Title: How to do Cochrane systematic reviews: Introductory workshop
1How to do Cochrane systematic reviews
Introductory workshop
- Dr Jeremy Grimshaw
- Canadian Cochrane Network and Centre
- University of Ottawa
2Outline of the course
- Introductions
- Introduction to systematic reviews
- Introduction to The Cochrane Collaboration
- Steps in doing a Cochrane review
- Coffee Break
- Advanced methods
- Appraising the risk of bias
- Locating and selecting studies for your
systematic review
3Introductions
- Name
- Background
- Previous experience of systematic reviews
- What you want to get out of the training?
4Session 1 Outline
- Introduction to systematic reviews
- Introduction to The Cochrane Collaboration
- Rationale for developing a protocol
- Format of a protocol
- Resources to support reviewers
5Problems of information management in healthcare
- Over 20,000 health journals published per year
- Published research of variable quality and
relevance - Research users (consumers, health care
professionals and policy makers) often poorly
trained in critical appraisal skills - Average time professionals have available to read
lt1 hour/week
6Systematic reviews
- Systematic reviews use rigorous scientific
methods to identify, assess and synthesise the
worldwide available evidence
7Systematic reviews
- A review of a clearly formulated question that
uses systematic and explicit methods to identify,
select, and critically appraise relevant
research, and to collect and analyse data from
the studies that are included in the review. - Cochrane Collaboration (2005) Glossary of Terms
in The Cochrane Collaboration
8Meta-analysis
- The use of statistical techniques in a systematic
review to integrate the results of included
studies. - Cochrane Collaboration (2005) Glossary of Terms
in The Cochrane Collaboration
9Meta-analysis
10Systematic reviews
- Systematic reviews involve
- stating the objectives of the research
- defining eligibility criteria for studies to be
included - identifying (all) potentially eligible studies
- applying eligibility criteria
- assembling the most complete dataset feasible
- analysing this dataset, using statistical
synthesis and sensitivity analyses, if
appropriate and possible - preparing a structured report of the research.
11Systematic reviews
- Reduce bias
- Reduce random error
- Explore variability
- Provide reliable basis for making decisions
- Inform and influence future research
12Traditional vs systematic reviews
- Traditional
- lack rigor
- methodology not transparent
- different reviewers reach different conclusions
- become out of date
- Systematic
- scientific rigor to minimise bias
- explicit and reproducible methodology
- regularly updated (Cochrane)
13The Cochrane Collaboration
- It is surely a great criticism of our profession
that we have not organised a critical summary, by
specialty or subspecialty, adapted periodically,
of all relevant randomized controlled trials. - Archie Cochrane, 1979
14The Prototype - Effective Care in Pregnancy and
Childbirth
15The Cochrane Collaboration
- The Cochrane Collaboration is an international
organization that aims to help people make
well-informed decisions about healthcare by
preparing, maintaining, and promoting the
accessibility of systematic reviews of the
effects of healthcare interventions.
16Structure of The Cochrane Collaboration
Centres
Review Groups
Steering Group
Consumer Network
Fields
Methods Groups
17- Promote awareness and support of The Cochrane
Collaboration on a geographical basis - support individual review authors
- support review groups in that area
- provide a unique function e.g. software
development, consumer support - link to government and other agencies
- not a production house for reviews
18Canadian Cochrane Centre staff
19- International and multidisciplinary
- Produce reviews
- Focused around health problems
- 51, covering most of health care
- Editorial base facilitates process
20Cochrane Review Groups
- Worldwide there are over 11,000 review authors
working with 51 registered review groups - 5 review groups have their editorial base in
Canada - Back Group
- Effective Practice and Organisation of Care Group
(EPOC) - Hypertension Group
- Inflammatory Bowel Disease and Functional Bowel
Disorders Group - Musculoskeletal Group
21- Represents a population, group, or type of care
that overlaps multiple review group areas - E.g., primary care, health promotion and public
health, health care of older people - Child Health Field and the Health Equity Field
are located in Canada - Ensures that their priorities are reflected in
the work of review groups
22- Develops methodology and advise the Collaboration
on how the validity and precision of systematic
reviews can be improved - Examples Statistical Methods, Non-Randomized
Studies, Information Retrieval - Reporting Bias Methods Group is located in Canada
23Cochrane Consumer Network
- Facilitates the dissemination of information to
patients, their families, friends and advocates
throughout the world - Plays an integral part to all activities within
The Cochrane Collaboration - Supports the role of consumers within The
Cochrane Collaboration - Liz Whamond (NB) current chair of The Cochrane
Consumer Network
24Cochrane reviews
- Cochrane reviews are conducted using standardised
rigorous approaches including detailed critical
appraisal of all studies - Cochrane reviews are reported using standardised
structure including abstract and consumer summary
25Steps of a Cochrane systematic review
- STEP 1 formulate the problem and register the
title with CRG - STEP 2 write protocol, submit for peer review
and publish - STEP 3 locate and select studies
- STEP 4 critically appraise studies for risk of
bias - STEP 5 collect data
- STEP 6 analyse and present results
- STEP 7 interpret results and write review
- STEP 8 improve and update review
26Title registration
- Decide which Cochrane CRG would be responsible
- relevant CRG may be obvious
- most groups are disease (or organ) focused
- if unsure which CRG, contact CCN/C for advice
- Information on CRGs
- The Cochrane Library
- www.cochrane.org
- Contact CRG directly
- Complete title registration form (from CRG)
27Tips for titles
- Titles for Cochrane reviews have a set format
- Format intervention for problem in category
- Caffeine for daytime sluggishness in adults
(hypothetical for today) - Keep it short
28Why have a protocol?
