Describe the features of evidence-based medicine, including
levels of evidence (e.g. NHMRC),meta-analysis, and systematic review
Randomised Control Trial
A prospective randomised controlled trial is the gold standard of experimental research.
It involves allocating patients randomly to either an intervention or a reference (control) group, and measuring the outcome of interest. Allocation can be performed in three ways:
Individuals allocated randomly. This may lead to uneven group sizes.
Allocation is performed within blocks such that group sizes will remain close in size
Groups are randomised within a category (i.e. men and women are randomised separately).
- Only study design which can establish causation
- Eliminates confounding
Randomisation controls for both known and unknown confounding factors, as these should be randomly allocated between groups.
- Blinding can be performed in a standardised fashion
- Decreases selection bias
- Not appropriate for all study designs
- Ethical concerns
e.g. Adrenaline in ALS
- Practical concerns
Small patient population or uncommon disease may cause recruitment difficulties
- Ethical concerns
The process of evaluating all of the (quality) literature to answer a specific clinical question. This:
- Does not necessarily involve statistical analysis
If it involves statistical analysis of multiple trials to generate a combined estimate of effect, it is known as a meta-analysis.
Mathematical technique of combining the results of different trials to derive a single pooled estimate of effect. Can be performed by:
- Pooling the results of each trial
- Combining all of the raw data and conducting a reanalysis
- Meta-analyses usually use random-effects models, which assumes there will be a variety of similar treatment effects
- Individual trials are summarised with an odds ratio, and weighted, usually predominantly by sample size
Stages of a [meta-analysis] and systematic review:
- Inclusion and exclusion criteria are predefined
- Search: including online databases, reference lists, citations, and experts
- Validation of potentially eligible trials (critique of interval validity, i.e. trial quality)
- [Heterogeneity Analysis]
- Reliability of result determined
i.e. Consistency across studies, statistical significance, large effect size, biological plausibility.
- Sensitivity analysis
Repeating the analysis with an alternative model, excluding borderline trials or outliers. If the result is unchanged, then the findings are robust.
For the pooling of results to be valid, the trials need to be similar. Differences between trials is known as heterogeneity. Heterogeneity can be either:
- Statistical; where the effects of the intervention are more different than would be expected to occur through chance alone. Heterogeneity analysis affects the type of model that can be used (fixed or mixed effects) and highly heterogenous data is not appropriate for meta-analysis.
- Clinical; where, due to trial design, it would be inappropriate to pool the results. For example, conducting a meta-analysis on the effects of the same drug in a paediatric and adult population would be inappropriate, as these are two different populations.
- Methodological; Where the methods used in different trials are too different to allow pooling of the data.
Results of meta-analyses are presented in a blobbogram, or more boringly, a Forest Plot.
- The x-axis plots the odds ratio, remembering that an OR of 1 indicates no difference
- The y-axis lists the studies included, and the overall summary statistic
- The square indicates the point estimate (from its x-location) and the weight given to the study (by its size)
- The horizontal line indicates the upper and lower bounds of the confidence interval
- The diamond indicates the overall point estimate and (by its width) the confidence interval for the point estimate
- The result of the heterogeneity test should also be displayed. P < 0.1 indicates significant heterogeneity.
A graphical tool to detect publication bias. Due to statistical power, larger studies should be a closer representation of the true effect. When evaluating an number of studies, one would expect that large studies cluster around the 'true effect' and smaller studies to have more scatter.
Strengths and weaknesses of meta-analyses
|Enhanced precision of estimates of effect||Publication bias|
|Useful when large trials have not been done or are not feasible||Duplicate publication|
|Generate clinically relevant measures (NNT, NNH)||Heterogeneity|
|Inclusion of outdated studies|
Because of these weaknesses, positive meta-analyses should be considered largely hypothesis-generating, and should be confirmed by (a large) RCT. Negative meta-analyses can probably be accepted.
- Myles PS, Gin T. Statistical methods for anaesthesia and intensive care. 1st ed. Oxford: Butterworth-Heinemann, 2001.