165 6 Cochrane review on integrated disease management for COPD exception of one outcome (hospital admission (in days)), funnel plots did not indicate that publication bias is likely. Observed asymmetry of the funnel plot for hospital admission is probably caused by the poor methodological quality of Farrero 2001 . Other potential sources of bias We included nine cluster-randomised trials, three of which introduced bias (Lou 2015; Rea 2004; Wood-Baker 2006). In Wood-Baker 2006, there was noticeable imbalance in di erences between groups at baseline. Wood-Baker 2006 and Lou 2015 did not account for clustering in statistical analyses of dichotomous outcomes. This may lead to over-precise results and can result in much more weight in a meta-analysis (Higgins 2011). Therefore, in our meta-analyses, we adjusted for the design e ect by reducing the size of the trial to its “e ective sample size” for all dichotomous outcomes (Rao 1992), and we used the adjusted MD via the GIV approach for all continuous outcomes. In Rea 2004, there was loss to follow-up of ve clusters (four control and one intervention cluster). Other potential sources of bias were found in Titova 2017 Kessler 2018 Lenferink 2019 Vasilopoulou 2017 Vianello 2016, and Lou 2015. Lou 2015 was performed across four geographically distinct regions and based randomisation on geographical location, thereby potentially introducing cluster e ects.
RkJQdWJsaXNoZXIy MTk4NDMw