First-trimester multi-modal cell-free DNA analysis for prediction of preterm and term preeclampsia Authors: Rebecca Ertl, Argyro Syngelaki, Olga Frank, Lukas Lüftinger, Eva Lukáčová, Casper Lumby, Adrian Stütz, Stephan Beisken, Andreas E. Posch, Kypros H. Nicolaides Published: 2026 ABSTRACT BACKGROUNDPreeclampsia, which is a leading cause of maternal and perinatal mortality and morbidity, represents a biologically heterogeneous syndrome. First-trimester screening with the Fetal Medicine Foundation (FMF) competing-risks model for preterm preeclampsia enables earlier identification of at-risk pregnancies, allowing risk reduction through aspirin prophylaxis, but depends on Doppler velocimetry and biochemical measurements that limit scalability and offer limited discrimination for term disease. A unified, molecular first-trimester test capable of stratifying risk across the full clinical spectrum of preeclampsia has not been established.OBJECTIVETo determine whether multi-modal, tissue-resolved analysis of first-trimester circulating cell-free DNA (cfDNA), obtained during routine noninvasive prenatal testing (NIPT), enables early prediction of both preterm and term preeclampsia.STUDY DESIGNThis nested case-control study included 125 singleton pregnancies sampled at 11 to 14 weeks’ gestation after quality control (48 controls, 30 preterm preeclampsia, 47 term preeclampsia). For 80 pregnancies, matched placental villi and maternal buffy coat samples were available to derive tissue reference profiles. Plasma cfDNA underwent multi-modal sequencing using Oxford Nanopore Technologies, enabling tissue-resolved analysis of fragmentomic and epigenetic signatures. Separate ensemble machine-learning classifiers were developed to capture distinct disease pathways for preterm (<37 weeks) and term (≥37 weeks) preeclampsia using stratified 10-fold cross-validation. Model discrimination was evaluated using area under the receiver operating characteristic curve (AUROC), sensitivity at predefined specificity thresholds, and comparison with the FMF first-trimester risk score. A population-level simulation of 100,000 pregnancies, applying incidence point estimates of 2.5% for preterm and 7.5% for term PE, was used to derive predictive values and likelihood ratios.RESULTSThe multi-modal cfDNA classifier achieved an AUROC (95% CI) of 0.85 (0.77–0.91) for preterm preeclampsia and 0.84 (0.76–0.91) for term preeclampsia. The FMF score yielded an AUROC of 0.80 (0.70–0.89) for preterm and 0.53 (0.43–0.63) for term PE. At 80% specificity, cfDNA sensitivity was 70.5% for preterm and 72.1% for term preeclampsia, demonstrating improved discrimination for term disease compared with FMF screening. In simulated population-level analysis, positive likelihood ratios were 4.25 (preterm) and 3.83 (term), with negative likelihood ratios of 0.21 and 0.34, respectively, supporting meaningful post-test risk stratification and strong rule-out performance.CONCLUSIONFirst-trimester multi-modal, tissue-resolved cfDNA analysis enables early risk stratification across the full clinical spectrum of preeclampsia from a single routine blood sample. Compared with FMF screening, this approach can potentially improve discrimination for term preeclampsia while providing incremental improvement for preterm disease. The potential for integration into existing NIPT workflows offers a scalable pathway toward unified precision prevention, supporting timely aspirin prophylaxis for preterm preeclampsia and risk-adapted surveillance strategies for term disease. Introduction Preeclampsia is a pregnancy-specific syndrome with complex causes, mechanisms, prediction, management, and prevention challenges.1 It remains a leading cause of maternal and perinatal morbidity and mortality worldwide.2 Short-term US healthcare costs attributable to preeclampsia were estimated at $2.18 billion in 2012 at 3.8% incidence.3 With incidence rising to 7.9% nation-wide in 2021 as well as exceeding 12% in recent multicenter data, and after adjustment for medical cost inflation, the projected annual short-term burden exceeds $7 to 10 billion by 2026 in the United States alone.4, 5, 6 Beyond this immediate impact, preeclampsia is associated with a two- to fourfold increase in long-term cardiometabolic disease risk among affected women and increased vulnerability to prematurity-related, later cardiovascular and neurodevelopmental disorders in their offspring.1,7 Effective prevention requires early risk stratification. Low-dose aspirin initiated before 16 weeks reduces preterm preeclampsia risk.8,9 The Fetal Medicine Foundation (FMF) first-trimester algorithm achieves robust detection of preterm disease but performs less well for term preeclampsia and relies on Doppler and biochemical testing.10,11 Later-pregnancy biomarkers improve short-term prediction but are operationally and economically not suitable for population-wide early screening.12 Preeclampsia is a heterogeneous syndrome in which diverse etiologic pathways converge on endothelial dysfunction, inflammation, and syncytiotrophoblast stress.1,13 Placental ischemia and angiogenic imbalance predominate in preterm disease, whereas term preeclampsia comprises subgroups