77 - Systematic Reviews and Meta-Analyses Flashcards
50 packet years
Equivalent of smoking a peck per day for 50 years
Definition of a systematic review
Literature review focussed on a single question
Identifies, appraises, selects and synthesises high-quality evidence relevant to question.
Narrow area of review
Meta-analysis
Statistical aspect of a systematic review
Analysis of combined data from multiple studies
Purposes of meta-analyses 1 2 3 4
1) Increase power
2) Resolve uncertainty
3) Improve estimates of effect size (precision)
4) Answer other questions
Highest level of evidence
Systematic review
How can studies for a systematic review be identified?
Use PICOT method (population, intervention, comparator, outcome, time)
MESH
Medical subject headings.
Tags on pieces of research
Data sources for systematic reviews
1
2
3
1) MEDLINE, EMBASE, CENTRAL, CINAHL, DARE, etc.
2) Reference lists of publications can be helpful
3) Grey literature - mightn’t be in mainstream literature, but might still be relevant (EG: negative findings, no statistically significant findings)
Inclusion/exclusion criteria
1
2
3
1) PICOT (population, intervention, comparator, outcome, time)
2) Other parameters (EG: only include studies over a certain sample size)
3) Need to be careful not to introduce selection bias
How are studies selected for systematic reviews? 1 2 3 4 5
1) At least two people independently select papers
2) Read all abstracts
3) Apply inclusion/exclusion criteria
4) Obtain full papers
5) Assess for quality
Example of consolidated standards for reporting trials
CONSORT guidelines for assessing trial quality
How can risk of bias be assessed?
Look at methods, see if methods are conducive to minimising bias
Example of statistical software
STATA
Key statistical issues in a meta-analysis
1
2
3
1) Outcome (weighted average effect size)
2) Weighting of individual studies
3) Heterogeneity (variability in effect sizes)
Examples of weighting of average effect size
Relative measure (RR, OR) Absolute measure (mean difference)
How can heterogeneity be measured?
Chi squared tests
Fixed-effects modelling
Forest plot
Combines data from several studies.
Compares studies, with weighting for sample size (represented by size of square beside study)
Compares, combines measures such as rate ratio of different studies
*Forest plot example
FOREST PLOT
Height of square represents sample size.
Lines beside squares represent confidence interval
Things to look for on a Forest plot
Small CI on pooled effect.
Want a good Chi squared test result in test for heterogeneity. This means higher p value, to indicate higher similarity.
Heterogeneity 1 2 3 4
1) Validity of meta-analysis relies on whether component studies are similar enough to be pooled
2) Need statistical effect sizes and variances to be similar
3) Need non-statistical similarities (EG: PICOT)
4) Non-statistical heterogeneity cannot be objectively assessed
PRISMA
Preferred-reporting items for systematic reviews and meta-analyses.
A checklist, flowchart to illustrate how a systematic review is to be carried out.
Diagnostic tests
Used when someone is suspected to have a condition, and you want to test for it. Test to confirm specific cause of problem.
Screening tests
Routinely test for an outcome in an at-risk population, EG: mammography for breast cancer in over-50 women
Why are positive and negative predictive values reliant on the prevalence of disease?
If not many cases of an outcome, can make positive predictive value very small
Positive predictive value
Proportion of those who tested positive and had the disease
Negative predictive value
Proportion of those who tested negative who didn’t have disease