In this retrospective cohort study, we collected data of women giving birth at “Fondazione Policlinico Universitario A. Gemelli (FPG) IRCCS” in the period between January 2017 and December 2019. The STROBE guidelines for cohort studies have been followed. As the aim of the study was to compare the possible effect of free-birthing position on obstetrics outcome in nulliparous women with or without epidural, we only recruited women at their first pregnancy aged between 18 and 40 years old, giving birth at a gestational age > 37 weeks, singletons, cephalic presentation with an estimated fetal weight and a neonatal weight appropriate for gestational age (AGA). We excluded all patients undergoing a caesarean section, pregnancies complicated by fetal anomalies or those with preexisting chronic illness.
All the patients admitted in labor in the delivery room were asked to give written informed consent to use the data of the pregnancy and outcome for research; we stored all pregnancy and delivery characteristics in a registry and we subsequently performed a retrospective analysis of the data extracted anonymously.
All the patients had the opportunity to choose the position the felt more comfortable with during the second stage of labor, according to the CTG trace and the progress of labor. In particular we documented the position the patient choose for pushing. If more than one position has been assumed during the second stage, we considered the one when the child was born.
Women choosing to have analgesia received epidural analgesia using ropivacaine at different concentration basing on cervical dilatation (from 0.1% starting at 4 cm of dilatation until 0.2% with full dilatation) with sufentanil 10 µg·ml in the first administration.
We investigated the association between selected maternal and gestational characteristics and several categorical and continuous outcomes. The primary outcome was the onset of perineal trauma during delivery, defined by three stages of increasing perineal damage according to the classification of Sultan et al. (a) first-degree vaginal tear; (b) second-degree vaginal tear; (c) episiotomy - versus women with intact perineum. Secondary outcomes were: (1) assisted vaginal delivery - as opposed to spontaneous vaginal delivery; (2) estimated blood loss during delivery; (3) duration of fetal descent during labor; (4) 1-minute and (5) 5-minutes Apgar scores.
The potential associations between the recorded maternal and gestational parameters and the two categorical outcomes − (1) all types of perineal trauma onset versus no trauma; (2) type of vaginal delivery - were first evaluated with standard univariate analyses: chi-squared test for categorical variables; t-test and Kruskal-Wallis test for normally distributed and non-normally distributed continuous variables, respectively (distribution assessed with Shapiro-Wilk test). The potential independent predictors of each categorical outcome were then evaluated using multivariate logistic regression. In all models, covariates were included in a stepwise forward process using the following criteria: clinical relevance; p < 0.2 at univariate analysis; age, gestational age, body mass index, maternal position at delivery and type of anesthesia forced to entry. Anesthesia and - when included as covariate in the multivariate models - perineal trauma were treated either as dichotomous variables (epidural versus no/local anesthesia; episiotomy versus no episiotomy, respectively) and ordinally, including the different types of anesthesia (none; local; epidural) and the above mentioned four levels of perineal trauma as dummy variables. The goodness-of-fit was checked using Hosmer-Lemeshow test, and the predictive power assessed through C-statistics (area under the Receiving Operator Curve). Standard post-estimation tests were used to check the final model validity, performing multicollinearity and influential observation analyses (using standardized residuals, change in Pearson and deviance chi-square), and testing for potential statistical interactions between outcomes and included covariates. We found very few variables which were collinear (1-minute and 5-minutes Apgar scores), and chose to include only the first in the final models. Also, less than 10% of influential observations were found for each model, and when analyses were repeated excluding the outliers, no relevant changes were observed, and no observation was thus excluded. Missing values were less than 5% for all variables, therefore no missing imputation technique was adopted.
To further explore the relationship between the recorded maternal and gestational characteristics and the onset of perineal trauma at delivery, six additional univariate analyses were run, using the same approach previously described. In each analysis, the recorded variables were compared between: (1) women with intact perineum and women with first-degree vaginal tear; (2) women with intact perineum and women with second-degree vaginal tear; (3) women with intact perineum and women undergoing episiotomy; (4) women with first-degree vaginal tear and women with second-degree vaginal tear; (5) women with first-degree vaginal tear and women undergoing episiotomy; (6) women with second-degree vaginal tear and women undergoing episiotomy. As a separate, additional analysis, a polytomous logistic model was used to test the independent association between all recorded covariates and perineal trauma. Three odds ratios (ORs) were therefore obtained for each predictor variable: given intact perineum as the reference category, the first OR was referred to women with first-degree vaginal tear; the second to women with second-degree tear; the third to women with episiotomy.
Finally, we evaluated the potential association between all recorded maternal and gestational characteristics and the four continuous outcomes. A Spearman correlation coefficient between the outcomes and each continuous parameter was first computed, and four multiple regression models were then fit. Separate analyses were initially made for 1-minute and 5-minutes Apgar scores. However, the two outcomes showed collinearity (Spearman rho = 0.70), and the regression coefficients, both raw and adjusted, were almost coincident. Thus, only the analyses related to the 1-minute score were reported to avoid redundancy. The same above specified criteria were used for model building, and the validity of final regression models was assessed as follows. The assumption of constant error variance was checked graphically, plotting Pearson residuals vs. fitted values, and formally, using the Cook-Weisberg test for heteroskedasticity. High leverage observations were identified by computing Pearson, standardized and studentized residuals, and Cook's D influence. In all models, we found less than 10 high-leverage observations, excluding which we noted no substantial changes. Statistical significance was defined as a two-sided p-value < 0.05, and all analyses were carried out using Stata, version 13.1 (Stata Corp., College Station, Texas, USA, 2013).