with and without an antiangiogenic profile that differ in inflammatory signatures and clinical trajectories.14 This heterogeneity likely contributes to the uneven performance of current screening strategies. Circulating cell-free DNA (cfDNA), obtained through routine noninvasive prenatal testing (NIPT), offers a scalable molecular platform for early risk assessment.15, 16, 17 Beyond fetal aneuploidy detection, cfDNA contains fragmentomic and epigenetic tissue-of-origin signals reflective of placental dysfunction and maternal vascular stress. Prior studies, however, have largely focused on single molecular modalities and preterm preeclampsia, with limited applicability to term disease.18, 19, 20, 21, 22 Whether multi-modal, tissue-resolved first-trimester cfDNA analysis can enable unified risk stratification across the full preeclampsia spectrum remains unknown. We evaluate a cfDNA framework for early prediction of both preterm and term preeclampsia (Figure 1). A single first-trimester blood test could operationalize the “turning the pyramid of care” paradigm, shifting care from reactive management to proactive, personalized prevention and surveillance of maternal and fetal risk.23Figure 1Schematic overview of the analytical framework evaluated in this studyA tissue-of-origin read classifier integrates epigenetic and fragmentomic features at the individual cfDNA read level to infer placental and maternal tissue contributions. The tissue-resolved preeclampsia prediction model combines maternal characteristics, differentially methylated CpG segments (DMS), and fragmentomic features to generate risk estimates for preterm and term preeclampsia.Figure 1 dummy alt textErtl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Schematic overview of the analytical framework evaluated in this study A tissue-of-origin read classifier integrates epigenetic and fragmentomic features at the individual cfDNA read level to infer placental and maternal tissue contributions. The tissue-resolved preeclampsia prediction model combines maternal characteristics, differentially methylated CpG segments (DMS), and fragmentomic features to generate risk estimates for preterm and term preeclampsia. Material and methods Study design This study was conducted as a nested case-control analysis within a pregnancy cohort recruited at King’s College Hospital, London, UK. Biobanked samples from women with singleton pregnancies were selected if a routine first-trimester hospital visit had been performed, and complete data on pregnancy outcome, maternal characteristics, medical history, and clinical measurements were available. In line with the study objective of developing and evaluating tissue-resolved cfDNA-based prediction models, the cohort was enriched for matched sample availability. Pregnancies with multifetal gestation, major fetal structural anomalies, or fetal aneuploidy were excluded. Cases were defined as pregnancies complicated by preeclampsia, diagnosed according to the criteria of the American College of Obstetricians and Gynecologists. Preeclampsia was defined as new-onset hypertension after 20 weeks’ gestation or, in women with pre-existing chronic hypertension, as worsening hypertension, with other systemic features including thrombocytopenia, renal insufficiency, impaired liver function, pulmonary oedema, or cerebral or visual symptoms.24 Cases were stratified by gestational age at delivery into preterm (<37+0 weeks) and term preeclampsia (≥37+0 weeks). Control pregnancies were defined as those without hypertensive pregnancy disorders and delivery at ≥37+0 weeks. This study was conducted in accordance with the ethical standards for human research established by the Declaration of Helsinki. The original data collection received approval from the London-Surrey Borders Research Ethics Committee. Written informed consent was obtained from all participants prior to inclusion in the study. Sample collection and processing Maternal peripheral blood was collected at the routine first-trimester hospital visit. Blood was drawn into EDTA tubes and processed according to a standardized protocol. Plasma and, for a subset of samples, buffy coat were separated within 30 minutes of collection, aliquoted, and stored at −80 °C until further analysis. For the same subset, placental tissue samples were collected by chorionic villus sampling. All tissue samples were processed and stored using standardized procedures to preserve DNA integrity. Clinical data included maternal demographic characteristics, obstetric and medical history, and clinical measurements obtained at the first-trimester visit. Pregnancy outcomes, including gestational age at delivery, birth weight, and diagnosis of preeclampsia, were recorded after delivery. The final cohort included 133 pregnancies: 52 controls, 50 term, and 31 preterm preeclampsia; matched placental and maternal buffy coat samples were available for 80 pregnancies and processed to generate reference fragmentomic and epigenetic profiles. Samples failing quality control due to identity inconsistencies, genomic DNA contamination or degradation, or insufficient sequencing output were excluded. High-quality cfDNA and reference tissue