Their goal was to identify a small signal detectable in the first 16 weeks of pregnancy that could form the basis of a simple, low-cost diagnostic test for use in low-, middle-, and high-income countries. To estimate the accuracy of the machine learning model, the researchers initially built the models with data from the discovery cohort, then confirmed the results by testing their performance on data from women in the validation cohort.
When you reduce preeclampsia, you likely also reduce preterm birth. It’s a double whammy to good effect.
A prediction model using a set of nine urine metabolites was highly accurate, the researchers found. These urine markers, in samples collected before 16 weeks of pregnancy, strongly predicted who later developed preeclampsia. Test performance was measured by a statistical measure known as the area under the characteristic curve used in machine learning. An AUC of 1 indicates perfect prediction for a test with two possible outcomes, whereas an AUC of 0.5 indicates no predictive value, similar to the outcome obtained from a coin toss. For urine markers, the AUC was 0.88 in the discovery cohort and 0.83 in the validation cohort, indicating high predictive power.
Measuring the same set of urine metabolites in samples collected during pregnancy produced similar predictive power, with an AUC of 0.89 in the discovery cohort and 0.87 in the validation cohort.
The researchers confirmed that their model had stronger predictive power than clinical characteristics associated with a pregnant woman’s risk of preeclampsia, such as chronic hypertension, high body mass index, and carrying twins.
A set of nine proteins measured in blood performed almost as strongly, with an AUC of 0.84.
The researchers also developed a predictive model that combined participants’ clinical characteristics with urinary metabolites, which enabled them to predict preeclampsia early in pregnancy with an AUC of 0.96. Clinical characteristics in the combined model are data that are already collected as part of the standard medical record, such as patient age, height, body mass index, and prepregnancy hypertension.
“This data collection is routine and can serve as the first level of triage,” says Aghipour. “We envision that patients whose data show them to be at risk may receive more extensive urine testing.”
Unraveling disease biology
Stanford Medicine researchers are also opening windows into the biology of preeclampsia. Another study, published in February the natureCell-free RNA measurements have been used to reveal biological clues about how preeclampsia arises.
“The ability to listen to conversations during pregnancy, measuring molecules simultaneously from the pregnant woman, the fetus and the placenta, is very helpful in giving us clues about what biological changes contribute to the disease,” said Mira Mofarez, Ph.D. the nature Paper, who was a graduate student in bioengineering when the study was conducted. The paper’s senior author is Stephen Koek, DPhil, professor of bioengineering and applied physics.
“The most striking changes occurred before 20 weeks’ gestation, whereas a preeclampsia diagnosis is usually made at more than 30 weeks’ gestation,” Mofarrez said. “It was amazing. We would expect changes in gene signaling when you see clinical symptoms, and that was happening long before pregnancy.”
Using 404 blood samples from 199 pregnant women, Mofarrez and his colleagues identified a set of 18 genes whose activity early in pregnancy predicted the development of preeclampsia.
The genes are consistent with what is known about the biology of how the disorder develops, he noted.
Scientists hypothesize that in pre-eclamptic pregnancies, the placenta is not fully developed; Its blood vessels may be very small. First, this is because the fetus is small and does not require much nutrition.
“But later in pregnancy, the fetus has grown, sending signals for more nutrients,” Mofarez said. “At that point, the only solution to the small blood vessels is more blood flow, so we see high blood pressure.” In severe cases, the pressure can cause a premature separation of the placenta from the uterine lining, creating an emergency where the baby must be delivered immediately.
The gene activity signals that Moufarez and his colleagues identified came from genes involved in pathways consistent with the development of preeclampsia, such as tissues related to the endothelial system, placenta and brain. (The brain is relevant because full-blown eclampsia causes seizures.) The scientists plan to use the work as a basis for future research into how the condition develops.
Scientists involved in both studies will validate their predictive tests in a much larger, more diverse population of women with the goal of developing tests for universal use.
Knowing more about how preeclampsia develops and how to predict it could have profound benefits for the world’s most vulnerable mothers, the researchers said, with an estimated 86% of global maternal deaths occurring in Asia and sub-Saharan Africa.
“This is where this type of testing is really needed, where resources are scarce,” Marich said. Unlike women in high-income countries, many women in low-income regions give birth far from hospitals, limiting their access to emergency care when they show symptoms of preeclampsia or eclampsia. “If we can identify which pregnancies are at high risk, we can help move those women to health care facilities and prevent deaths.”
The patterns was supported by the March of Dimes Prematurity Research Center at Stanford University School of Medicine, the Stanford Maternal and Child Health Research Institute, the Christopher Hess Research Fund, the National Institutes of Health (grants 1R01HL139844, 5RM1HG00773507 and R35GM383507 to the Wellford Fund). , Alfred E. Mann Foundation, the Bill and Melinda Gates Foundation, the Thomas C. and Joan M. Merrigan Endowment of Stanford University, and the Chan Zuckerberg BioHub Microbiome Initiative.
The the nature The research was supported by Chan Zuckerberg BioHub, Global Alliance to Prematurity and Stillbirth, March of Dimes Foundation, National Science Foundation (Grant DGE 1656518), Benchmark Stanford Graduate Fellowship, Stanford ChEM-H Chemistry Biology Interface Training Program. , and H&H Evergreen Fund.