- Systematic reviews are scientific research
- Plan methods
- Reduce bias
- Access to peer review
- Avoid duplication of effort
- Rest of review follows
- focus for thinking about review
- planning and allocating tasks
29Format of a Cochrane protocol
- Standard format
- to help review authors be systematic
- to help people reading the reviews to find
information quickly
30Format of a Cochrane protocol
- Title/review authors
- Background
- Objectives
- Selection criteria
- Search strategy
- Methods
31Background, objectives and selection criteria
32Background (Handbook Section 3)
- Description of the condition and its significance
- Description of the intervention and its role in
practice - How the intervention might work
- Why it is important to do the review
33Objectives (Handbook Section 3)
- Follow naturally from the background
- What are the questions?
- does it work?
- is the effect consistent or variable?
- Questions should be clearly framed
34Lumping and splitting
- Systematic review question can be framed broadly
or narrowly - Effects of audit and feedback to improve chronic
diseases across all health care settings and
professionals - Effects of audit and feedback to improve chronic
diseases within primary care - Effects of audit and feedback to improve diabetes
care within primary care
35Lumping and splitting
- The lumping rationale - systematic reviews aim
to identify the common generalisable features
within similar interventions, minor differences
in trial design are not important. - Lumped reviews allow generalisability and
consistency of findings to be assessed across
wide range of settings and populations this
reduces risk of bias or chance results - The lumping principle is that the results of
two interventions should be combined unless there
are good grounds to believe they will have
opposing effects.
36Lumping and splitting
- The splitting rationale is that it is only
appropriate to combine trials which are very
similar in design, patient selection,
intervention characteristics and outcome
recording. - Split reviews avoid combining apples and
oranges - Reviews can be split by participants,
interventions or outcome - Very narrowly focused reviews are de facto
subgroup analyses
37Lumping and splitting
- Practical considerations
- Lumped reviews
- Logistically challenging (large number of
included studies) - Analytically challenging
- Heterogeneity expected
- Interpretation maybe challenging seeing the
woods for the trees - Split reviews
- Easier, quicker, cleaner
38Lumping and splitting
- Effects of audit and feedback (85 studies)
- Effects of audit and feedback to improve chronic
diseases across all health care settings and
professionals (18 studies) - Effects of audit and feedback to improve chronic
diseases within primary care (14 studies) - Effects of audit and feedback to improve diabetes
care within primary care (3 studies)
39Selection criteria (Handbook Section 4)
- Follow naturally from the objectives
- type of participants
- type of interventions
- type of outcomes
- type of studies
40Purpose of selection criteria
- Help you design the search
- Selecting studies
- Start thinking about the analysis
- One way to minimise bias
41Small group exercises
42Writing the methods and completing the protocol
(Handbook sections 6,7,8)
43Purpose of methods (Handbook sections 6,7,8)
- Plan what you will do before you start
- Divide work among review authors and establish
timeline - Minimise bias
- Start thinking about the analysis
44Methods (Handbook sections 6,7,8)
- Inclusion of trials
- - how?
- - by who?
45Methods(Handbook sections 6,7,8)
- Assessment of risk of bias
- - type of assessment
- - by who?
- - how to incorporate in the review?
46Methods (Handbook sections 6,7,8)
- Data extraction
- - what will be extracted?
- - by who?
- - how to deal with missing data?