datasets were retained for downstream fragmentomic, methylation, and risk prediction analyses (Figure 2).Figure 2Study flow diagramMaternal plasma samples from 133 pregnancies (52 controls, 50 term preeclampsia, and 31 preterm preeclampsia) were processed for cfDNA extraction, sequencing, and bioinformatic analysis. For a subset of 80 pregnancies, matched maternal buffy coat and placental villi samples were additionally sequenced. All samples underwent standardized quality control, and samples failing quality thresholds were excluded prior to downstream analyses.Figure 2 dummy alt textErtl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Study flow diagram Maternal plasma samples from 133 pregnancies (52 controls, 50 term preeclampsia, and 31 preterm preeclampsia) were processed for cfDNA extraction, sequencing, and bioinformatic analysis. For a subset of 80 pregnancies, matched maternal buffy coat and placental villi samples were additionally sequenced. All samples underwent standardized quality control, and samples failing quality thresholds were excluded prior to downstream analyses. DNA extraction, sequencing, and processing cfDNA was extracted from maternal plasma using the aitios cfDNA standardized protocol v1—a validated extraction workflow preserving native cfDNA fragment length distributions—and sequenced on an Oxford Nanopore Technologies platform (Supplemental Data: DNA Extraction and Sequencing). Sequencing data derived from both cfDNA and tissue gDNA samples were processed using a unified bioinformatics pipeline (Supplemental Data: Bioinformatics Processing). Multi-modal preeclampsia prediction The multi-modal, tissue-resolved preeclampsia classifier inferred risk for preterm and term disease from cfDNA fragmentomic and epigenomic signals together with maternal characteristics (Supplemental Data: Multi-Modal Preeclampsia Prediction). Separate classifiers were trained for preterm and term preeclampsia to enable disease subtyping, with pregnancies affected by the respective subtype as the positive class and all remaining samples as controls. Model training followed a modular ensemble framework (Supplemental Data: Model Development). Model evaluation Model performance was assessed using stratified 10-fold cross-validation. Performance was quantified using the area under the receiver operating characteristic curve (AUROC). Sensitivity and accuracy were reported at specificity thresholds of 70%, 80%, and 90% from cross-validated model scores. For benchmarking, ROC curves were constructed using the FMF first-trimester risk score. AUROCs of the cfDNA-based models and the FMF risk score were compared for preterm and term preeclampsia using DeLong’s test for correlated ROC curves. ROC curves were also generated for a combined model including cfDNA features (including maternal characteristics) and the FMF risk score to assess complementary predictive value. To estimate clinically relevant predictive values, we performed a population-level screening simulation. The training dataset was bootstrapped to 100,000 pregnancies, assuming prevalences of 2.5% for preterm and 7.5% for term preeclampsia, reflecting contemporary US incidence estimates and the higher prevalence of term compared with preterm preeclampsia.5,25 Confusion matrices at predefined specificity thresholds were used to calculate predictive values, likelihood ratios, post-test probabilities, and post-test odds. To explore the biological processes captured by predictive cfDNA models, functional enrichment analysis was performed in plasma cfDNA methylation changes (Supplemental Data: Functional Enrichment Analysis). Notably, pathway-level analysis identified potential disease biology mechanisms but did not outperform the integrated ensemble models, which captured higher-order interactions across feature spaces. Results Cohort characterization Maternal plasma samples were obtained from 133 singleton pregnancies at 11+0 to 13+6 weeks’ gestation. Eight plasma cfDNA samples were excluded after quality control (two genomic DNA contamination, one genetic discordance with matched buffy coat, and five insufficient sequencing yield), leaving 125 plasma samples: 48 controls, 30 preterm preeclampsia, and 47 term preeclampsia (Table 1).Table 1Cohort characteristics of plasma samples included for downstream analysesTable 1 dummy alt textCohortControln=48Pretermn=30Termn=47CharacteristicMedian|NIQR|%Median|NIQR|%Median|NIQR|%Age (y)35.132.3–37.835.632.0–39.033.4a28.7–36.8Weight (kg)68.862.0–77.869.564.5–77.475.5a65.5–90.1Height (cm)165160–169165160–168164160–172GA at sampling (wk)13.012.6–13.412.712.2–13.513.012.5–13.4EthnicityWhite2960.48b26.71940.4Black1225.018b60.022a46.8South Asian510.4310.024.3East Asian24.20036.4Mixed0013.312.1Medical historyChronic hypertension0010c33.36a12.8DM Type 1000000DM Type 20026.712.1SLE/APS000000Smoker24.213.312.1Family history of PE24.2310.000Method of conceptionSpontaneous4797.92686.74391.5In vitro fertilization12.1310.048.5Ovulation drugs0013.300ParityNulliparous1939.61343.32757.4Parous, no PE2756.21033.31531.9Parous, PE24.2723.3510.6Clinical characteristicsMAP84.881.5–89.897.2c89.6–105.293.4c85.8–101.5PAPP-A