Diastolic Function and Positive Airway Pressure: More Research Is Warranted

Dear Editor,We read with interest the meta-analysis on the value of chronic treatment with noninvasive positive airway pressure (PAP) on left ventricular diastolic function (LVDF) in patients with obstructive sleep apnea [1]. The authors included 9 randomized studies (833 patients) and concluded that noninvasive PAP is associated with improvement in LVDF, as suggested by a decrease in deceleration time (DT), isovolumic relaxation time (IVRT), and the ratio of E/e'. While these results are physiologically sounded [2-4], we have some concerns on the methodology of this meta-analysis and on the authors' conclusions.The first consideration that warrant caution when interpreting the results of the meta-analysis [5] is the authors' choice to conduct meta-analysis with a fixed-effect rather than a random effect (RE) model. The fixed-effect approach assumes that the "true effect" is the same across all studies. However, it is unlikely that all included studies have similar "true effect" due to the heterogeneity of the included populations. More importantly, the fixed-effect model should not be used when there is statistical heterogeneity as in the case of DT (I2 = 94%) or E/e' ratio (I2 = 69%). In such cases, it is strongly advisable to use a RE model, which better balances the weights of the included studies [6].The second consideration is that, as the number of studies reporting data on the three echocardiography parameters is limited, conducting a trial sequential analysis (TSA) on the primary outcomes before drawing strong conclusions is strongly advisable. Indeed, performing TSAs would allow to decrease the risks of type I statistical error and to understand the robustness of the authors' findings. Therefore, the study by Al-Sadawi et al. [1] would strongly benefit from a TSA gathering results on the required "information size" (sample of patients needed) for the investigated outcomes. We imported outcomes data in the TSA Software (Copenhagen Trial Unit's TSA Software®; Copenhagen, Denmark). The information size was computed assuming an alpha risk of 5% with a power of 80%. We used a RE model, and as estimated effects on the investigated outcomes we used the mean difference reported by Al-Sadawi et al. [1]. Further details on TSA and its interpretation are available elsewhere [7]. Therefore, we conducted three TSAs in total to investigate the robustness of PAP-induced improvements in DT, IVRT, and E/e' ratio. Our three TSAs showed that the required information size (or sample size needed) was never reached for the outcomes that suggest some improvement in LVDF: DT, n = 223/810 (Fig. 1); IVRT, n = 223/317 (Fig. 2); E/e' ratio, 118/351 (Fig. 3).In summary, the authors conducted a meta-analysis on an interesting topic, but it is important to apply the RE model when heterogeneity is present and to perform TSAs to confirm the robustness of the investigated outcomes. More research is definitely needed on this topic.The authors have no conflicts of interest to declare.No funding was received.All the authors read the meta-analysis and commented it. F.S., L.L.V., and M.Z. performed a check on the values of echocardiography variables; F.S. and L.L.V. conducted the TSA; C.C. and M.A. assessed the appropriateness of random- and fixed-effect model; FS wrote the draft of the letter; L.L.V., M.Z., C.C., and M.A. critically revised the letter; all the authors approved the final version.