous outcomes). As we have noted, although we en- courage the use of this or a similar format and consistency, those preparing SoF tables should consider their target audience and the specific characteristics of the underlying evidence when deciding on the optimal format for a SoF ta- ble. Future editions of GRADEpro will include additional options for the preparation of EPs and SoF tables reflecting this flexibility [10].

13. Modifications of GRADE

Some organizations have used modified versions of the GRADE approach. We recommend against such modifica- tions because the elements of the GRADE process are inter- linked because modifications may confuse some users of evidence summaries and guidelines, and because such changes compromise the goal of a single system with which clinicians, policy makers, and patients can become familiar.

14. GRADE’s Limitations

Those who want to use GRADE should consider five im- portant limitations of the GRADE system. First, as noted previously, GRADE has been developed to address ques- tions about alternative management strategies, interven- tions, or policies. It has not been developed for questions about risk or prognosis, although evidence regarding risk or prognosis may be relevant to estimating the magnitude of intervention effects or providing indirect evidence link- ing surrogate to patient-important outcomes.

Second, attempted application of GRADE to an ill- defined set of recommendations that one may call ‘‘mother- hood statements’’ or ‘‘good practice recommendations’’ will prove problematic. A guideline panel may want to issue such recommendations relating to interventions that represent necessary and standard procedures of the clinical encounter or health care systemdsuch as history taking and physical

examination, helping patients to make informed decisions, obtaining written consent, or the importance of good commu- nication. Some of these recommendations may not be help- ful, and when they are helpful, it may not be a useful exercise to rate the quality of evidence or grade the strength of the recommendations. Other recommendations may be confused with good practice recommendations but may in fact require grading.

Recommendations that are unhelpful include those that aretoovaguetobe implemented(e.g.,‘‘takeacomprehensive history’’ or ‘‘complete a detailed physical examination’’). Some interpretations of such recommendations might lead to inefficient or counterproductive behavior. Guideline panels should issue recommendations only when they are both specific and actionable.

Recommendations that may be helpful but do not need grading are typically those in which it is sufficiently obvi- ous that desirable effects outweigh undesirable effects that no direct evidence is available because no one would be foolish enough to conduct a study addressing the implicit clinical question. Typically, such recommendations are sup- ported by a great deal of indirect evidence, but teasing out the nature of the indirect evidence would be challenging and a waste of time and energy. One way of recognizing such questions is that if one made the alternative explicit, it would be bizarre or laughable.

Procedures may be sufficiently ingrained in standard clinical practice that guideline panels would be inclined to consider them good practice recommendations when in fact a dispassionate consideration would suggest that legit- imate doubt remains regarding the balance of desirable and undesirable consequences. Such recommendations should undergo formal rating of quality of evidence and grading of strength of recommendations. Table 3 provides examples of unhelpful good practice recommendations, helpful good practice recommendations, and recommendations that might be confused with good practice recommendations but require rating of quality of evidence and grading of recommendations.

Third, as illustrated in Fig. 3, preparing a guideline en- tails several steps both before and after those steps to which the GRADE system applies. It is important for review au- thors and guideline developers to understand where GRADE fits into the overall process and to look elsewhere for guidance related to those other steps [17,18]. We do, however, in later articles in this series, provide our view of how the GRADE system is best implemented in the con- text of these other steps.

Fourth, the overwhelming experience with GRADE thus far is in evaluation of preventive and therapeutic interven- tions and in addressing clinical questions rather than public health and health systems questions. Those applying GRADE to questions about diagnostic tests, to public health, or to health systems questions will face some spe- cial challenges [8,19]. We will address these challenges, particularly those related to diagnostic tests, later in this

Fig. 3. Where GRADE fits in to the guideline development process. Abbreviation: GRADE, Grading of Recommendations Assessment, Development, and

Evaluation. a Some aspects of the development and appraisal of systematic reviews fall clearly within the GRADE process and others do not. Particularly

relevant to GRADE are the definition of the question and most particularly the definition of the outcomes, including the choice of the most important;

the specification of a priori hypotheses to explain possible heterogeneity; and the interpretation of the results, in particular the generation of estimates of

absolute effect and the interpretation of subgroup analyses.