MoM0.70.4–0.90.60.6–0.80.80.4–1.1UtA-PI1.81.5–2.12.3b1.9–2.81.91.4–2.2CRL67.862.1–73.263.256.0–75.166.960.6–72.4GA at delivery (wk)39.638.8–40.933.9c30.1–36.139.0a38.1–40.1Birth weight (g)34193148–36851515c1120–23433155b2725–3523Maternal and pregnancy characteristics are presented as median (interquartile range, IQR) for continuous variables and count N (percentage, %) for categorical variables. Dichotomous variables were compared between case and control groups using Fisher’s exact test (≥5 observations per group); continuous variables were assessed using the Mann–Whitney U test (≥10 observations per group).APS, antiphospholipid syndrome; CRL, Crown–Rump length; DM, diabetes mellitus; GA, gestational age; MAP, mean arterial pressure; NS, not significant; PAPP-A, pregnancy-associated plasma protein-A; SLE, systemic lupus erythematosus; UtA-PI, uterine artery pulsatility index.aP<.05bP<.01cP<.001.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Table 1Cohort characteristics of plasma samples included for downstream analysesTable 1 dummy alt textCohortControln=48Pretermn=30Termn=47CharacteristicMedian|NIQR|%Median|NIQR|%Median|NIQR|%Age (y)35.132.3–37.835.632.0–39.033.4a28.7–36.8Weight (kg)68.862.0–77.869.564.5–77.475.5a65.5–90.1Height (cm)165160–169165160–168164160–172GA at sampling (wk)13.012.6–13.412.712.2–13.513.012.5–13.4EthnicityWhite2960.48b26.71940.4Black1225.018b60.022a46.8South Asian510.4310.024.3East Asian24.20036.4Mixed0013.312.1Medical historyChronic hypertension0010c33.36a12.8DM Type 1000000DM Type 20026.712.1SLE/APS000000Smoker24.213.312.1Family history of PE24.2310.000Method of conceptionSpontaneous4797.92686.74391.5In vitro fertilization12.1310.048.5Ovulation drugs0013.300ParityNulliparous1939.61343.32757.4Parous, no PE2756.21033.31531.9Parous, PE24.2723.3510.6Clinical characteristicsMAP84.881.5–89.897.2c89.6–105.293.4c85.8–101.5PAPP-A MoM0.70.4–0.90.60.6–0.80.80.4–1.1UtA-PI1.81.5–2.12.3b1.9–2.81.91.4–2.2CRL67.862.1–73.263.256.0–75.166.960.6–72.4GA at delivery (wk)39.638.8–40.933.9c30.1–36.139.0a38.1–40.1Birth weight (g)34193148–36851515c1120–23433155b2725–3523Maternal and pregnancy characteristics are presented as median (interquartile range, IQR) for continuous variables and count N (percentage, %) for categorical variables. Dichotomous variables were compared between case and control groups using Fisher’s exact test (≥5 observations per group); continuous variables were assessed using the Mann–Whitney U test (≥10 observations per group).APS, antiphospholipid syndrome; CRL, Crown–Rump length; DM, diabetes mellitus; GA, gestational age; MAP, mean arterial pressure; NS, not significant; PAPP-A, pregnancy-associated plasma protein-A; SLE, systemic lupus erythematosus; UtA-PI, uterine artery pulsatility index.aP<.05bP<.01cP<.001.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Cohort characteristics of plasma samples included for downstream analyses Maternal and pregnancy characteristics are presented as median (interquartile range, IQR) for continuous variables and count N (percentage, %) for categorical variables. Dichotomous variables were compared between case and control groups using Fisher’s exact test (≥5 observations per group); continuous variables were assessed using the Mann–Whitney U test (≥10 observations per group). APS, antiphospholipid syndrome; CRL, Crown–Rump length; DM, diabetes mellitus; GA, gestational age; MAP, mean arterial pressure; NS, not significant; PAPP-A, pregnancy-associated plasma protein-A; SLE, systemic lupus erythematosus; UtA-PI, uterine artery pulsatility index. P<.05 P<.01 P<.001. Control pregnancies resulted in delivery of phenotypically normal neonates at term with birthweights appropriate for gestational age; none of the control participants had a history of chronic hypertension. Compared with controls, both preterm and term preeclampsia groups showed significant differences in first-trimester risk factors, including mean arterial pressure (P<.001 for both), and a higher proportion of participants self-identifying as Black ethnicity (preterm: P<.01; term: P<.05). Differences specific to the preterm preeclampsia group included elevated uterine artery pulsatility index (P<.01). Maternal age and weight were modestly higher in the term preeclampsia group compared with controls (both P<.05), while gestational age at sampling did not differ between groups. Perinatal outcomes differed markedly across cohorts. Preterm preeclampsia was associated with significantly earlier delivery and lower birthweight than controls (both P<.001). Term preeclampsia pregnancies delivered earlier and had lower birthweights (P<.05 and P<.01, respectively). No other comparisons were made for groups with less than 5 observations for dichotomous variables and 10 observations for continuous variables due to a lack of statistical power. Sample processing The optimized cfDNA extraction workflow increased cfDNA recovery from patient plasma relative to the QIAamp MinElute ccfDNA Midi Kit, with a median ∼1.7-fold increase in total yield across matched samples (paired t test, P<.0001). Recovery of longer cfDNA fragments also increased (∼2.5-fold). Libraries generated using this workflow produced a median of 89.09 Gb raw and 69.84 Gb aligned