Early prediction of preeclampsia from clinical, multi-omics and laboratory data using random forest model
Comprehensive Overview
This study, published in BMC Pregnancy and Childbirth (2025), investigated whether combining routine blood tests with detailed molecular profiling of a pregnant woman's blood could help predict preeclampsia, a dangerous pregnancy complication marked by sudden high blood pressure after 20 weeks of gestation, well before symptoms typically appear. Preeclampsia affects roughly 1 in 20 pregnancies worldwide and is a leading cause of maternal and infant death, yet accurately predicting who will develop it early in pregnancy has remained difficult, and existing screening tools using only standard clinical measurements have performed poorly. Researchers retrospectively studied 198 pregnant women from three hospitals in China, including those who later developed early-onset or late-onset preeclampsia and those with healthy pregnancies, using blood samples collected in early pregnancy. In addition to standard clinical and laboratory measurements, researchers analyzed hundreds of proteins and metabolites, small molecules produced during metabolism, circulating in each woman's blood plasma. They then used a machine-learning method called random forest, which combines many decision-making models to improve accuracy, to build prediction models and identify which combinations of clinical, protein, and metabolite markers most reliably distinguished women who would go on to develop preeclampsia.
Outcomes and Implications
The most notable finding was that combining clinical characteristics with protein and metabolite biomarkers produced substantially more accurate predictions of preeclampsia than clinical information or laboratory tests alone, particularly for the more severe early-onset form of the disease. The best-performing model for early-onset preeclampsia, which combined seven clinical factors, seven blood metabolites, and five routine laboratory measurements, correctly identified affected women about 9 times out of 10 and correctly ruled out unaffected women at a similarly high rate. A separate model built for the later-onset form of the disease also performed well, though somewhat less accurately at identifying affected women, reflecting the fact that late-onset preeclampsia tends to have subtler, more varied underlying biology. Researchers identified dozens of specific proteins and metabolites that differed meaningfully between women who developed preeclampsia and those who did not, including several molecules never previously linked to the condition, offering new clues about the biological processes driving early-onset and late-onset preeclampsia separately. Known risk factors, including twin pregnancy, higher body mass index, and prior recurrent pregnancy loss, were also confirmed to raise risk. This matters because reliably identifying high-risk women very early in pregnancy could allow doctors to start preventive treatments, such as low-dose aspirin, sooner. The authors conclude that combining clinical, protein, and metabolite data offers a meaningfully more accurate approach to early preeclampsia prediction than currently used methods, though larger studies across more diverse populations are still needed.