392 G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

series. Aware that work remains to be done in refining the GRADE process and addressing areas of uncertainty, the GRADE working group continues to meet regularly and continues to welcome new members to participate in the discussions.

Finally, GRADE will disappoint those who hope for a framework that eliminates disagreements in interpreting evidence and in deciding on the best among alternative courses of action. Although the GRADE system makes judgments about quality of evidence and strength of recom- mendations in a more systematic and transparent manner, it does not eliminate the need for judgments.

15. Where from here

The next article in this series will describe GRADE’s ap- proach to framing the question that a systematic review or guideline is addressing and deciding on the importance of outcomes. The next set of articles in the series will address in detail the decisions required to generate EPs and SoF ta- bles, such as those presented in Tables 1 and 2. The series will then address special challenges related to diagnostic tests and resource use and the process of going from evidence to rec- ommendations. The series will conclude by commenting on issues of applying GRADE in guideline panels.

Appendix. Explanations for SoFs tables (Table 2) and EPs (Table 1)

Examples from table Explanations

Outcomes Outcomes

The tables provide the findings for the most important outcomes for someone making a decision. These include

potential benefits and harms, whether the included studies provide data for these outcomes or not. Additional

findings may be reported elsewhere in the review.

Absolute risks Absolute risks

Risk is the probability of an outcome occurring. The estimated risks columns in the SoF table present the best

estimate of the risk in the control group (control risk in the EP) and the risk in the intervention group

(intervention risk antibiotics), with a CI around the risk in the intervention group. If one wants to know the

difference in absolute risk or the CI around the difference in risk, this requires subtraction. In the EP, the risk

difference is presented directly.

185 per 1,000 (159e213) Confidence interval

ACI is a range around an estimate that conveys how precise the estimate is; in this example, the result is the estimate

of the intervention risk (see in the following). The CI is a guide to how sure we can be about the quantity we are

interested in (here the true absolute effect). The narrower the range between the two numbers, the more confident

we can be about what the truevalue is; thewider the range, the less surewe can be. Thewidth of the CI reflects the

extent to which chance may be responsible for the observed estimate (with a wider interval reflecting more

chance).

(95% CI) 95% CI

As explained previously, the CI indicates the extent to which chance may be responsible for the observed numbers.

In the simplest terms, a 95% CI means that we can be 95% confident that the true size of effect is between the

lower and upper confidence limit (e.g., 0.62 and 0.83 in the example of a relative effect of pain at 2e7 d in Table

2). Conversely, there is a 5% chance that the true effect is outside of this range.

Estimated risk control

257 per 1,000

Estimated control risk (without the intervention)

Estimated risks control (control rate in the EP) are typical rates of an outcome occurring without the inter-

vention. They will ideally be based on studies of incidence in representative populations. Alternatively, if such

observational studies are not available, they can be based on control group risks in comparative studies. When

only one control group risk is provided, it is normally the median control group risk across the studies that

provided data for that outcome.

In this example (pain at 2e7 d), the risk of 257 events occurring in every 1,000 people indicates what would

happen in a typical control group population. When relevant, the tables will provide information for more than

one population, for instance differentiating between people at low and high risk when there are potentially

important differences.

Intervention risk antibiotics 185

per 1,000 (159e213)

Intervention risk

In this example, the estimated risk in the control group was 257 events in every 1,000 persons. Implementing the

intervention in this population would result in a intervention intervention group risk of 185 events in every

1,000 people, given the pooled risk ratio (RR) across studies. If the table provides more than one control risk

for an outcome, for instance differentiating between people at low and high risk, then a intervention risk is

provided for each population.

Determining the effect of the intervention requires subtraction. In the EP, the subtraction has been done for you.

The intervention results in 72 fewer children in every 1,000 experiencing pain at 2e7 d.

