Development and Validation of a Logistic Regression Algorithm to Predict the Risk of Obstetric Anal Sphincter Injury.

Overview

An obstetric anal sphincter injury (OASI) occurs during the final stage of a vaginal delivery. This tissue laceration, even if adequately sutured, poses a substantial threat to bowel continence in women.1,2 In a recent register-based study we showed that following an OASI at the first birth, the risk of a repeat injury almost tripled and that the long-term prevalence of fecal incontinence (FI) doubled in women with 1 OASI and tripled in those with 2 consecutive OASIs, in comparison with nulliparous women not affected by childbirth.3 Most OASIs occur seemingly by chance in the absence of known risk markers, and there is still no prediction model that is of use to avoid OASI in the clinical setting.4 Therefore, these injuries are often excused as inevitable and impossible to foresee. The aim of this study is to develop and validate prediction models for the risk of an OASI in high- and low-risk scenarios.

Research plan for development and validation of a logistic regression derived algorithm to estimate the risk of obstetric anal sphincter injury Aim The aim of this study is to develop and validate prediction models for the risk of an OASI in high- and low-risk scenarios. Study design The study will be restricted to cover 10 years from 2009 to 2018. The primary data source will be the Swedish Medical Birth Register (MBR). In addition, data will be retrieved from the National Patient Register, which comprises data on health care episodes in inpatient (hospital) and outpatient specialist care maintained by the Swedish National Board of Health and Welfare and the Swedish Longitudinal Integrated Database for Health Insurance and Labor Market Studies maintained by Statistics Sweden. The following inclusion criteria will be applied: * The 1st and the 2nd vaginal delivery, with secured information on the number of births and the mode of delivery * Singleton pregnancies * Gestational week ≥37+0 Three separate study cohorts are planned to be analyzed: 1. The first vaginal birth in nulliparous women (high-risk) 2. The first vaginal birth after one or more prior cesarean sections (VBAC) (high-risk) 3. The second vaginal delivery (low-risk) Women from cohort 1 may contribute with their second birth to cohort 3. The main difference between the cohorts is the additional pre-natal information available in cohorts 2 and 3, compared with cohort 1. Definition of obstetric anal sphincter injury outcome The Swedish medical registers follow the International Classification of Diseases, 10th revision (ICD-10), for OASI. For identification of the outcome OASI, the following codes will be used. * ICD 070.2, O70.2C, O70.2D, O70.2E, O70.2F, O70.2X (third-degree tear, which involves part of or the entire anal sphincter), and/or * ICD O70.3 (fourth-degree tear, which extends further to the rectal mucosa) and/or * the variable "Sphincter, (SFINKTER)" is "1" * the surgical Diagnosis Related Groups code, MBC33 (suture of a third- or a fourth-degree perineal tear). Predictors of obstetric anal sphincter injury The selection of candidate predictor variables for OASI will be based on our previous works on OASI, a search of systematic reviews and meta-analyses in the literature, and clinically relevant and retrievable information in the registers. Candidate predictors: Maternal demographics Age Weight in early pregnancy Weight gain since prior pregnancy Height Body mass index Smoking habits during pregnancy Education Income Country of birth Prior obstetric information (maternal and infant) Emergency C-section Elective C-section Apgar score Large-for-gestational-age Infant birth weight Infant head circumference Obstetric anal sphincter injury Vacuum delivery Forceps delivery Labor induction Labor augmentation Episiotomy Weight gain since prior pregnancy Maternal diseases (current) Recurrent cystitis Chronic kidney diseases Diabetes type I and II, pre- pregnancy Epilepsy Asthma Inflammatory bowel disease Systemic lupus erythematosus Hypertensive diseases Pregnancy diabetes Labor Labor induction\* Labor augmentation Episiotomy Vacuum delivery Forceps delivery Epidural anesthesia\* Spinal anesthesia\* Infant birth characteristics Birth weight\* Head circumference\* Gestational age\* Large-for-gestational-age\* Gender\* Presentation at delivery Time of day of birth Conditional pre-natal predictors, variables in Cohort 1, which only can (or could) be determined or planned pre-birth, include maternal demographics and diseases, gestational age, fetal position, male gender, induction of labor, and, in some cases, the presence of macrosomia. \*At present, the infant birth weight, head circumference, and gestational age at delivery are documented post-partum. However, the ultrasonographic biometry technique for estimating fetal weight and head circumference is developing, so we will include these variables in all models. Statistical analysis plan The split-sample validation approach will be used with a temporal split according to the period for birth \[development/training data set 2012-2018 (\~70%) vs. validation/test data set 2009-2011 (\~30%)\]. Logistic regression statistics will be used as the main predictor modeling tool. A model will be developed from the training dataset based on the minimization of the Bayesian Information Criteria (BIC) using the "best subset selection" approach. Model performance and stability will be evaluated using a bootstrap approach with 200 samples developing models in each sample and comparing with the model from the whole training dataset and the global model. Non-linear effects will be evaluated using natural cubic splines with 5 knots at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles. Linear and splines effects will be included simultaneously in the model selection procedure, enabling a data-driven selection between linear and non-linear trends. Transformations and cut-offs of predictor variables may be tested if needed. The model developed on the training data will then be compared with the validation data using calibration plots. Risk scoring systems will be generated from the final models to obtain individual risk scores and predicted probabilities. Women will be divided into three risk groups: low, medium, and high risk. Three different models per cohort will be constructed (resulting in 3x2 = 6 models). 1. Predictors known pre-pregnancy (Maternal variables) 2. Predictors known post-partum (Maternal variables + Pregnancy + Neonatal + Labour) 3. Predictors known post-partum (Maternal variables + Conditional pre-natal predictors)