Data analysis
Categorical variables were expressed as absolute numbers and compared among groups with χ2 or Fisher exact test as appropriate.
We evaluated the association between each group and negative outcome(s), CS, and OVD using multiple logistic regression models adjusting for baseline characteristics (e.g., age, parity, education, BMI, neonatal weight). Results of logistic regression are also presented for CS and OVD since they were evaluated as clinical outcomes related to failed induction in Sri Lanka.26 A one-sided Cochran-Armitage test for trend was performed to assess the influence of changes of clinical protocols and staff training practices27, 28over different semesters of the study on CS and OVD.
As secondary analyses we compared IOL at 40 GW to a group composed of IOL at 41 GW and SOL, in line with analyses by Rydahl and collegues.16 This allowed comparison between IOL group at 40 GW and spontaneous labour at the same gestational age, and simultaneously took into account the risks of the ongoing pregnancy including all births at 41 GW, reducing possible bias.
In addition, we performed a sensitivity analysis including all cases with reported hypertensive disorders (pregestational hypertension, preeclampsia, eclampsia, HELLP syndrome), chorioamnionitis, oligohydramnios, APH, and signs of potentially impaired foetal wellbeing (non-reassuring or pathological cardiotocography, reduced foetal movement, meconiumstained amniotic fluid) from 41 GW, considering these as negative birth outcomes rather than as possible risk factors.
Data were analysed using STATA version 14.0 (Stata Corporation, College Station TX) and SAS/STAT® software version 9. All statistical tests were two-sided and a p-value less than 0.05 was considered statistically significant.