Relative effect (95% CI) Relative effect or RR

Relative effects are ratios. Here the relative effect is expressed as a RR.

Risk is the probability of an outcome occurring. A RR is the ratio between the risk in the intervention group and the

risk in the control group. If the risk in the intervention group is 1% (10 per 1,000) and the risk in the control group

is 10% (100 per 1,000), the relative effect is 10/100 or 0.10.

If the RR is exactly 1.0, this means that there is no difference between the occurrence of the outcome in the

intervention and the control group. It is unusual for the RR to be exactly 1.0, and what it means if it is above

or below this value depends on whether the outcome being counted is judged to be good or bad.

If the RR is greater than 1.0, the intervention increases the risk of the outcome. If it is a good outcome (for

example, the birth of a healthy baby), an RR greater than 1.0 indicates a desirable effect for the intervention;

whereas, if the outcome is bad (for example, death), an RR greater than 1.0 would indicate an undesirable

effect.

If the RR is less than 1.0, the intervention decreases the risk of the outcome. This indicates a desirable effect, if it

is a bad outcome (for example, death) and an undesirable effect if it is a good outcome (for example, birth of

a healthy baby).

RR 0.72 (0.62e0.83)

The mean edema score in the

intervention groups was on

average 4.7 lower (95%

CI �4.5, �4.9).

There are no mean scores in this example (but this is what it would look like if there were).

(Continued)

393G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

Appendix. Continued

Examples from table Explanations

2,791 (10 studies) Number of participants (studies)

The table provides the total number (no.) of participants across studies (2,791 in this example) and the number of

studies (10) that provided data for that outcome. This indicates how much evidence there is for the outcome.

The EP includes columns that provide the number of events and number of patients, in each of the control

(241/1,605) and intervention (223/1,624) groups

Quality of the evidence

(GRADE)

Quality of the evidence

The quality of the evidence is a judgment about the extent to which we can be confident that the estimates of

effect are correct. These judgments are made using the GRADE system and are provided for each outcome.

The judgments are based on the type of study design (randomized trials vs. observational studies), the risk of

bias, the consistency of the results across studies, and the precision of the overall estimate across studies. For

each outcome, the quality of the evidence is rated as high, moderate, low, or very low.

A blank space indicates that the information is not relevant.

What is the difference between the risks presented in the shaded columns and the relative effect?

The effect of an intervention can be described by comparing the risk of the control group with the risk of the

intervention group. Such a comparison can be made in different ways.

One way to compare two risks is to calculate the difference between the risks. This is the absolute effect. The

absolute effect can be found in the SOFs table by calculating the difference between the numbers in the

shaded columnsdthe control risk in the control group on the left and the intervention risk in the intervention

group on the right. The EP does the subtraction for you.

Here is an example: Consider the risk for blindness in a patient with diabetes over a 5-year period. If the risk for

blindness is found to be 20 in 1,000 (2%) in a group of patients treated conventionally and 10 in 1,000 (1%) in

patients treated with a new drug, the absolute effect is derived by subtracting the intervention group risk from

the control group risk: 2%�1% 5 1%. Expressed in this way, it can be said that the new drug reduces the 5-year risk for blindness by 1% (absolute effect is 10 fewer per 1,000).

Another way to compare risks is to calculate the ratio of the two risks. Given the data above, the relative effect is

derived by dividing the two risks, with the intervention risk being divided by the control risk: 1% O 2% 5 ½

(0.50). Expressed in this way, as the ‘‘relative effect,’’ the 5-year risk for blindness with the new drug is

one-half the risk with the conventional drug.

Here the table presents risks as times per 1,000 instead of percentage, as this tends to be easier to understand.

Whenever possible, the table presents the relative effect as the RR.

Usually the absolute effect is different for groups that are at high and low risk, whereas the relative effect often is

the same. Therefore, when it is relevant, GRADE tables report risks for groups at different levels of risk.

394 G. Guyatt et al. / Journal of Clinical Epidemiology 64 (2011) 383e394

References

[1] Field M, Lohr K. Clinical practice guidelines: directions for a new

program. Washington, DC: National Academic Press; 1990.

