Development and validation of a multivariable prediction model of spontaneous preterm delivery and microbial invasion of the amniotic cavity in women with preterm labor

Abstract

BACKGROUND: Early spontaneous preterm delivery is often associ- ated with microbial invasion of the amniotic cavity and/or intraamniotic inflammation. OBJECTIVE: The objective of the study was to develop and validate clinically feasible multivariable prediction models of spontaneous delivery within 7 days and microbial invasion of the amniotic cavity in women admitted with diagnose of preterm labor and intact membranes below 34 weeks. STUDY DESIGN: We used data from a cohort of women admitted from 2012 to 2018 with diagnosis of preterm labor below 34 weeks who had undergone amniocentesis to rule out microbial invasion of the amniotic cavity. The main outcome was spontaneous delivery within 7 days from admission. The secondary outcome was microbial invasion of the amniotic cavity, defined by a positive culture and/or 16S ribosomal RNA gene in the amniotic fluid. The sample (n 1⁄4 358) was divided into derivation (2012e2016) and validation cohorts (2017e2018). Logistic regression models using a stepwise selection of variables were developed for the outcomes evaluated. We explored as predictive variables ultrasound cervical length measurement at admission, maternal C-reactive protein, gestational age, amniotic fluid glucose, and interleukin-6 (expressed as log units). Models were developed in the derivation cohort and applied to the validation cohort and diagnostic performance was calculated. RESULTS: The derivation cohort included 263 women and the valida- tion cohort 95 women. One hundred five of the women (39%, 105 of 268) spontaneously delivered in the following 7 days and 68 (19%, 68 of 358) had microbial invasion of the amniotic cavity. For spontaneous delivery within 7 days after admission, 4 predictors were identified: cervical length at admission, gestational age, amniotic fluid glucose, and interleukin-6. The diagnostic performance of the model was assessed in the validation cohort using the receiver operating characteristic curve and showed an area under curve of 0.86 (95% confidence interval, 0.77e0.95) with a detection rate of spontaneous delivery within 7 days of 87%, a false- positive rate of 33%, a negative predictive value of 80%, and a negative likelihood ratio of 0.1908. For microbial invasion of the amniotic cavity, 2 independent predictors of the amniotic cavity were identified: amniotic fluid glucose and maternal C-reactive protein. The receiver operating characteristic curve and an area under curve in the validation cohort was 0.83 (95% confidence interval, 0.70e0.96) with a detection rate of 76%, a false-positive rate of 8%, a negative predictive value of 93%, and a negative likelihood ratio of 0.2591. CONCLUSION: In women with preterm labor, we propose 2 clinically feasible prediction models to classify as low vs high risk of spontaneous delivery within 7 days and of microbial invasion of the amniotic cavity. The models showed a high diagnostic performance and could be of value to optimize clinical management.