Method 1
SMDs. Weights based on adjusted standard errors
|
All available data combined
Clustering accounted for at level of randomisation
|
Mixture of different outcomes and formats likely to lead to
heterogeneity
May be difficult to interpret
Inconsistent units of analysis (patient/HCP/site)
Estimation assumptions may not hold
|
Method 2
Separate analyses for binary and continuous data. Weights based
on adjusted standard errors
|
Likely to lead to less heterogeneity than method 1 as more similar
measures are being combined.
Little manipulation or estimation required
|
Does not combine all available information in a single analysis, which
leads to loss of power and multiplicity
Two analyses may give conflicting results
|
Method 3
SMDs. Weighting by number of HCP
|
Consistent units – weighted by health care professional
|
Number of health care professionals not always reported, requiring an
estimate to be imputed
Weighting may be related to quality of reporting; e.g. poorly reported
studies get less weight.
Unit of analysis error when not randomised at level of analysis
Weights related to the size of the study but not the
variability/precision
Issues with SMDs as above
|
Method 4
Albatross plot
|
May include additional studies that report p-value only
No assumptions
|
Difficult to check that p-values are correct if not accompanied by other
summary data
P-values prone to selective reporting
Need to adjust sample size in some way for cluster trials
|