Accurate and actionable prediction of impending labor using deep learning on maternal physiological data

Every pregnancy is assigned a “due date.” However, this date is not an accurate or personalized
guide for when labor will begin, or when the baby will be born. The “estimated due date” (EDD)
represents forty completed weeks of pregnancy, calculated from the first day of the last menstrual
period. Instead of being useful for predicting or planning, 40 weeks is an average duration of
pregnancy across populations. Mothers and infants with a duration of pregnancy under 37 weeks or
over 42 weeks are both at risk for birth complications, morbidity, or mortality. However, even
across ‘normal’ term gestation, uncertainty in planning for birth can arise from unexpected
complications, cause added anxiety, and lead to greater use of costly intervention or
hospitalization. For rural residents or for those with high-risk pregnancies who should not undergo
labor, the risk of uncertainty can be overtly dangerous. Our team has developed a method to
interpret physiological vital sign patterns during pregnancy to create an accurate prediction of
when labor will start. The proposed 8-month study will enhance and improve our existing work,
using artificial intelligence methods on data from non-invasive wearable sensors, making the
prediction of labor more accurate. We will also operationalize a method to provide families or care
providers with a time frame when labor is likely to occur in real-time. This tool will then be applied
to a large validation trial of the method in pursuit of FDA-approval.

Grant Number: 1R41HD117576-01A1
NIH Institute/Center: NIH
Principal Investigator: Chinmai Basavaraj