[2] Schunemann HJ, Woodhead M, Anzueto A, Buist S, Macnee W,

Rabe KF, et al. A vision statement on guideline development for re-

spiratory disease: the example of COPD. Lancet 2009;373:774e9.

[3] Guyatt G, Vist G, Falck-Ytter Y, Kunz R, Magrini N, Schunemann H.

An emerging consensus on grading recommendations? ACP J Club

2006;144(1):A8e9. [4] Guyatt GH, Oxman AD, Kunz R, Jaeschke R, Helfand M, Liberati A,

et al. Incorporating considerations of resources use into grading rec-

ommendations. BMJ 2008;336:1170e3.

[5] Guyatt GH, Oxman AD, Vist GE, Kunz R, Falck-Ytter Y,

Alonso-Coello P, et al. GRADE: an emerging consensus on rating

quality of evidence and strength of recommendations. BMJ

2008;336:924e6. [6] Guyatt GH, Oxman AD, Kunz R, Vist GE, Falck-Ytter Y,

Schunemann HJ. What is ‘‘quality of evidence’’ and why is it impor-

tant to clinicians? BMJ 2008;336:995e8.

[7] Guyatt GH, Oxman AD, Kunz R, Falck-Ytter Y, Vist GE, Liberati A,

et al. Going from evidence to recommendations. BMJ 2008;336:

1049e51.

[8] Schunemann HJ, Oxman AD, Brozek J, Glasziou P, Jaeschke R,

Vist GE, et al. Grading quality of evidence and strength of recommen-

dations for diagnostic tests and strategies. BMJ 2008;336:1106e10.

[9] Jaeschke R, Guyatt GH, Dellinger P, Schunemann H, Levy MM,

Kunz R, et al. Use of GRADE grid to reach decisions on clinical

practice guidelines when consensus is elusive. BMJ 2008;337:

a744.

[10] Brozek J, Oxman A, Sch€unemann HJ. GRADEpro. [Computer pro- gram]. Version 3.2 for Windows. Available at http://mcmaster.

flintbox.com/technology.asp?Page53993 and http://www.cc-ims.net/

revman/gradepro. Accessed October 21, 2010.

[11] Sch€unemann H, Brozek J, Guyatt G, Oxman A. GRADE handbook for grading quality of evidence and strength of recommendation; 2010.

[12] Oxman AD, Sackett DL, Guyatt GH. Users’ guides to the medical lit-

erature. I. How to get started. The Evidence-Based Medicine Work-

ing Group. JAMA 1993;270:2093e5.

[13] Akl EA, Maroun N, Guyatt G, Oxman AD, Alonso-Coello P, Vist GE,

et al. Symbols were superior to numbers for presenting strength of

recommendations to health care consumers: a randomized trial. J Clin

Epidemiol 2007;60:1298e305.

[14] GRADE Working Group. Systems for grading the quality of evidence

and the strength of recommendations II: Pilot study of a new system.

BMC Health Serv Res 2005;5:25.

[15] Rosenbaum SE, Glenton C, Nylund HK, Oxman AD. User testing

and stakeholder feedback contributed to the development of under-

standable and useful Summary of Findings tables for Cochrane re-

views. J Clin Epidemiol 2010;63:607e19.

[16] Rosenbaum S, Glenton C, Oxman A. Evaluation of summary of find-

ings tables for Cochrane reviews. J Clin Epidemiol 2010;63:620e6.

[17] Schunemann H, Fretheim A, Oxman AD. Improving the use of re-

search evidence in guideline development: 1. Guidelines for guide-

lines. Health Res Policy Syst 2006;4:13.

[18] Cochrane Handbook for Systematic Reviews of Interventions Version

5.0.0. [www.cochrane-handbook.org].

[19] Sch€unemann H, Fretheim A, Oxman AD. Improving the use of re-

search evidence in guideline development: 9. Grading evidence and

recommendations. Health Res Policy Syst 2006;4:21.