155 6 Cochrane review on integrated disease management for COPD recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Chapter 23.1.3) (Higgins 2019b). In case of a unit of analysis error in cluster-RCTs, we adjusted for the design e ect by reducing the size of the trial to its “e ective sample size” (Rao 1992). The e ective sample size of a single intervention group in a cluster-randomised trial is its original sample size divided by a quantity called the ‘design e ect’. The design e ect is 1 + (M - 1) * ICC, where M is the average cluster size, and ICC is the intra-cluster correlation coe cient. For dichotomous data, both the total number of participants and the number of participants experiencing the event were divided by the design e ect. For continuous data, for which the GIV method could not be used, only sample sizes were reduced, and means and SDs were left unchanged (Higgins 2011). Dealing with missing data When a study paper missed important statistical information required for analysis, or required additional calculations that needed to be clari ed, we attempted to contact study authors to gather the required information. When authors had not calculated relevant statistics but presented supporting data, we conducted calculations using methods described in the 2019 Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2019a). When studies did not report SDs for change from baseline but did provide information on means, standard errors (SEs), 95% CIs, P values, and population sizes across groups, we calculated SDs for change from baseline using the RevMan 5 internal calculator. When we could not directly calculate the SD for change from baseline, we imputed the SD using a correlation coe cient as described in the Cochrane Handbook for Systematic Reviews of Interventions (Chapter 6.5.2.8) (Higgins 2019a). We calculated the correlation coe cient by using the weighted mean (based on size of the study) of two or more studies that reported results for the respective variable in su cient detail. In the case that fewer than two studies provided su cient information, a weighted mean correlation coe cient could not be calculated. In that case, we used data on post-intervention measurements, as they are considered to be more precise. For studies that reported a median instead of a mean, we estimated the mean and the SD using the method and open-access calculator provided in Wan 2014 . Assessment of heterogeneity We assessed heterogeneity in each meta-analysis both visually through inspection of forest plots and statistically using tau², I², and the T statistic (Higgins 2019). We regarded heterogeneity as substantial when I² was greater than 50% or a low P value (< 0.10) was reported for the Chi² test for heterogeneity. We reported heterogeneity and explored the possible causes. In cases of substantial (I² > 50%) or considerable (I² > 75%) heterogeneity, we investigated sources for heterogeneity by conducting subgroup analyses (see Subgroup analysis and investigation of heterogeneity).
RkJQdWJsaXNoZXIy MTk4NDMw