Holger Schünemann Professor of Clinical Epidemiology , Biostatistics and Medicine

Holger Schünemann Professor of ClinicalEpidemiology, Biostatistics and Medicine McMaster University, Hamilton, Canada Italian NCI „Regina Elena“, Rome, Italy Principles guideline development and the GRADE system

Content • Describe the grade of recommendation and what each category means: strong/weak and optional language • How the quality of evidence can be upgraded/down-graded • What happens when you’re recommending something not be done? • Maybe provide some ID-type examples if possible - I’m attaching our clinical questions that may be used as examples? • Provide a quick tutorial of GRADEPro • Questions

The GRADE approach Clear separation of 2 issues: 1) 4 categories of quality of evidence: very low, low, moderate, or high quality? • methodological considerations • likelihood of systematic deviation from truth • by outcome 2) Recommendation: 2 grades – weak/conditional or strong (for or against)? • Quality of evidence only one factor • Influenced by magnitude of effect(s) – balance of benefits and harms, values and preferences, cost *www.GradeWorkingGroup.org

Implications of a strong recommendation • Patients: Most people in this situation would want the recommended course of action and only a small proportion would not • Clinicians: Most patients should receive the recommended course of action • Policy makers: The recommendation can be adapted as a policy in most situations

Implications of a weak recommendation • Patients: The majority of people in this situation would want the recommended course of action, but many would not • Clinicians: Be prepared to help patients to make a decision that is consistent with their own values/decision aids and shared decision making • Policy makers: There is a need for substantial debate and involvement of stakeholders

Answer Same type of interpretation

Determinants of quality - For body of evidence - • RCTs start high • observational studies start low • 5 factors that can lower quality • limitations of detailed design and execution • inconsistency • Indirectness/applicability • publication bias • Imprecision • 3 factors can increase quality • large magnitude of effect • all plausible confounding may be working to reduce the demonstrated effect or increase the effect if no effect was observed • dose-response gradient

Assessing the quality of evidence

1. Design and Execution • limitations • lack of concealment • intention to treat principle violated • inadequate blinding • loss to follow-up • early stopping for benefit • selective outcome reporting • Example: RCT suggests that danaparoid sodium is of benefit in treating HIT complicated by thrombosis • Key outcome: clinicians’ assessment of when the thromboembolism had resolved • Not blinded – subjective judgement

2. Inconsistency of results (Heterogeneity) • if inconsistency, look for explanation • patients, intervention, outcome, methods • unexplained inconsistency downgrade quality • Bleeding in thrombosis-prophylaxed hospitalized patients • seven RCTs • 4 lower, 3 higher risk

Example: Bleeding in the hospital Dentali et al. Ann Int Med, 2007

Judgment • variation in size of effect • overlap in confidence intervals • statistical significance of heterogeneity • I2

Heparin or vitamin K antagonists for survival in patients with cancer Akl E, Barba M, Rohilla S, Terrenato I, Sperati F, Schünemann HJ. “Anticoagulation for the long term treatment of venous thromboembolism in patients with cancer”. Cochrane Database Syst Rev. 2008 Apr 16;(2):CD006650.

Non-steroidal drug use and risk of pancreatic cancer Capurso G, Schünemann HJ, Terrenato I, Moretti A, Koch M, Muti P, Capurso L, Delle Fave G. Meta-analysis: the use of non-steroidal anti-inflammatory drugs and pancreatic cancer risk for different exposure categories. Aliment Pharmacol Ther. 2007 Oct 15;26(8):1089-99.

3. Directness of Evidence • differences in • populations/patients (mild versus severe COPD, older, sicker or more co-morbidity) • interventions (all inhaled steroids, new vs. old) • outcomes (important vs. surrogate; long-term health-related quality of life, short –term functional capacity, laboratory exercise, spirometry) • indirect comparisons • interested in A versus B • have A versus C and B versus C • formoterol versus salmeterol versus tiotropium

Directness interested in A versus B available data A vs C, B vs C Alendronate Risedronate Placebo

4. Publication Bias & Imprecision • Publication bias • number of small studies

ISIS-4Lancet 1995 I.V. Mg in acute myocardial infarction Meta-analysisYusuf S.Circulation 1993 Publication bias Egger M, Smith DS. BMJ 1995;310:752-54

Funnel plot 0 Symmetrical: No publication bias 1 Standard Error 2 3 0.1 0.3 0.6 1 3 10 Odds ratio Egger M, Cochrane Colloquium Lyon 2001

Funnel plot 0 Asymmetrical: Publication bias? 1 Standard Error 2 3 0.1 0.3 0.6 1 3 10 Odds ratio Egger M, Cochrane Colloquium Lyon 2001

ISIS-4Lancet 1995 I.V. Mg in acute myocardial infarction Meta-analysisYusuf S.Circulation 1993 Publication bias Egger M, Smith DS. BMJ 1995;310:752-54

