Data extraction and outcome indicators
Two authors (Guo ZH and Qiu YJ) independently screened the titles and abstracts of the studies identified based on inclusion and exclusion criteria. Full texts were then read to determine which studies would be finally included in this meta-analysis. The Cochrane Risk of Bias and the tool Newcastle–Ottawa Scale were used to evaluate bias and quality. Statistical heterogeneity in the included studies was assessed via I2 test and Cochran’s Q test, and funnel plots and Egger’s test were applied to assess publication bias10. The I2 > 50% and Cochran’s Q test p < 0.1 suggest statistically significant heterogeneity11.
Data were extracted from all included studies independently by two authors, presented by a standard form that included author, study type, control group, sample size, mean gestational age, mean birth weight, and corrected gestational age at extubation. Any disagreements in data extraction were settled via discussion. The indicator of the primary outcome was extubation failure, which was often defined as requiring reintubation and invasive ventilatory support within 72 h. The secondary outcomes were rates of related adverse clinical events under NIV-NAVA compared with conventional NIV in extremely preterm infants after extubation.
Data synthesis and statistical analysis
The expression of dichotomous outcome data analyses are risk ratios (RRs) with 95% confidence intervals (CIs)12, and the expression of continuous outcome data analyses are mean differences (MDs) with 95% CIs. Data reported as medians and interquartile ranges were converted to estimated means and standard deviations via a standard method13,14. Review Manager (RevMan 5.3; Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Denmark) software was applied to process meta‑analyses. Variables were analyzed with a fixed-effects model or a random-effects model depending on statistical heterogeneity. When p> 0.10 or I2 < 50% a fixed-effects model was used, otherwise a random-effects model was used.
Trial sequential analysis
The meta-analysis is often called interim analysis when applied to new data that are constantly being updated, and some false-positive conclusions or false-negative conclusions are unavoidable due to sparse or accumulating data15. The application of trial sequential analysis (TSA) may ameliorate these problems in analysis with adjustment of CIs and restricted thresholds16. In the current study TSA software (version 0.9.5.10 Beta; Copenhagen Trial Unit, Copenhagen, Denmark) was used to construct adjusted significance test boundaries by alpha-spending17. The type 1 error was set as 5% and boundary type was set as two-sided for conventional boundaries. An O’Brien Fleming power of 80% was used in alpha-spending boundaries. Relative risk reduction of 71% was estimated, 20% for the incidence in the control arm, and the heterogeneity correction was based on model variance17,18.