Method 3: SMDs, weighting by number of health care professionals
Where the population of interest is the health care professional, it may be desirable to weight the results by the number of health care professionals included6,21to aid population inference. This method utilises commonly reported summary information without adjustment for clustering. In this method of meta-analysis the studies are weighted by the number of health care professionals as an alternative to the commonly used inverse variance method which uses weights based on standard errors. As pointed out in table 4 there are weaknesses to this approach which we discuss below.
Not every trial in the SOCIAL review reported the number of health care professionals – for example a cluster trial where an intervention was directed at all staff on a hospital ward. Where no information was given about the number of health care professionals, Ivers et al.6 used the number of practices/hospitals/communities instead, and we followed that method here. An alternative might be to estimate the number of health care professionals using data from similar studies – e.g. using mean number of GPs per surgery or mean number of nursing staff on a hospital ward. Note that this method needs to be combined with method 1 or method 2 above or an alternative way of summarising mixed outcome measures; here we combined it with method 1 to summarise standardised mean differences.