Preeclampsia (PE) is a pregnancy complication defined by new onset hypertension and proteinuria or other maternal organ damage after 20 weeks of gestation. Although non-invasive prenatal testing (NIPT) has been widely used to detect fetal chromosomal abnormalities during pregnancy, its performance in combination with maternal risk factors to screen for PE has not been extensively validated. Our aim was to develop and validate classifiers that predict early- or late-onset PE using the maternal plasma cell-free DNA (cfDNA) profile and clinical risk factors.
We retrospectively collected and analyzed NIPT data of 2,727 pregnant women aged 24–45 years from four hospitals in China, which had previously been used to screen for fetal aneuploidy at 12 + 0 ~ 22 + 6 weeks of gestation. According to the diagnostic criteria for PE and the time of diagnosis (34 weeks of gestation), a total of 143 early-, 580 late-onset PE samples and 2,004 healthy controls were included. The wilcoxon rank sum test was used to identify the cfDNA profile for PE prediction. The Fisher’s exact test and Mann–Whitney U-test were used to compare categorical and continuous variables of clinical risk factors between PE samples and healthy controls, respectively. Machine learning methods were performed to develop and validate PE classifiers based on the cfDNA profile and clinical risk factors.
By using NIPT data to analyze cfDNA coverages in promoter regions, we found the cfDNA profile, which was differential cfDNA coverages in gene promoter regions between PE and healthy controls, could be used to predict early- and late-onset PE. Maternal age, body mass index, parity, past medical histories and method of conception were significantly differential between PE and healthy pregnant women. With a false positive rate of 10%, the classifiers based on the combination of the cfDNA profile and clinical risk factors predicted early- and late-onset PE in four datasets with an average accuracy of 89 and 80% and an average sensitivity of 63 and 48%, respectively.
Incorporating cfDNA profiles in classifiers might reduce performance variations in PE models based only on clinical risk factors, potentially expanding the application of NIPT in PE screening in the future.