OBJECTIVE

To compare the prognostic models for spontaneous preterm birth.

MATERIAL AND METHODS

A retrospective cohort study was based on the Arkhangelsk County Birth Registry data. We analyzed 50581 spontaneous births in 2012—2018. We developed 6 models (three each for logistic regression and classification trees): for prediction of preterm birth at pregnancy planning stage, at the first antenatal visit and at the beginning of the second trimester of pregnancy (20—22 weeks of gestation).

RESULTS

In 2012—2018, there were 2167 spontaneous preterm births (4.28%). The outcomes of previous pregnancies, some preexisting diseases (diabetes mellitus, epilepsy, obesity), behavioral factors and characteristics of the current pregnancy (multiple pregnancy, isthmic-cervical insufficiency, threat of abortion) were significant for prediction. Depending on predictors, AUC for regression models and models based on classification trees varied from 0.623 (95%CI: 0.610—0.636) to 0.703 (95%CI: 0.691—0.715) and from 0.561 (95%CI: 0.552—0.571) to 0.620 (95%CI: 0.608—0.632), respectively.

CONCLUSION

Binary logistic regression had higher predictive performance compared to classification trees for prediction of spontaneous preterm delivery. Maximum prognostic significance was achieved after inclusion of factors of data on anamnesis, lifestyle and health of the pregnant woman, as well as peculiarities of the current pregnancy.