Continuous data are described in Chapter 9, Section 9.2.3, and their meta-analysis is discussed in Chapter 9, Section 9.4.5. To perform a meta-analysis of continuous data using either mean differences or standardized mean differences review authors should seek:

 

Due to poor and variable reporting it may be difficult or impossible to obtain the necessary information from the data summaries presented. Studies vary in the statistics they use to summarize the average (sometimes using medians rather than means) and variation (sometimes using standard errors, confidence intervals, interquartile ranges and ranges rather than standard deviations). They also vary in the scale chosen to analyse the data (e.g. post-intervention measurements versus change from baseline; raw scale versus logarithmic scale).

 

A particularly misleading error is to misinterpret a standard error as a standard deviation. Unfortunately it is not always clear what is being reported and some intelligent reasoning, and comparison with other studies, may be required. Standard deviations and standard errors are occasionally confused in the reports of studies, and the terminology is used inconsistently.

 

When needed, missing information and clarification about the statistics presented should always be sought from the authors. However, for several of the measures of variation there is an approximate or direct algebraic relationship with the standard deviation, so it may be possible to obtain the required statistic even if it is not published in the paper, as explained in Sections 7.7.3.2 to 7.7.3.7. More details and examples are available elsewhere (Deeks 1997a, Deeks 1997b). Chapter 16 (Section 16.1.3) discusses options if standard deviations remain missing after attempts to obtain them.

 

Sometimes the numbers of participants, means and standard deviations are not available, but an effect estimate such as a mean difference or standardized mean difference may be reported, for example in a conference abstract. Such data may be included in meta-analyses using the generic inverse variance method only if they are accompanied by measures of uncertainty such as a standard error, 95% confidence interval or an exact P value. A suitable standard error from a confidence interval for a mean difference should be obtained using the early steps of the process described in Section 7.7.3.3. For standardized mean differences, see Section 7.7.7.