data per sample, exceeding PromethION vendor benchmarks for cfDNA (Supplemental Data: Results—Sample Processing). Multi-modal preeclampsia prediction We developed a multi-modal classifier integrating fragmentomic, epigenetic, and clinical features to predict preterm and term preeclampsia. As no single modality fully separated cases from controls, an ensemble approach combining multiple feature spaces was used to capture complementary biological signals. Individual component models showed limited discrimination (AUROCs 0.60–0.70), whereas the ensemble approach improved performance to 0.85 (95% CI 0.77–0.91) for preterm and 0.84 (0.76–0.91) for term preeclampsia (Figure 3).Figure 3Predictive performance of multi-modal cfDNA models for preterm and term preeclampsiaReceiver operating characteristic (ROC) curves comparing the performance of the Fetal Medicine Foundation (FMF) first-trimester risk score (blue), the cfDNA-based multi-modal model (turquoise), and the combined model integrating cfDNA features with the FMF risk score (purple). (A) ROC curves for prediction of preterm preeclampsia. (B) ROC curves for prediction of term preeclampsia. The area under the ROC curve is reported for each model, with corresponding 95% confidence intervals.Figure 3 dummy alt textErtl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Predictive performance of multi-modal cfDNA models for preterm and term preeclampsia Receiver operating characteristic (ROC) curves comparing the performance of the Fetal Medicine Foundation (FMF) first-trimester risk score (blue), the cfDNA-based multi-modal model (turquoise), and the combined model integrating cfDNA features with the FMF risk score (purple). (A) ROC curves for prediction of preterm preeclampsia. (B) ROC curves for prediction of term preeclampsia. The area under the ROC curve is reported for each model, with corresponding 95% confidence intervals. The FMF score, calculated based on parameters shown in Table 1 (excluding PlGF), achieved 0.80 (0.70–0.89) for preterm preeclampsia and 0.53 (0.43–0.63) for term preeclampsia. The combined model improved AUROC by ∼9 percentage points for preterm disease and ∼29 percentage points for term disease relative to the FMF score alone. The multi-modal cfDNA classifier showed higher sensitivity than the FMF score at 70% and 80% specificity (Table 2). At 90% specificity, the FMF score retained a modest sensitivity advantage (∼6 percentage points), consistent with its design. In contrast, cfDNA-derived signals showed a more continuous risk distribution, providing more balanced performance across clinically relevant specificity ranges, particularly at 70% to 80%.Table 2Discriminatory performance of FMF, cfDNA-based, and combined models for preterm and term preeclampsiaTable 2 dummy alt textPreterm preeclampsiaTerm preeclampsiaSensitivity at % specificitySensitivity at % specificityModelAUROC (95% CI)708090AUROC (95% CI)708090FMF risk score0.80 (0.70–0.89)69.967.458.10.53 (0.43–0.63)33.617.03.0Multi-modal cfDNA0.85 (0.77–0.91)87.370.552.10.84 (0.76–0.91)83.172.164.7Multi-modal cfDNA+FMF0.89 (0.82–0.94)95.081.156.70.82 (0.74–0.88)74.065.556.2Area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals is shown for each prediction model for preterm and term preeclampsia. Sensitivity values are reported at predefined specificity thresholds of 70%, 80%, and 90%. All pairwise models comparisons using DeLong’s test for correlated ROC curves were significant (P<.001) except for the pairs multi-modal cfDNA vs multi-modal cfDNA+FMF for both conditions.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Table 2Discriminatory performance of FMF, cfDNA-based, and combined models for preterm and term preeclampsiaTable 2 dummy alt textPreterm preeclampsiaTerm preeclampsiaSensitivity at % specificitySensitivity at % specificityModelAUROC (95% CI)708090AUROC (95% CI)708090FMF risk score0.80 (0.70–0.89)69.967.458.10.53 (0.43–0.63)33.617.03.0Multi-modal cfDNA0.85 (0.77–0.91)87.370.552.10.84 (0.76–0.91)83.172.164.7Multi-modal cfDNA+FMF0.89 (0.82–0.94)95.081.156.70.82 (0.74–0.88)74.065.556.2Area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals is shown for each prediction model for preterm and term preeclampsia. Sensitivity values are reported at predefined specificity thresholds of 70%, 80%, and 90%. All pairwise models comparisons using DeLong’s test for correlated ROC curves were significant (P<.001) except for the pairs multi-modal cfDNA vs multi-modal cfDNA+FMF for both conditions.