Machine-learning used to develop test to predict premature births

MACHINE-LEARNING HAS been used to develop a pioneering test which accurately predicted potential premature births in almost three-quarters of women with an asymptomatic high risk.

Researchers at the University of Warwick “trained” a device to look for chemical vapour patterns associated with pre-term birth, using vaginal swabs taken during routine examinations.

After analysing swabs from 216 asymptomatic women, it forecast an outcome of premature delivery in 73 per cent of cases, set out in findings published in Scientific

Reports. It is hoped the technology could lead to a cost-effective, non-invasive, point-of-care test for women identified as at risk of premature delivery, and consequently reduce risks to both mother and baby.

Pre-term birth is the leading cause of death in children under five and there are few accurate tools to predict who is going to have a premature baby.

The technology focused on analysis of volatile organic compounds (VOCs) present in the vagina for a condition called bacterial vaginosis. Previous studies have showed presence of the condition is associated with increased risk of premature births.

Lead author and obstetrics and gynaecology registrar Dr Lauren Lacey, of Warwick Medical School, said: “We’ve demonstrated the technology has good diagnostic accuracy, and in the future it could form part of a care pathway to determine who would deliver pre-term. Although the first test taken earlier in pregnancy is diagnostically less accurate, it could allow interventions to be put in place to reduce the risk of pre-term delivery.”