Marjon Borgert

124 Chapter 6 nationwide improvement programs which positively in uences patient outcomes as well. The context in which a program for quality improvement is launched contributes to di erent outcomes. 83,85 Limitations Our systematic review is hampered by several limitations. There is a chance that we missed some relevant studies, because di erent terms are given to care bundles. However, a broad search strategy was used and we have completed the search with a hand search. Two criteria for selecting studies, i.e. compliance rates and implementation strategies, were not (clearly) reported in abstracts, while these criteria were described in the full text. We included any article to the phase of full-text screening if there was any uncertainty about one of the inclusion criteria. Furthermore, our review was restricted to the inclusion of English language publications only and relevant studies published in other languages could have been missed. However, evidence for the e ect of language restrictions on systematic bias remains inconclusive. Another important issue is that no studies with randomized designs were included. The majority of the studies included were quality improvements and before-and-after studies without controls. Thus, observed changes could be in uenced by secular trends. 86 Furthermore, the overall methodology of the included studies was poor, involving an increased risk of bias. 86 Therefore, the results should be interpreted with caution. An important problem hampering a meta-analysis was due to the heterogeneity of the available data (Supplementary File 5). There was a high variability in study design, methodological quality, bundle characteristics, compliance measurements and the calculation of compliance within a speci c time frame. Therefore, it was not possible to point out the superior implementation strategy. Moreover, complete data of compliance was lacking, e.g. most studies only reported compliance as a percentage, without explicitly reporting numerators and denominators. Although not all included studies show high compliance levels, publication bias could still have in uenced our results since all included studies show positive results. Since compliance was reported as secondary outcome, the quality of reporting could have been in uenced by this fact. Future research Further research is needed to identify the best strategy to implement care bundles to achieve high levels of compliance. To investigate the e ects of implementation strategies on compliance levels, there is a need for robust study methods in implementation or quality improvement research. Studies using randomized designs should be considered to increase the internal and external validity, especially when the intervention is considered for widespread implementation. 87 However, randomization