47Methods (Handbook sections 6,7,8)
- Data synthesis
- - what type of synthesis to do?
- - what summary measures to use?
- - what comparisons to make?
- plans to investigate for differences in study
results (heterogeneity) or potential for missing
studies (publication bias) - plans for subgroup or sensitivity analyses?
48Writing your protocol
- Make sure all sections are completed
- - title/review authors, background, objectives,
selection criteria, search strategy, methods - Dont forget the other bits and pieces
- - contact details of review authors,
acknowledgements, conflict of interest, sources
of support, references, additional tables
49Resources for Cochrane reviewers
- The Cochrane Handbook
- Open learning support
- Revman
- Support from CRGs
- Trial search co-ordinators
- Review group co-ordinators
- Contact editor
- Support from CCN/C
- Training
- Whatever else we can do!
50After protocol is complete
- Submit to CRG for editorial process
- Internal and external peer review 1-2 months
- May receive critical comments to respond to
- Protocol published on The Cochrane Library
- Reward yourself then start the review
- Note - if you need to change your plan report
it and justify it (be transparent)
51 52Session 2 Outline
- Locating and selecting studies for your
systematic review - Appraising the risk of bias
53Locating and selecting studies for your
systematic review
54Why is a rigorous approach to searching important
- To achieve up-to-date, relevant, unbiased
systematic reviews of the effects of
interventions - To overcome problems with limited searching
- bias
- lack of precision and generalisability
55Finding as much as possible and being strategic
- Use multiple sources and start with the highest
yield - Electronic databases
- Cochrane Review Group register
- Cochrane Central Register of Controlled Trials
(CENTRAL) - MEDLINE
- EMBASE
- Other databases depending on your topic
- Reference lists of trials and reviews
- Handsearching journals and conferences
- Personal communication
56Designing a search strategy
- Work with TSC of review group
- Formulate your question (PICO)
- Identify your search terms
- Put it all together
57Formulate the question
- Translate the clinical problem into a structured
question and identify the key concepts - Participants
- Interventions
- Outcomes
- Studies
- Develop search terms (commonly based on
participants, interventions, study design)
58Searching tip
- Searching is often a matter of trying terms and
seeing what result you get - Find additional terms from the citations you find
that are helpful how are they indexed what are
the keywords used?
59Documenting your search
- Important for you, so you can remember what
youve done - Important for users of your review
- Document
- what you searched
- when you searched
- how you searched
60Additional points
- No right time to stop searching be guided by
returns and resources - Ideas for search terms from citations you know
of try them and check returns - Use a referencing database to manage citations
61Minimising bias when searching
- Dont limit search on
- language
- source (which journal, database etc)
- publication status (full article, conference
abstract, unpublished etc) - publication date
62Inclusion and exclusion of studies
- Develop and pilot a form based on your selection
criteria - Scan and retrieve possibly relevant studies
- Decide if studies meet inclusion criteria and
record reasons for exclusion - How reproducible are the decisions?
- Done by two independent review authors
- Beware of duplicates
63Appraising the risk of bias
64What is bias?
- Bias is systematic (non-random) error arising
from the design or conduct of a trial - Previously termed methodological quality
updated to susceptibility to bias or risk of
bias
65Bias vs Quality
Bias
Quality
Systematic error influencing trial estimate of
the true effect of an intervention
Based on the conduct of the trial did trialists
go to sufficient lengths to minimise potential
bias?
66Why assess susceptibility to bias?
- To determine whether biases could have affected
the results - Studies with less susceptibility to bias better
estimate the truth about the effect of an
intervention - Biased primary studies that provide misleading
results could generate misleading systematic
reviews
67(No Transcript)
68- Which biases make a difference to trial results?
69(No Transcript)
70(No Transcript)
71Target Population
Random sequence Allocation concealment
Selection
Performance
Detection
Attrition
72Target Population
Selection
Performance
Detection
Attrition
73Selection bias
- systematic differences in participant
characteristics at the start of a trial
74Selection bias
- Occurs during allocation into comparison groups
- Avoiding selection bias
- random unpredictable sequence
- allocation concealment (protection of the
sequence most important)
75Selection bias
- Adequate allocation concealment
- central computer randomisation
- on-site computer from which assignment can only
be determined after entering patient data - serially numbered, sealed, opaque envelopes
- Inadequate allocation concealment
- non opaque envelopes
- odd or even date of birth or medical record
number
76Target Population
Selection
Performance
Detection
Attrition
77Performance bias
- Occurs during the intervention phase
- Systematic differences, other than the
intervention being investigated, in the treatment
of the two groups - Participant expectations influencing their
response to treatment - Avoiding performance bias
- - blinding the provider
- - blinding the participant
Schulz KF Grimes DA 2002 Lancet
78Allocation concealment vs blinding
79Target Population
Selection
Performance
Detection
Attrition
80Detection bias
- Occurs during the follow up phase
- Systematic differences in the outcome assessment
between groups - Avoiding detection bias
- - blinding the outcome assessor
- - blinding data analyst/investigator?
