Rethinking the efficacy of awake prone positioning in COVID-19-related acute hypoxaemic respiratory failure.

Jie Li and colleagues examined the efficacy of awake prone positioning on patients with COVID-19-related acute hypoxaemic respiratory failure. However, we have serious concerns about the assessment of the risk of bias, the data analysis, and the rating of certainty of evidence in the two cluster randomised controlled trials (RCTs) and eight individual RCTs that were included in the study.

First, the tool the authors used to assess the risk of bias in cluster RCTs is not optimal. In contrast to individual RCTs, in which individual participants are randomly assigned to a group, cluster RCTs are designed to evaluate interventions delivered at the group level. This design increases the possibility of some types of bias, such as identification or recruitment bias and loss of clusters. Therefore, the Cochrane collaboration recommends a specific tool, the risk-of-bias 2.0 for cluster-randomised trials tool for assessing the risk of bias in cluster RCTs. Li and colleagues, however, used the conventional tool intended for individual RCTs. We reappraised the cluster RCTs that were included using the recommended tool, and found that more sources of bias were classified as high compared with the findings of Li and colleagues, especially sources related to the cluster design (appendix). Therefore, the overall risk of bias in these cluster RCTs should be graded as high.

Second, Li and colleagues did not take into account the design effect of clustering. After adjusting for design effect, we found that the results on the need for escalating respiratory support (relative risk 1·03, 95% CI 0·78 to 1·36), admission to an intensive care unit (relative risk 0·80, 95% CI 0·52 to 1·22), and hospital length of stay (mean difference +0·55 days, 95% CI –0·52 to +1·62) differed from those calculated by Li and colleagues (appendix).

Finally, we found that Li and colleagues did not appropriately consider the risk of bias when evaluating the certainty of the evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. Absent or unclear blinding has been shown to have a substantial effect on the results of RCTs, exaggerating the intervention effects by up to 13%. According to the Cochrane criteria, the overall risk of bias in all the RCTs included in the study should be categorised as high because of the absence of blinding. Therefore, the certainty of the evidence on all outcomes should be downgraded by one GRADE level from those reported by Li and colleagues because of the serious risk of bias (appendix). The low certainty of evidence indicates that new research is likely to change the estimate. Therefore, caution is needed when interpreting the efficacy of awake prone positioning and when using awake prone positioning in clinical practice.

We declare no competing interests.