16.3.6  Approximate analyses of cluster-randomized trials for a meta-analysis: inflating standard errors

A clear disadvantage of the method described in Section 16.3.4 is the need to round the effective sample sizes to whole numbers. A slightly more flexible approach, which is equivalent to calculating effective sample sizes, is to multiply the standard error of the effect estimate (from an analysis ignoring clustering) by the square root of the design effect. The standard error may be calculated from a confidence interval (see Chapter 7, Section 7.7.7). Standard analyses of dichotomous or continuous outcomes may be used to obtain these confidence intervals using RevMan. The meta-analysis using the inflated variances may be performed using RevMan and the generic inverse-variance method.

 

As an example, the odds ratio (OR) from a study with the results

Intervention: 63/295

Control: 84/330

is OR = 0.795 (95% CI 0.548 to 1.154). Using methods described in Chapter 7 (Section 7.7.7.3), we can determine from these results that the log odds ratio is lnOR = –0.23 with standard error 0.19. Using the same design effect of 1.576 as in Section 16.3.5, an inflated standard error that accounts for clustering is given by 0.19 × Ö1.576 = 0.24. The log odds ratio (–0.23) and this inflated standard error (0.24) may be entered into RevMan under a generic inverse-variance outcome.