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Discriminatory performance of FMF, cfDNA-based, and combined models for preterm and term preeclampsia Area under the receiver operating characteristic curve (AUROC) with 95% confidence intervals is shown for each prediction model for preterm and term preeclampsia. Sensitivity values are reported at predefined specificity thresholds of 70%, 80%, and 90%. All pairwise models comparisons using DeLong’s test for correlated ROC curves were significant (P<.001) except for the pairs multi-modal cfDNA vs multi-modal cfDNA+FMF for both conditions. Subgroup analysis stratified by baseline risk using the FMF first-trimester risk score (low vs high risk threshold 1 in 100) assessed robustness across risk strata. Within the cohort, FMF classified 7 of 30 preterm and 18 of 47 term preeclampsia cases as low risk, indicating that 23.3% and 38.3% of cases would not have been identified by FMF-based screening alone. The cfDNA-based model maintained discriminative performance across baseline risk strata and identified additional cases not captured by FMF risk stratification (2 preterm and 13 term), while preserving specificity and achieving even higher specificity in high-risk pregnancies. Simulated population-level screening performance In the simulated real-world screening cohort, the positive likelihood ratio was 4.25 for preterm preeclampsia and 3.83 for term preeclampsia. For preterm disease, this exceeded the FMF score (positive likelihood ratio 3.72). The negative likelihood ratio for preterm preeclampsia was 0.214, lower than the FMF value of 0.325, indicating greater residual risk reduction after a negative cfDNA-based test. Using these likelihood ratios, the negative predictive value for preterm preeclampsia was 99.5% and the positive predictive value 9.8%. For term preeclampsia, the negative and positive predictive values were 97.3% and 23.7%, respectively (Table 3). These results reflect the low population prevalence of both conditions while highlighting strong rule-out performance of the cfDNA-based classifier, particularly for placental-associated preterm disease.Table 3Simulated population-level screening performance for preterm and term preeclampsiaTable 3 dummy alt textPreterm preeclampsiaTerm preeclampsiaPrevalence2.5%7.5%Accuracy80.6%80.4%Sensitivity82.8%72.5%Specificity80.5%81.1%PPV9.8%23.7%NPV99.5%97.3%Screen positive rate21.1%23.0%Pretest prob. (odds)2.5% (0.026)7.5% (0.081)Post-test +Prob. (odds)9.8% (0.109)23.7% (0.311)Post-test −Prob. (odds)0.5% (0.005)2.7% (0.027)Likelihood ratio+4.2493.830Likelihood ratio−0.2140.339Reported metrics include accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV), screen-positive rate, and corresponding pre- and post-test probabilities and odds. Likelihood ratios for positive and negative test results are shown to quantify the change from pretest to post-test probability.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Table 3Simulated population-level screening performance for preterm and term preeclampsiaTable 3 dummy alt textPreterm preeclampsiaTerm preeclampsiaPrevalence2.5%7.5%Accuracy80.6%80.4%Sensitivity82.8%72.5%Specificity80.5%81.1%PPV9.8%23.7%NPV99.5%97.3%Screen positive rate21.1%23.0%Pretest prob. (odds)2.5% (0.026)7.5% (0.081)Post-test +Prob. (odds)9.8% (0.109)23.7% (0.311)Post-test −Prob. (odds)0.5% (0.005)2.7% (0.027)Likelihood ratio+4.2493.830Likelihood ratio−0.2140.339Reported metrics include accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV), screen-positive rate, and corresponding pre- and post-test probabilities and odds. Likelihood ratios for positive and negative test results are shown to quantify the change from pretest to post-test probability.Ertl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Simulated population-level screening performance for preterm and term preeclampsia Reported metrics include accuracy, sensitivity, specificity, positive and negative predictive values (PPV, NPV), screen-positive rate, and corresponding pre- and post-test probabilities and odds. Likelihood ratios for positive and negative test results are shown to quantify the change from pretest to post-test probability. Biological motivation Tissue-resolved features improved classification in a subtype-specific manner. Fragmentomic features contributed most to preterm prediction (McNemar’s P<.05), consistent with placental pathology, whereas maternal-derived accessibility signals were more informative for term disease (P<.01). Feature importance analysis showed minimal correlation between models (P<.0001; Figure 4), supporting distinct early biological processes.Figure 4Subtype-specific feature learning and differential cfDNA methylation patterns(A) Bland–Altman plot comparing ranked nucleosome accessibility feature importances derived from the preterm and term preeclampsia classifiers, shown for an illustrative feature modality. The absence of systematic agreement and the wide dispersion around zero indicate that nucleosome accessibility features are weighted differently between models, consistent with subtype-specific signal learning. (B) Volcano plot showing differentially methylated segments (DMSs) in first-trimester plasma cfDNA comparing preterm preeclampsia cases vs controls. (C) Volcano plot showing DMSs comparing term preeclampsia cases vs controls. DMSs passing a false discovery rate (FDR) threshold of 5% are highlighted, with hypomethylated segments shown in turquoise and hypermethylated segments in orange. The horizontal blue line indicates the FDR-adjusted significance threshold.Figure 4 dummy alt textErtl. First-trimester multi-modal cell-free DNA analysis for prediction. AJOG Glob Rep 2026. Subtype-specific feature learning and differential cfDNA methylation patterns (A) Bland–Altman plot comparing ranked nucleosome accessibility feature importances derived from the preterm and term preeclampsia classifiers, shown