Statistical analysis
For each variable (demographic, in-labour and delivery) differences across delivery groups (1-5) were compared using either ANOVA or Kruskal-Wallace ANOVA for normal or non-normally distributed data respectively. 95% confidence intervals and odds ratios were calculated using univariate logistic regression analysis to examine 1) the association between instrumental delivery mode (groups 1-4) and spontaneous vaginal delivery, and 2) the relationship between all delivery modes (groups 1-5) and overall composite risk outcome.
To adjust for baseline risk, a stepwise, multivariate regression analysis was performed. Indices were included based on clinical plausibility and/ or a significant association with the following dependent variables in univariate analysis: successful vaginal delivery, maternal composite risk outcome, neonatal composite risk outcome and overall composite risk outcome. These were inputted alongside delivery mode to produce two final multivariate binomial regression models: one with a dependent variable of successful vaginal delivery (delivery groups 1-4) and another with a dependent variable of overall composite risk outcome (delivery groups 1-5). Receiver operating characteristic curve analysis was performed on these final models with an area under the curve (AUC) of 0.80 considered to represent reasonable prediction. All data was collected into Microsoft Excel and analysis performed using IBM SPSS statistical software Version 26.

Results

Figure 1 displays the breakdown of delivery types occurring at the hospital over the study period (2014-2018). Of 24,756 deliveries 66.5% were spontaneous vaginal deliveries, 22.48% were Caesarean deliveries and 10.6% were instrumental deliveries. From the 2631 instrumental deliveries performed over the study period, 991 (37.6%) were performed in obstetric theatre as a trial of instrumental delivery. Excluding unavailable or twin pregnancy data (20 datasets), remaining cases were organised into 5 groups according to the initial delivery attempt from which 285 (29.3%) were KFD, 300 (30.8%) were DFD, 163 (16.7%) were MR+FD, 116 (11.9%) were VD, and 107 (11%) were pEmCS.
Across groups 1 to 5, the data collected for demographic variables is displayed in Table 1. Differences between groups were observed for both body mass index (BMI) and weight, with the pEmCS group having a significantly higher mean average weight. As height did not vary significantly between groups, this led to a weight-dependent difference in BMI between groups. This factor was accounted for with the inclusion of BMI as a final variable in the multivariate models described below.
The data representing all in-labour and delivery variables can be observed in Supplementary Table 1. There were a greater proportion of multiparous women in both the KFD and DFD groups compared with pEmCS; the deliverer was more often a doctor of greater experience undertaking the Keilland’s deliveries compared with other groups and pEmCS were significantly more likely than other groups to be performed for fetal distress and to be performed under general anaesthetic. The direct forceps delivery group had a greater proportion of occipito-anterior position and low cavity (fetal station 2+ and below) deliveries than other groups. Use of 2 instruments was greater in the ventouse delivery group compared with other instrumental delivery groups.
The primary outcome of vaginal delivery was investigated initially via univariate regression analysis, with the instrumental delivery groups (1-4) compared, using group 1 (KFD) as the reference (Figure 2A). Attempt at manual rotation and direct forceps was associated with the greatest percentage of completed vaginal delivery (92%), with direct forceps (89.7%) and Keilland’s (83.2%) second and third respectively. Attempt at ventouse delivery displayed the lowest percentage completion rate for vaginal delivery at 75%, consistent with ventouse delivery completion rates seen in existing literature 15. When compared with the reference group of Keilland’s forceps, raw odds ratios suggested that use of direct forceps delivery (OR 1.75, 95%CI: 1.08 – 2.05) and manual rotation plus forceps delivery (OR 2.33, 95%CI: 1.22-4.45) might increase chances of vaginal delivery; with ventouse delivery less likely to achieve vaginal delivery (OR 0.60, 95%CI: 0.36-1.02). A stepwise, multivariate regression analysis was performed to further investigate this trend and adjust for baseline confounding and risk factors. The final model adjusted for maternal BMI, birth weight, parity, analgesia, experience of deliverer, fetal position and fetal station. Ventouse delivery was significantly less likely to succeed at vaginal delivery when compared to KFD in this adjusted model (OR 0.426, 95%CI 0.227-0.797). To assess the robustness of predictive value for this model a receiver operator characteristic curve was produced (Supplementary Figure 1A) which demonstrated an area under the curve of 0.768.
The data demonstrating immediate perinatal adverse outcomes are presented in Table 2. There were two neonatal deaths in the cohort, one in the Keilland’s group and one in the ventouse group, there were no significant neonatal injuries noted in any groups. Of note, instrumental deliveries other than Keilland’s forceps were associated with higher rates of 3rd and 4th degree tears, with the highest proportion in the ventouse group at 10.3%. Primary emergency full dilatation Caesarean was associated with the highest proportion of maternal injuries during Caesarean (8.4%) and a greater frequency of babies with Apgar score of <7 at 5 minutes (9.3%), the latter possibly associated with a much higher proportion of these deliveries being performed under general anaesthetic (22.4%) than was observed in the other delivery groups (Supplementary Table 1).
The secondary outcome of overall composite risk score was investigated via univariate regression analysis, with the all delivery groups (1-5) compared, using group 5 (pEmCS) as the reference group, the results of which are displayed in Figure 2B. An outcome associated with immediate risk occurred in 22.8% of all deliveries in the cohort, indicating the high-risk nature of such full dilatation delivery trials. Keilland’s forceps deliveries were associated with the lowest proportion of composite risk outcome occurrence (17.5%), and ventouse deliveries the highest (29.3%). When compared with the reference group of pEmCS, there were no significant associations between mode of delivery and composite risk outcome. This analysis was further modified using multivariate logistic regression to adjust for baseline risk factors. The final variables included in this model were: maternal BMI, birth weight, analgesia, experience of trial decision maker, indication for trial, experience of deliverer, fetal position and fetal station. The adjusted model demonstrated significant differences (P=0.016) for composite risk outcome occurrence between the pEmCS and KFD delivery groups (OR 0.37, 95% CI: 0.16-0.81), however the receiver operating characteristic curve produced from this model demonstrated low predictive value with an area under the curve of 0.64 (Supplementary Figure 1B).

Discussion