Preeclampsia, characterized by arterial hypertension accompanied by proteinuria, is a pregnancy complication that most often arises after the twentieth to twenty-fifth week of gestation. It represents a major cause of morbidity and mortality for both mother and fetus, affecting an estimated 5 to 7% of pregnancies worldwide and accounting for more than 70,000 maternal deaths and 500,000 fetal deaths each year. In France, over the 2010–2016 period, approximately 5.2% of pregnancies were complicated by gestational hypertension and 2% by preeclampsia or eclampsia. Current early-screening methods, particularly those based on the combined detection of biomarkers such as sFLT1, sEng, and PlGF, remain imperfect, as their predictive performance declines as pregnancy progresses. It is within this context that an approach allowing for earlier diagnosis was sought.
The peroxisome proliferator-activated receptor gamma (PPARγ) plays an essential role in placental development, and certain of its single-nucleotide polymorphisms (SNPs) have been associated with increased susceptibility to pregnancy disorders. The study examined three PPARγ SNPs — Pro12Ala, C1431T, and C681G — combined with nine clinical factors, using data from 1,648 women drawn from the EDEN cohort, of whom 35 had experienced a preeclamptic pregnancy and 1,613 a normal pregnancy. Genotyping was performed on DNA extracted from leukocytes using two techniques (LightCycler hybridization probes followed by a TaqMan procedure), with a success rate above 98% for each of the three SNPs.
Univariate analysis identified the C1431T SNP as the only factor significantly associated with preeclampsia (p < 0.05), with an odds ratio ranging from 4.90 to 8.75 (95% confidence interval). In parallel, three multivariate variable-selection methods retained seven potential predictors: the maternal C1431T and C681G variants, obesity, body mass index, number of pregnancies, primiparity, smoking, and educational level. These seven variables were used to train eight machine learning algorithms. The boosted decision tree (boost tree) model proved the most effective, with an accuracy and area under the ROC curve of 0.971 and 0.991 on the training set, and 0.951 and 0.701 on the test set. Building on this model, the authors constructed a decision tree confirming the significant association of the PPARγ C1431T variant with susceptibility to preeclampsia. According to them, this tool could contribute to the screening of at-risk pregnancies from the very earliest stages and serve as a decision-support reference in clinical practice.