81Target Population
Selection
Performance
Detection
Attrition
82Attrition bias
- Occurs over the duration of follow up
- Systematic differences in the loss of
participants to follow up between groups - Completeness of follow up
- - participants lost to follow up/ not included
in outcome assessment, could be different from
those who remained in the trial - avoiding attrition bias
- - describe proportion of participants lost to
follow-up - - use intention-to-treat analyses
83Attrition bias
- Intention-to-treat (ITT) analysis
- - once randomised, analysed
- - regardless of whether participants
- received the intervention
- deviated from the protocol
- withdrew from the study
- - maintains randomised allocation of participants
84- What is the evidence for these biases influencing
trial results?
85Empirical evidence that bias affects results
- Selection bias
- Inadequate/unclear allocation concealment
associated with exaggerated treatment effect
(average 29 increased benefit)1,2,3,4 - Performance bias
- Open trials (no blinding) exaggerate benefits
of treatment (average 14 increased benefit)
1,2,3,4
- Schulz et al, JAMA 1995
- Moher et al, Lancet, 1998
- Kjaergard et al, Ann Intern Med, 2001
- Egger et al, HTA, 2003
- Juni et al, JAMA, 1999
86Empirical evidence that bias affects results
- Detection bias
- Open outcome assessment exaggerated treatment
effects (171 and 355) - Attrition bias
- Handling of dropouts did not affect results1,5
however, compared whole group differences (not
between group differences). Not an exclusion but
important to report. - Schulz et al, JAMA 1995
- Moher et al, Lancet, 1998
- Kjaergard et al, Ann Intern Med, 2001
- Egger et al, HTA, 2003
- Juni et al, JAMA, 1999
87Assessing susceptibility to bias
- How do we assess susceptibility to bias?
- - Scales
- - Checklists
88Beware of quality scales.
- Different scales are available (gt50)
- - vary in dimensions covered, size, complexity
- - some provide composite numerical score
- Reliability and validity of many scales not
established - May lead to different conclusions regarding the
susceptibility to bias of the same trial (Juni
1999)
89Treatment estimates from high-quality trials
Treatment estimates from low-quality trials
Same group of trials divided into high and low
quality using 25 different Quality Assessment
Scales
Juni P, Witschi A, Bloch R, Egger M. The hazards
of scoring the quality of clinical trials for
meta-analysis. JAMA 1999282(11)1054-60
90How to assess susceptibility to bias
- Component approach assessment against individual
criteria
91Assessing susceptibility to bias in reviews
- Was the allocation sequence adequately generated
(randomised)? - Was the allocation adequately concealed?
- Was there blinding of participants, personnel
- Was there blinding of outcome assessors?
- Were dropouts and exclusions adequately
addressed? - Other (including selective outcome reporting and
other criteria for addressing topic-specific or
design-specific biases)
92- How do we incorporate susceptibility to bias in
meta-analysis?
93Incorporating susceptibility to bias in
meta-analysis
- Exclude trials with high susceptibility to bias
- - critical a priori cutoff for inclusion (e.g.
only trials with adequate allocation concealment
or blinding) - - may be problematic if few trials
- Sensitivity analysis
- - analyse separately above and below a certain a
priori level and compare results (does it change
conclusions?) - - for instance, analyse the impact of allocation
concealment, blinding, use of ITT analysis, on
results
94Example of sensitivity analysis
95But beware
- Scant empirical evidence about which biases are
important - Reporting of susceptibility to bias is not always
the same as the susceptibility to bias in the
conduct of the trial - - allocation concealment unclear in 87 of
trials as per published report (Pildal et al,
2005) - - allocation concealment unclear in about 80 of
trials (Middleton 2001) - CONSORT statement (1994, 2001) www.consort-stateme
nt.org - Possible discrimination against older trials
96Important points for your review
- Plan criteria and how you will incorporate
susceptibility to bias at the protocol stage - Assessment undertaken once trials are included
and prior to data extraction - 2 independent review authors