Meta-analysis Flashcards
Meta-analysis
- A systematic method for combining the results of multiple similar studies addressing a similar clinical question
- Allows more accurate conclusions to be drawn from a larger pooled number of participants
Aim of meta-analysis
To estimate the treatment effect with the greatest possible power and precision
Meta-analysis increases the sample size
• Reduces the risk of type 2 error/ false negative results
• Produces estimates that better approximate the population parameters
• Standard error = standard deviation/ √sample size
• 95% Confidence intervals = sample mean ± (standard error x 1.96)
• i.e. the bigger the sample size the smaller the standard error and confidence intervals and the more precise the estimate
Process
- Set the study question
- Literature search
- Study selection
- Data extraction and quality assessment
- Statistical analysis
Places to look for study
• Electronic medical databases e.g.
– Embase, Medline
• Published books, conference material
• Relevant internet sites e.g. specialist societies, guidelines
• Research registries, unpublished studies
• Research and clinical affairs departments of pharmaceutical companies
• General internet search e.g. Google/ google scholar
Study selection
• Pre-defined criteria to include: – Study types – Participant inclusion and exclusion criteria – Treatment and control – Treatment time
Publication bias
• Journals prefer to publish trials with significant positive findings rather than negative or indifferent trial results
– Need to try and identify unpublished trials
• Funnel plot – used to try and identify the existence of publication bias
• Funnel plot
used to try and identify the existence of publication bias
Statistical analysis
• Combines information from different studies – Aggregate trial results – Individual patient data level • Basic requirements – Common measure of treatment effect • Odds or risk ratio • Mean difference – Variance or standard error of treatment effect must be calculated for each study
fixed effect model
assumes meta-analysis is trying to estimate an overall treatment effect
used where studies match closely in design and methodology
variability within studies but not between
trials contribute to estimat eaccording to their weight
random effects model
assumes a different underlying treatment effecr for each study
used where studies do not match
variability within and between studies
given mroe weight to smaller studies
fixed effect model details
assumes eacht rial provides an estimate of the same population value
requires estimate of treatment effect and variance for each trial
trials are weighted
big trials have low standard error hence carry more weight
weight = 1/variance
1/se^2
pooled estimates
for each trial estimate of treatment, effect is multiplied by its trial weight
weighted estimates are added and total divided by sum of weights
null hypothesis - pooled estimate is 0
testing for heterogenity
q test and i2
q test (chi)
tests null hyptothesis that all studies have same treatment effect
i2
tests how much heterogeneity there is
z
tests the null hypothesis that effect size is 0
tau t - for random effects model
estimate of variance between studies in random effects model
limitations of meta-analysis
• Only as good as the available studies
– e.g. limited if studies are small, poorly performed, very heterogeneous, have poor internal validity
• Many potential sources of bias
– Study selection, missing studies (more likely to be ‘negative’), bias within individual studies, choice of statistical methods
• Pooled estimates may not be directly meaningful/ applicable to real life clinical practice
• Complex
– Requires expertise to navigate the process
addressing limitations of meta-analysis
- Clear protocol before meta-analysis starts
- Full transparent reporting of methods and results including study selection
- Awareness and discussion of bias
- Developing expertise in conduct and interpretation of meta-analysis
