New screening method for prediction of preterm delivery in singleton pregnancies

Headnote

Keywords: Preterm, screening, delivery, risk factors

[OP-102]

Objective: To establish a new predictive model for screening the risk of preterm delivery (PTD) before 37 weeks of gestation

Material and Methods: This is a cohort study. Between 2013-2015 all singleton pregnancies (n: 1453) with (38%, 557/1453) or without (62%, 896/1453) PTD risk factors were recruited in the study. Obstetric history, cervical length and nomogram for PTD risk factors were evaluated to establish a new screening risk assessment for PTD. All clinical variables were assessed for an association with 37 week PTD using univariable logistic regression models. Multivariable logistic regression analysis was used to identify a set of predictors. Associations were summarized using the odds ratio (OR) and corresponding 95% confidence interval (CI) estimated from the models. Type-I error rate was taken as 0.25 for candidate variables for multivariate logistic regression. Discrimination was assessed with 300 bootstrap resamples. For each bootstrap sample, a logistic regression model was fit using the variables identified in the final model and the concordance index (c-index) was calculated. The c-index is a measure of a model's predictive accuracy (discrimination). An unbiased estimate of the c-index was obtained based on averaging the 300 c-indices. Calibration was assessed graphically by examining how far the predicted probabilities are from the actual observed proportion with 37 week PTD. Statistical analyses were performed using R software. A nomogram constructed from the final model is presented in Fig. 1. For a given patient, points are assigned to each of the variables and a total score is derived. The total points score corresponds to a predicted probability of 37 week PTD. The performance of the final model was assessed through calibration and discrimination. The models had good calibration as illustrated in the calibration plot in Fig. 2. The unbiased estimate of the c-index derived from bootstrap resamples was excellent (c-index=0.746).

Results: PTD rates were 22.1%. Of those 15,8% pregnancies had at least one risk factor for PTD. Univariable analysis revealed that the most common risk factors for PTD were maternal age (74.5%), in-vitro- fertilization (29.4%), placenta previa (3.9%) respectively. The relationship between PTD and risk factors only gestational diabetes mellitus (GDM) (p:0.057, CI 0.9 - 14.1), PPROM (preterm premature rupture of membranes) (p:0.000, CI 4.2-17.5), eraly onset of uterine contractions (p:0.000, CI 2.5-9.7), second/third trimester bleeding (p:0.000, CI 4.2-17.5), 3.854-50.672), preeclampsia (p:0.000, CI 4.7-19.4), advance maternal age (p:0.000, CI 0.2-0.5), oligo/anhidramnios (p:0.000, CI 1.5- 5.5), maternal hypothyroidism (p:0.000, CI 1.124-3.386) were found statistically significant (p<0.05).

Conclusion: Recently the screening method of PTD is depends on patient obstetric history and cervical length measurement by transvaginal ultrasonography. However clinical symptoms of the pregnancy such as; polihidramnios, GDM etc. are not objectively assessed. We conducted a unique statistical model for PTD. This a new model consists of not only obstetric history and short cervical length but also of the clinical symptoms to evaluate the risk of PTD more sensitive, specific and accurate. Further studies are needed to answer which preventative measures such as; cerclage, cervical pessary, progesterone on the specific conditions and the symptoms is better.