Estimating pregnant women’s’ risk of pre-eclampsia using artificial intelligence – with a focus on expectant mothers around the world

The leading cause of maternal deaths continues to be pre-eclampsia. The disease causes 46,000 maternal deaths annually as well as preterm birth, low birth weight and stillbirths. Although living in a country with a lower gross domestic product increases the risk of pre-eclampsia, pregnant women from these countries are underrepresented in pre-eclampsia research. To address this issue, a new risk prediction model called PIERS-ML was developed, including more than 10,000 women from the Global North and Global South. Every woman in the group with suspected pre-eclampsia is examined and the model then predicts her individual risk of adverse pre-eclampsia outcomes within the next two days . In the future, mothers showing symptoms can either be reassured not to worry if their risk for adverse preeclampsia outcomes is classified as low, or they can receive special surveillance and adapted care if their risk is identified as high.

Pre-eclampsia is the new onset of high blood pressure after 19 weeks of gestation and high levels of protein in urin (proteinuria). It leads to organ damage or uteroplacental dysfunction. Globally, it is one of the leading causes of maternal death affecting 2-4% of all pregnancies. This leads to a total of 46,000 pre-eclampsia-related maternal deaths per year. The earlier the onset of pre-eclampsia in the pregnancy, the greater the risk for the expectant mother. More than 99% of pre-eclampsia-related maternal deaths occur in low- and middle-income countries (LMICs). In addition, women with pre-eclampsia tend to give birth earlier, which means that the disease also increases the risk of preterm birth and/or the risk of having a baby with low birth weight. Stillbirths are also more common in sick mothers.

This is why it is so important to quickly assess the risk of every symptomatic pregnant woman in order to prevent severe pre-eclampsia as early as possible. There are some artificial intelligence models that try to predict this individual risk; however, they are rarely tested in LMICs or not externally validated. Therefore, the PIERS-ML model was developed to better predict pre-eclampsia risk levels. The model is an interdisciplinary collaboration between scientists with backgrounds in gynaecology and obstetrics, foetal medicine, and mathematics and statistics from Canada and the UK.

Accurate risk prediction thanks to many and diverse participants

To develop the PIERS-ML, a total of 8,843 women from eleven countries were included to obtain as much data as possible. Mothers from six countries in the Global North as well as from Brazil, Fiji, Pakistan, South Africa, and Uganda were included as LMICs. In addition, 2,900 women from the UK were used for external validation of the model, which increases the model’s accuracy.

Thanks to the large group of participants, the PIERS-ML model proved successful in predicting the risk of adverse maternal outcomes within the first two days after assessment. However, the data were less precise in predicting adverse outcomes within a longer time frame for low-risk mothers, but still accurate for high-risk women.

PIERS-ML can guide prevention and treatment in the future

The data show that various factors influence a pregnant woman’s risk of pre-eclampsia. Women had a higher risk of adverse outcomes if they were younger, had multiple pregnancies, and were treated in a country with lower per capita gross domestic product. The model also revealed more detailed indicators for pre-eclampsia. These include among other things a lower maternal weight, lower oxygen saturation, and higher dipstick proteinuria.

In the future, the model can help reassure low-risk women that they are very unlikely to develop adverse outcomes and relieve their stress and fear. For the women categorised as high-risk, a timely clinical response may be justified. The place of care, any transfer, monitoring, co-interventions, and timed birth can then be tailored to the individual mother. Magnesium sulphate prophylaxis could be adapted according to risk.

The model is so promising because machine learning is suitable to manage many different variables and, to the authors’ knowledge, is the first model for risk stratisfaction in pre-eclampsia that includes data from LMICs and high-income countries.

Paper available at: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10983826/

Ful list of authors: Tünde Montgomery-Csobán, Kimberley Kavanagh, Paul Murray, Chris Robertson, Sarah J E Barry, U Vivian Ukah, Beth A Payne, Kypros H Nicolaides, Argyro Syngelaki, Olivia Ionescu, Ranjit Akolekar, Jennifer A Hutcheon, Laura A Magee, Peter von Dadelszen

DOI: https://doi.org/10.1016%2FS2589-7500(23)00267-4

More information on pre-eclampsia is available in our factsheet