for an illustrative feature modality. The absence of systematic agreement and the wide dispersion around zero indicate that nucleosome accessibility features are weighted differently between models, consistent with subtype-specific signal learning. (B) Volcano plot showing differentially methylated segments (DMSs) in first-trimester plasma cfDNA comparing preterm preeclampsia cases vs controls. (C) Volcano plot showing DMSs comparing term preeclampsia cases vs controls. DMSs passing a false discovery rate (FDR) threshold of 5% are highlighted, with hypomethylated segments shown in turquoise and hypermethylated segments in orange. The horizontal blue line indicates the FDR-adjusted significance threshold. Ranked feature importances across ensemble components revealed minimal or weak negative correlation between preterm and term preeclampsia models (Spearman correlation, P<.0001; Figure 4), indicating that the classifiers learned distinct signal combinations per subtype, supporting hypothesis that subtypes reflect partially divergent biological processes detectable as early as the first-trimester. Functional enrichment analysis of differentially methylated segments in plasma identified 2621 hypomethylated and 681 hypermethylated segments for preterm preeclampsia among ∼1.4 million candidate DMS, compared with 569 hypomethylated and 42 hypermethylated segments for term preeclampsia. Consistent with greater placental involvement in preterm disease, differential methylation—particularly hypomethylation—is more pronounced in preterm than term preeclampsia, which shows a more limited and heterogeneous signal. These findings support subtype-specific rather than uniform epigenetic patterns. For preterm preeclampsia, enrichment analysis highlighted angiogenesis and placental development pathways, including the RHO GTPase cycle (FDR<0.01) and upstream Signaling by Receptor Tyrosine Kinases (FDR<0.05). Pathways involving cadherin-11 expression and function were also enriched, implicating calcium-dependent cell-cell adhesion mechanisms required for trophoblast invasion and placental morphogenesis.26 Term preeclampsia showed fewer enriched pathways overall. Most prominently, pathways related to gastrulation and early developmental patterning were overrepresented (FDR<0.05). Although gastrulation is completed early in embryogenesis, these pathways include regulatory programs reused in placental differentiation and maternal-fetal interface remodeling.27 Comment Principal findings This study demonstrates the feasibility of predicting both preterm and term preeclampsia from first-trimester cfDNA derived from routine prenatal plasma samples. By integrating fragmentomic and methylation-derived signals with tissue-resolved features and maternal clinical characteristics, the multi-modal classifier extends prior cfDNA approaches that have largely focused on preterm preeclampsia or single molecular modalities.19, 20, 21, 22 To our knowledge, these findings provide the first evidence that cfDNA captures biologically relevant information across the full clinical spectrum of preeclampsia, including term disease, from a routine first-trimester blood sample. Unlike conventional biochemical or single-marker assays, parallel modeling of preterm and term pathways using high-dimensional cfDNA features enables improved detection across both phenotypes from a single assay, supporting risk-adapted prevention and management strategies at the population level. Results in the context of what is known cfDNA fragmentation and methylation patterns reflect cell-type-specific chromatin organization and regulatory states.21 The predictive contribution of these features is consistent with early placental dysfunction and maternal vascular or immune perturbations preceding clinical disease. Tissue-resolved cfDNA features allow attribution of signals to maternal or placental origins, helping resolve the differing pathophysiologies of preterm and term preeclampsia and supporting the view of preeclampsia as a biologically heterogeneous disorder.28 Improved molecular subtyping may ultimately inform targeted preeclampsia therapies beyond current approaches. Functional enrichment analysis provides biological context for the subtype-specific signals captured by the classifiers. For preterm preeclampsia, enrichment of RHO GTPase signaling, receptor tyrosine kinase pathways, and cadherin-mediated adhesion is consistent with impaired placentation, including disrupted trophoblast invasion, angiogenesis, and endothelial function.29 Term preeclampsia showed fewer enriched pathways, including developmental patterning programs, consistent with a more heterogeneous and less placenta-centric pathophysiology.19 Together, these observations support the biological plausibility of the multi-modal cfDNA framework and the integration of fragmentomic, methylation, and tissue-resolved signals for early subtype-specific risk prediction. Clinical implications All cfDNA features were derived from plasma samples and sequencing data compatible with standard first-trimester NIPT workflows, enabling screening from a single blood draw. This approach may facilitate early risk stratification and help