Topic 9: Meta Analysis Flashcards
What is Meta Analysis
The statistical combination of results from two or more separate studies. A systematic review.
Why might we want to do meta analysis
Individual studies may be too small, studies may address the same question but the trials differ in some ways.
Why might individual trials be too small
Failed to recruit their target, they overestimated the MCID.
In what ways can trials that address the same research question differ
Details of treatment, outcome, patient samples
Why is meta-analysis needed in grant applications
To demonstrate a new trial is justified, and the question cannot be answered using existing data.
What might make a trial difficult to include in meta analysis
Difficult to locate, trial might not have made it into literature, the trial is currently ongoing.
What are the advantages of meta analysis
Increase the power to detect treatment differences, increased precision for estimates of treatment effect. Allows results from small inconclusive studies to contribute. Gain the ability to answer questions not posed by individual studies (eg subgroup effects). It is reassuring to see consistency across a broad range of patients.
What do we need to consider for meta analysis to make sure it isn’t misleading
Specific study designs, within study biases, variation across studies, reporting bias.
What are the preliminary steps for a meta analysis
Formulate review question, specify eligibility criteria, identify, select and criticially appraise studies, collect appropriate data, decide what would be meaningful to analyse.
What does an assessment of quality of the study mainly look at
The risk of bias in the results: for example, over or under estimating a treatment effect.
Give an example of a tool for assessing bias
Cochranes
Give reasons a trial may not be considered of high enough quality to be included in a meta analysis
Small sample size, poor measurement of characteristics, poor intervention implications, non-random treatment assignment.
How do we decide which studies to exclude
A study is given a quality score, and we exclude studies based on a quality score cut off.
What are 2 other ways to assess the quality of trials being considered for the meta analysis
Weighting studies, down weighting those of a poorer quality. Or doing a sensitivity analysis where we compare results across pre-specified analyses.
What quantity do we usually investigate when it is binary data
Risk ratio, odds ratio
What quantity do we usually investigate when it is continuous data
Mean difference, standardised mean difference
What quantity do we usually investigate when it is ordinal data
Proportional odds ratio
What quantity do we usually investigate when it is counts and rates data
rate ratio
What quantity do we usually investigate when it is time to event data
Hazard ratio
What analysis method is used for meta analysis
Meta regression analysis
How does the fixed effects model work
The underlying true treatment effect is assumed to be equal for all trials, and then the treatment effect observed in each trial is this plus a sampling error.
What distribution does the fixed effects model assume for the sampling errors
normally distributed with mean 0 and variance.
What is an alternative to the fixed effects model
The random effects model
What does the random effects model say
allows the population treatment effects to vary across trials, since the idea that each observed treatment effect estimates that same underlying quality is not necessarily plausible.
What does meta regression analysis do
Models the variation between trials investigating why results from different trials might have differred.
What is heterogeneity in meta analysis
The variation in study outcomes between studies
What is I-squared
The percentage of total variation across studies due to heterogeneity and not chance.
What can we do before deciding whether to carry out a fixed or random effects model
Test for homogeneity
What is I-squared used as an alternative for and why
Homogeneity tests. Homogeneity tests have low power.
