STATS Lec 20- Systematic reveiw and meta analysis Flashcards

1
Q

Literature reviews

A
  • Narrative
    • Provides written summary of information- e.g. introduction to a project
    • Usually done for a purpose
    • Usually involves some critical element
    • May be comprehensive and unbiased, often not
  • Systematic review
    • A research method
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Why do a systematic review

A
  • Fergusson et al (2005) effect of aprotinin on the outcome of cardiac surgery
  • A CUMULATIVE systematic review of 64 RCT- first in 1987
  • Meta-analysis of first 12 trials (1987-1992)- Statistically significant effect of treatment
  • 52 further trials performed between 1992 and 202- would not have been required if meta-analysis had been done in 1992
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

Classical versus cumulative meta-analysis

Smoking cessation with nicotine patch at 6 months

A
  • 2nd box is a combination of all previous studies put together
  • Showing clear benefit with tight confidence intervals
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Systematic review: A research method

A
  • Process of identifying, interpreting and summarising information from studies in order to answer a structured question
  • Doing an experiment with existing data rather than with your own data- Prof Li Wan Po
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Compare these two

A
  • Narrative review
    • No defined method
    • Exploratory, creative
    • May be biased
    • Introduction to a project
  • Systematic review
    • Rigorous method
    • Transparent and replicable
    • Aims to eliminate bias
    • Research method
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Meta-analysis

A
  • Statistical- combination of the results from studies to obtain a more precise estimate of treatment effect
  • Can be included as part of a systematic review- usually end part
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Key points- to do a systematic review you need to

A
  • Have a good working knowledge and understanding of the field
  • Understand the systematic review methodology
  • Be able to follow the research method in a focussed, structured and transparent way
  • Undertake a rigorous, systematic, comprehensive and exhaustive search for ALL relavent literature
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Stages of systematic review

A
  1. Scope and Mapping- refine your research question
  2. Plan search, develop protocol
  3. Search and document- comprehensive and complete
  4. Apply inclusion and exclusion criteria- refine search
  5. Quality assessment- further exlude if necessary
  6. Data extraction- compliation into summary tables
  7. Synthesis- Meta-analysis if appropriate
  8. Write-up
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

search strategy: VITAL

A
  • Databases
  • Search terms and combinations
  • Should be inclusive- identify all potentially relevant studies
  • Limitations of search- recognise the effect on conclusions
  • Unpublished data- relevance
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

Inclusion/Exclusion criteria

A
  • Include only studies relavent to question
  • Consider
    • Study design
    • Population intervention
    • Outcomes
    • QUALITY
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Hierarchy of evidence a quality issue

A
  1. Randomised controlled trial
  2. Non-randomised trial
  3. Cohort study (prospective)
  4. Case-control study (retrospective)
  5. Cross-sectional study
  6. Surveillance data
  7. Case report
  • Systematic review and meta-analysis is often considered higher than RCT because it is many RCT summarised at once
  • Further down, you move away from robust, reliable data and start to introduce bias
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Assessing quality: checklist

A
  • Published checklists available
    • Modify to suit your question
    • What is a good quality/bad quality in your field
  • How to measure ‘quality’
  • Subjective measure
    • Must be transparent
  • Quality of reporting vs. quality of the study
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

DH NSF quality assessment scale (2008)

A
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

Next stage: summarise results

A
  • Descriptive summary and tables- show off data
  • Measure of effect (+CI) for each study
    • Concept of effect size
    • Depends on question and data
  • Test for heterogeneity
    • Can data be combined from different studies
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

Outcome measures: effect size

A
  • Varies depending on research question
    • Clinical
      • Odds ratio, RR, reduction in BP
      • See slides at the end
    • Economic
      • Cost/Benefit
    • Should be appropriate to your research question
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Further analysis: Meta-analysis

A
  • Calculation of POOLED EFFECT SIZE
  • Do not simply average data from a group of trials
  • Give more weight to studies which tell us more
  • Calculate a weighted average of the treatment effects from each study
17
Q

Can you combine data

A
  • Are the same measurements made
    • Reduction in BP
    • Reduction in stroke incidence
    • Decreased mortality
  • Can the data be converted to comparable measures
    • The number needed to treat NNT
    • Positive effects / negative effects (categorisation)
  • Is the data from a homogeneous population
    • The same all the way through
18
Q

May not be homogeneous if

A
  • The study designs are different
    • You need to have evaluated this
  • The sample of patients/subjects varies between trials / studies
    • You need to have noted this for all studies
  • Different outcome measures are used
    • Again, something to record for all studies
  • The quality or size of the studies are different
    • Did you do a robust quality assessment
19
Q

Homogeneous

A
20
Q

If

A
  • Your data does look suitable
  • Meta-analysis
  • Combine results from individual trials to produce summary outcome measure
    • Pooled effect size
21
Q

Forrest plot

A
22
Q

Meta-analysis models: Fixed-effects vs. random-effects

A
  • Fixed-effects
    • Assumes that each study is measuring the same treatment effect
      • All variation due to that effect
    • Did the intervention provide benefit, on average, in the studies included
  • Random-effects
    • Assumes that there is variation in treatment effects between trials
      • Reasons for variation can be investigated
    • Will the intervention provide benefit, on average
23
Q

Done by computer

A
  • Is the data homogeneous
  • Calculate a combined effect size
24
Q

What are the advantages

A
  • Reduction of bias
  • Improved precision of estimate
  • Better understanding of seemingly heterogeneous results
  • Hypothesis generation
25
Q

Draw conclusion

A
  • Address the original question
  • Consider the strength of evidence
  • Clinical (or other real) significance
26
Q

Outcome measures: some notes to help

A
  • RR
    • A proportion of events in the treatment group/proportion events in placebo group
    • e.g. 38/294 divided by 54/232 = 0.55
  • Odds ratio (OR)
    • Odds in treatment group/odds in control group
    • e.g. 38/256 divided by 54/178 = 0.49
27
Q

Outcome measures

A
  • Absolute risk reduction (ARR)
    • Proportion events in placebo group minus the proportion of events in the treatment group described as a %
    • e.g. (54/232 - 38/294) x 100
    • 10.3% in the treatment group
  • A number needed to treat (NNT)
    • 1/ARR x 100
    • 9.7 i.e. approximately 10 people will have to be treated (for the time equivalent to trial duration) to prevent 1 event
    • Maybe useful in the report as easier to explain than OR etc
28
Q

Dichotomous vs continuous data

A
  • Dichotomous- sick or not sick
  • Continuous
    • e.g. BP, ChE level
    • Effect on public health measures in MANY ways
  • Can dichotomise continuous data
    • e.g. ChE >5 or ChE <5
    • Positive, negative or no effect on public health