Meta-analysis confirmed by mega-trials Egger M, Smith DS. BMJ 1995;310:752-54

Publication bias (File Drawer Problem) • Faster and multiple publication of “positive” trials • Fewer and slower publication of “negative” trials

5. Imprecision • small sample size • small number of events • wide confidence intervals • uncertainty about magnitude of effect • how to decide if CI too wide? • grade down one level? • grade down two levels? • extent to which confidence in estimate of effect adequate to support decision

Example: Bleeding in the hospital Dentali et al. Ann Int Med, 2007

Offer all effective treatments? • atrial fib at risk of stroke • warfarin increases serious gi bleeding • 3% per year • 1,000 patients 1 less stroke • 30 more bleeds for each stroke prevented • 1,000 patients 100 less strokes • 3 strokes prevented for each bleed • where is your threshold? • how many strokes in 100 with 3% bleeding?

1.0% 0

What can raise quality? 1. large magnitude can upgrade (RRR 50%) • very large two levels (RRR 80%) • common criteria • everyone used to do badly • almost everyone does well • oral anticoagulation for mechanical heart valves • insulin for diabetic ketoacidosis • hip replacement for severe osteoarthritis 2. dose response relation (higher INR – increased bleeding) 3. all plausible confounding may be working to reduce the demonstrated effect or increase the effect if no effect was observed

All plausible confounding would result in an underestimate of the treatment effect • Higher death rates in private for-profit versus private not-for-profit hospitals • patients in the not-for-profit hospitals likely sicker than those in the for-profit hospitals • for-profit hospitals are likely to admit a larger proportion of well-insured patients than not-for-profit hospitals (and thus have more resources with a spill over effect)

All plausible biases would result in an overestimate of effect • Hypoglycaemic drug phenformin causes lactic acidosis • The related agent metformin is under suspicion for the same toxicity. • Large observational studies have failed to demonstrate an association • Clinicians would be more alert to lactic acidosis in the presence of the agent

Relevant clinical question?Example from a not so common disease Clinical question: Population: Avian Flu/influenza A (H5N1) patients Intervention: Oseltamivir (or Zanamivir) Comparison: No pharmacological intervention Outcomes: Mortality, hospitalizations, resource use, adverse outcomes, antimicrobial resistance Schunemann et al. The Lancet ID, 2007

Methods – WHO Rapid Advice Guidelines for management of Avian Flu • Applied findings of a recent systematic evaluation of guideline development for WHO/ACHR • Group composition (including panel of 13 voting members): • clinicians who treated influenza A(H5N1) patients • infectious disease experts • basic scientists • public health officers • methodologists • Independent scientific reviewers: • Identified systematic reviews, recent RCTs, case series, animal studies related to H5N1 infection

Evidence Profile • Oseltamivir for treatment of H5N1 infection: - -

Oseltamivir for Avian Flu Summary of findings: • No clinical trial of oseltamivir for treatment of H5N1 patients. • 4 systematic reviews and health technology assessments (HTA) reporting on 5 studies of oseltamivir in seasonal influenza. • Hospitalization: OR 0.22 (0.02 – 2.16) • Pneumonia: OR 0.15 (0.03 - 0.69) • 3 published case series. • Many in vitro and animal studies. • No alternative that is more promising at present. • Cost: ~ 40$ per treatment course Schunemann et al. Lancet ID, 2007 & PLOS Medicine 2007

Determinants of the strength of recommendation

Example: Oseltamivir for Avian Flu Recommendation: In patients with confirmed or strongly suspected infection with avian influenza A (H5N1) virus, clinicians should administer oseltamivir treatment as soon as possible (????? recommendation, very low quality evidence). Schunemann et al. The Lancet ID, 2007

Example: Oseltamivir for Avian Flu Recommendation: In patients with confirmed or strongly suspected infection with avian influenza A (H5N1) virus, clinicians should administer oseltamivir treatment as soon as possible (strong recommendation, very low quality evidence). • Values and Preferences • Remarks: This recommendation places a high value on the prevention of death in an illness with a high case fatality. It places relatively low values on adverse reactions, the development of resistance and costs of treatment. Schunemann et al. The Lancet ID, 2007

Other explanations Remarks: Despite the lack of controlled treatment data for H5N1, this is a strong recommendation, in part, because there is a lack of known effective alternative pharmacological interventions at this time. The panel voted on whether this recommendation should be strong or weak and there was one abstention and one dissenting vote.

Strength of recommendation • “The strength of a recommendation reflects the extent to which we can, across the range of patients for whom the recommendations are intended, be confident that desirable effects of a management strategy outweigh undesirable effects.” • Strong or weak/conditional

Quality of evidence & strength of recommendation • Linked but no automatism • Other factors beyond the quality of evidence influence our confidence that adherence to a recommendation causes more benefit than harm • Systems/approaches failed to make this explicit • GRADE separates quality of evidence from strength of recommendation

Creating a new GRADEpro file

Holger Schünemann Professor of Clinical Epidemiology , Biostatistics and Medicine