address gaps in preeclampsia prevention.30 Integration into existing NIPT workflows may be particularly relevant for term preeclampsia, where effective first-trimester screening remains limited, although prospective validation in larger, unselected populations is required. The scalability of cfDNA-based approaches relates primarily to workflow integration rather than intrinsic cost or infrastructure simplicity. While leveraging existing NIPT infrastructure may streamline implementation, multi-modal cfDNA analysis remains dependent on sequencing capacity and bioinformatics, potentially limiting accessibility in low-resource settings. Biochemical markers—such as inhibin-A, PAPP-A2, Sonic Hedgehog31, 32, 33—and emerging point-of-care approaches therefore remain important, particularly given their accessibility and evolving performance.34 Accordingly, cfDNA-based screening is best viewed as a complementary, NIPT-compatible platform for unified first-trimester molecular risk stratification rather than a replacement for existing strategies. Research implications The FMF algorithm was designed to identify early-onset and preterm preeclampsia and relies heavily on clinical and hemodynamic markers that perform well in high-risk populations.8,35 It was not intended to predict term disease, which is more heterogeneous and less tightly linked to early placental insufficiency. The complementary performance profiles observed here should therefore be interpreted considering these differing objectives rather than as a direct competition between methods. In this study, cfDNA-based models improved discrimination for term preeclampsia and provided incremental gains for preterm disease. These findings are limited by the exploratory design and sample size; prospective validation is required to define how cfDNA may complement or integrate with screening strategies such as the FMF algorithm. Further refinement of tissue-resolved modeling may enable improved biological disease subtyping and provide insight into early mechanisms underlying maternal and fetal complications beyond preeclampsia. Strengths and limitations This study represents, to our knowledge, the largest cfDNA-based preeclampsia prediction analysis from routine first-trimester samples in a heterogeneous multi-ancestry cohort and the first to apply tissue-resolved features derived from single-molecule sequencing. The availability of matched maternal and placental tissue enabled biological anchoring of cfDNA signals, strengthening tissue-of-origin inference and interpretability. However, the cohort was derived from Biobanked specimens enriched for well-characterized cases and matched sample availability and may not reflect an unselected screening population. A technical strength is the multi-modal approach, which preserves native cfDNA fragment lengths and endogenous methylation on the same molecules, enabling integrated fragmentomic and epigenetic analysis without PCR or conversion-induced bias. The primary limitation is cohort size, which restricts subgroup analyses, increases the risk of overfitting, and requires cautious interpretation of performance estimates. Population-level performance estimates are based on simulation and may overestimate real-world performance in the absence of external validation. No formal power calculation was performed, given the exploratory design. Larger prospective studies are required to validate performance, assess generalizability and robustness in the presence of confounders, and enable finer phenotypic stratification, including evaluation of differential response to preventive interventions such as aspirin. Conclusions First-trimester cfDNA sequencing enables prediction of both preterm and term preeclampsia by capturing molecular signals reflecting placental and maternal contributions across the disease spectrum. A full-spectrum early screening test could support earlier risk stratification and inform prevention, surveillance, and timing-of-birth strategies, particularly for term disease. By providing an objective molecular risk assessment early in pregnancy, this approach addresses the quality gap highlighted by the Society for Maternal-Fetal Medicine, namely the under-identification of at-risk pregnancies and the suboptimal uptake of evidence-based preventive interventions for preterm disease.30 Compatibility with routine NIPT workflows provides a practical and scalable pathway for combined fetal and maternal screening, positioning cfDNA-based approaches as a core component of next-generation prenatal care. CRediT authorship contribution statement Rebecca Ertl: Writing – original draft, Visualization, Investigation, Conceptualization. Argyro Syngelaki: Investigation, Data curation. Olga Frank: Methodology, Investigation. Lukas Lüftinger: Software, Data curation. Eva Lukáčová: Investigation. Casper Lumby: Software, Data curation. Adrian Stütz: Investigation. Stephan Beisken: Writing – review & editing, Writing – original draft, Validation, Methodology, Investigation. Andreas E. Posch: Writing – review & editing, Resources, Funding acquisition, Conceptualization. Kypros H. Nicolaides: Writing – review & editing, Supervision, Resources, Funding acquisition.