Data synthesis and analysis
A meta-analysis was performed using the random-effects model with the Mantel-Haenszel method using Review Manager (RevMan), version 5.4 (The Nordic Cochran Center, The Cochrane Collaboration). The effect estimates were reported as RRs and absolute risk differences (ARDs) with 95% confidence intervals (95% CIs). Statistical significance was set at p < 0.05. Heterogeneity was assessed by visual inspection of the forest plot using the χ2-test (significant if p<0.10) and I2 statistic (heterogeneity considered significant if I2 > 40%). Subgroup analyses by gestational age (<28 gestational weeks, 28-32 gestational weeks), respiratory support (invasive ventilation or non-invasive ventilation), and age after birth (acute phase; 0-27 days, chronic phase ≥ 28 days) were preplanned and performed.