First Part.MA.Oct202010

Meta-analysis

Adrian V. Hernandez, M.D., Ph.D.Assistant Professor of Medicine

Quantitative Health Sciences

October 21, 2010

Meta-analysis l October 10, 2012 l 2

FIRST PART (40 minutes)

Introduction, objectives, types of meta-analysis, definition of research question, getting information, inclusion/exclusion criteria

Break 10 minutes

OUTLINE

Meta-analysis l October 10, 2012 l 3

OUTLINE (2)

SECOND PART: 50 MINUTES

• Analysis (models, methods, heterogeneity, publication bias, quality, subgroup analysis)

• Reporting of meta-analysis (PRISMA, MOOSE guidelines)

Meta-analysis l October 10, 2012 l 4

META-ANALYSIS: FACTS

• Too much information is available

• Many meta-analyses published lately: 1989-1993: 1301 1994-1998: 2532 1999-2003: 4917 2004-2008: 10567

• Why meta-analyses?Saves money and effortEvaluates limitations of the evidenceDesigns future researchProvides evidence for regulatory processes

Meta-analysis l October 10, 2012 l 5

OBJECTIVES

• Summarize and integrate results of studies

• Analyze differences among studies

• Overcome small sample sizes

• Increase precision of effects

• Evaluate effects in subsets of patients

• Generate new hypotheses

Meta-analysis l October 10, 2012 l 6

TYPES OF META-ANALYSES

• Randomized controlled trials (RCTs)

• Observational Studies

• Diagnostic studies

Meta-analysis l October 10, 2012 l 7

CRITICAL ISSUES

• Identification and selection of studies

• Heterogeneity of results

• Analysis of data

• Reporting of results

• Interpretation of published results

Meta-analysis l October 10, 2012 l 8

IDENTIFICATION AND SELECTIONOF STUDIES

• The most critical step of a meta-analysis

• Clearly specified in protocol

• Phases: 1. Definition of research question

2. Literature search

3. Choice of relevant studies

Meta-analysis l October 10, 2012 l 9

DEFINITION OF RESEARCH QUESTION

• Are the beneficial and harmful effects of glycoprotein IIb/IIIa receptor blockers similar between younger and older NSTE-ACS patients?

• What is the risk of HF with the use of rosiglitazone and pioglitazone in patients at high risk of DM and with type 2 DM?

• Which are the risk factors associated with hypercapnia in obese patients with OSA and without COPD?

Meta-analysis l October 10, 2012 l 10

LITERATURE SEARCH

• Pubmed-Medline (www.pubmed.gov)

• Embase (www.embase.com)

• Ovid-Medline (www.ovid.com)

• The Web of Science (isiknowledge.com)

• Cochrane Library (www.cochrane.org)

• Scopus (www.scopus.com)

• Google Scholar (scholar.google.com)More on: Steinbrook R. NEJM 2006; 354:4-7.

Meta-analysis l October 10, 2012 l 11

LITERATURE SEARCH - BIASES

• Publication bias

• Search bias

• Selection bias

Meta-analysis l October 10, 2012 l 12

PUBLICATION BIAS

Positive results are more likely to be published than negative results

Meta-analysis l October 10, 2012 l 13

PUBLICATION BIAS (2)

Meta-analysis l October 10, 2012 l 14

PUBLICATION BIAS (3)

Meta-analysis l October 10, 2012 l 15

PUBLICATION BIAS (4)

How to avoid/diminish?

• Identify unpublished studies (e.g. Nissen SE et al. NEJM 2007)

• Search registries (e.g. NIH’s http://clinicaltrials.gov)

• Do not discard studies in other languages (e.g. German,

French, Spanish)

Meta-analysis l October 10, 2012 l 16

SEARCH BIAS

How happens? Limited number of search engines

Inappropriate keywords

How to avoid/diminish?

• At least 3 search engines

• Use relevant keywords and show them (e.g. for RCTs:

see Dickersin K et al. BMJ 1994; 309: 1286-91)

• Two or more researchers

Meta-analysis l October 10, 2012 l 17

SELECTION BIAS

How happens? Long list of potential articles

Selection necessary (similarity, -replication)

How to avoid/diminish?

• Define clear list of inclusion and exclusion criteria

• Two or more researchers

Meta-analysis l October 10, 2012 l 18

INCLUSION/EXCLUSION CRITERIA

• Objective

• Population studied

• Study design-Quality of data

• Sample size

• Treatment/Intervention

• Controls

• Duration of study

• Calendar time

Meta-analysis l October 10, 2012 l 19

Inclusion: phase III RCTs, Gum chewing vs. control (active/placebo) on time to flatus/LOS, elective colorectal surgery for localized disease (cancer or not), open or laparoscopic, >15 years, English, 1960-2008.

Exclusion: Non-randomized studies, surgery beyond colorectal, prior colonic surgery, emergency surgery.

J Gastrointest Surg (2009) 13:649–656DOI 10.1007/s11605-008-0756-8

Meta-analysis l October 10, 2012 l 20

COMBINATION OF STUDIES ALWAYS POSSIBLE?

Enough info?, quality?, heterogeneity?

Meta-analysis l October 10, 2012 l 21

AVAILABILITY OF INFORMATION

Summary effects only (OR, RR, HR, mean [SD]). Most of the cases → Limited analyses.

What to do? Contact authors, Patient level data

Meta-analysis l October 10, 2012 l 22

ALL STUDIES CAN BE COMBINED?

Only reasonably well conducted RCTs?

Observational studies also?

Similar results between RCTs and observational?

Similar results overtime?

Meta-analysis l October 10, 2012 l 23

MA OF RCTs vs. OBSERVATIONAL

• 3 RCTs, n=10731 diabetic/high risk of diabetes, >12 months f-up: OR 2.1 (95% CI: 1.1-4.1)

• 4 Retrospective cohorts, n=67382 diabetic patients: OR 1.6 (95% CI 1.3-1.8)

Meta-analysis l October 10, 2012 l 24

MAs OVERTIME: OBSERVATIONAL (1)

17 studies (3C, 14CC)

RR for ischemic stroke: 2.8 (95%CI 2.2-3.4)

Cohorts: 3.2 (2.0-5.3)

Case-control: 2.8 (2.2-3.5)

Meta-analysis l October 10, 2012 l 25

MAs OVERTIME: OBSERVATIONAL (2)

20 studies (4C, 16CC)

RR for all-stroke: 1.9 (95%CI 1.4-2.6)

Cohorts: 1.0 (0.5-1.8)

Case-control: 2.1 (1.6-2.9)

RR for ischemic stroke: 2.7 (95%CI 2.2-3.4)