JMIR Publications

ABSTRACT

Background:

Globally, the trends of preterm birth rate have been increasing over time. Ultrasonography cervical length assessment is considered to be the most effective screening method, however, universal cervical length screening in the whole population remains controversial because of the cost budget.

Objective:

In this work, obstetric data are used to analyze and assess the risk of preterm birth. The purpose is to screen high-risk groups of preterm birth in the early and second trimester of pregnancy, and targeted cervical screening is more in line with health economics.

Methods:

This study attempts to use continuous electronic medical records(EMRs) data of pregnant women to construct a preterm birth predicting classifier based on long short-term memory (LSTM) networks. The clinical data were collected from 5187 Chinese pregnant women with natural vaginal delivery, including more than 25,000 obstetric EMRs during the early trimester to 28 weeks of gestation. The area under ROC curve, accuracy, sensitivity and specificity were used to assess the performance of prediction model.

Results:

Compared with traditional cross-sectional study, LSTM model in time series study has better overall prediction ability, which has a lower misdiagnosis rate with the same detection rate, and the accuracy was 0.739, sensitivity was 0.407, specificity was 0.982, and AUC was 0.651. Feature importance identification indicated that blood pressure, blood glucose, lipids, uric acid and other metabolic factors were the important factors related to preterm birth.

Conclusions:

The results of this study are helpful to the formulation of guidelines for the prevention and treatment of preterm birth, and it can assist the clinicians to make correct decisions during the obstetric examinations. For the preterm birth prediction scenario, time series model has certain advantages.