Predictive Model for Preeclampsia Combining sFlt-1, PlGF, NT-proBNP, and Uric Acid as Biomarkers

Preeclampsia (PE), a hypertensive disorder affecting 2–5% of pregnancies, remains a major cause of maternal and perinatal morbidity and mortality. While improved obstetrical care has reduced mortality, complications such as preterm delivery and intrauterine growth restriction (IUGR) persist. PE's pathophysiology involves placental dysfunction, impaired trophoblast invasion, and an imbalance of angiogenic and antiangiogenic factors. Elevated soluble fms-like tyrosine kinase-1 (sFlt-1), reduced placental growth factor (PlGF), and an increased sFlt-1/PlGF ratio are associated with PE, both before and after clinical onset. The sFlt-1/PlGF ratio is used clinically, but its limitations in management and prognosis of women with abnormally high ratios remain. Pregnancy places stress on the cardiovascular system, and in PE, cardiac diastolic dysfunction leads to elevated N-terminal pro-brain natriuretic peptide (NT-proBNP) levels, correlating with PE severity. Serum uric acid is also consistently elevated in PE, potentially reflecting reduced glomerular filtration. While neither NT-proBNP nor uric acid alone are strong PE predictors, combining them with the sFlt-1/PlGF ratio shows promise. This study aimed to externally validate a previously published machine-learning algorithm predicting early-onset PE using these biomarkers and to compare its performance to the sFlt-1/PlGF ratio alone.

Existing literature highlights the sFlt-1/PlGF ratio as a valuable predictor of PE and its complications, increasing before clinical symptoms appear. However, other biomarkers, when used individually, haven't significantly improved prediction accuracy beyond the sFlt-1/PlGF ratio. Studies show that high sFlt-1/PlGF ratios correlate with shorter time to delivery, but high false-positive rates necessitate hospitalization and close monitoring. NT-proBNP levels are elevated in hypertensive pregnancy disorders, especially severe and preterm PE. Several studies have demonstrated this correlation, showing higher NT-proBNP in early-onset compared to late-onset PE. The role of uric acid in PE prediction is less clear; while hyperuricemia is associated with more severe disease and adverse outcomes, its predictive value is debated. A previous study by Lafuente-Ganuza et al. (2020) developed a machine-learning model incorporating the sFlt-1/PlGF ratio, NT-proBNP, and uric acid, demonstrating improved positive predictive values compared to the sFlt-1/PlGF ratio alone. However, external validation of this model was needed.

This real-world, multicenter observational study used data from the EuroPE study cohort (NCT03231657) including pregnant women (24+0 to 36+6 weeks) with suspected PE from seven Spanish University Hospitals (March 2018–December 2020). After exclusions (twins, loss to follow-up, term pregnancies, etc.), 597 participants and 936 serum samples were analyzed. PE diagnosis followed ISSHP criteria (systolic or diastolic BP >140/90 mmHg, proteinuria >300 mg/24h or 2+ on dipsticks after 20 weeks). Early-preterm PE was defined as onset before 33+6 weeks, late-preterm PE between 34+0 and 36+6 weeks. Suspected PE was defined based on high BP, new/worsening proteinuria, PE-related symptoms, low platelets, elevated liver enzymes, or abnormal Doppler findings. Adverse maternal and perinatal outcomes were defined according to standard criteria. Serum PlGF, sFlt-1, NT-proBNP, and uric acid were measured using automated immunoassays. The MLM included gestational age at admission, chronic hypertension history, and biomarker levels (corrected for gestational age). A random forest-based algorithm was used to predict PE risk (low, moderately low, moderately high, high). A patient was considered positive if any repeated measurements showed moderately high or high risk. Statistical analysis involved ANOVA, Chi-squared tests, and bootstrap methods for confidence intervals. The MLM's performance was compared to the sFlt-1/PlGF ratio alone in predicting preterm PE, early-preterm PE, and late-preterm PE, assessing sensitivity, specificity, PPV, NPV, and AUC.

The study included 597 women, with 34.7% (207/597) developing PE (15.1% early-preterm, 11.2% late-preterm). Women with PE had significantly higher sFlt-1, sFlt-1/PlGF ratio, NT-proBNP, and uric acid levels. The MLM demonstrated significantly better specificity (94.9% vs. 91.0%) and PPV (83.1% vs. 72.8%) in predicting preterm PE compared to the sFlt-1/PlGF ratio alone. While not statistically significant, the MLM also showed slightly better sensitivity (79.6% vs. 77.5%) and NPV (93.7% vs. 92.8%). The MLM's performance improved as the delivery date neared (Figure 2). In predicting early-preterm PE, the MLM showed significantly better specificity and PPV than the sFlt-1/PlGF ratio alone. Similar improvements in specificity and PPV were observed for late-preterm PE prediction with the MLM. Tables 2, 3, and 4 present detailed validation results for different PE subtypes and timeframes.

The study's results confirm the superiority of the combined biomarker model (sFlt-1/PlGF ratio, NT-proBNP, and uric acid) over the sFlt-1/PlGF ratio alone in predicting preterm PE. The improved PPV and specificity offer greater clinical precision, reducing unnecessary interventions in women with suspected PE. The model's increased predictive power, particularly for early-preterm PE and within shorter timeframes to delivery, highlights its potential to optimize management strategies. The findings support the hypothesis that incorporating cardiac (NT-proBNP) and renal (uric acid) markers alongside placental angiogenic factors enhances the predictive accuracy of preeclampsia. The improved NPV suggests the potential for safe outpatient management for a larger proportion of women with suspected PE. The model demonstrates better predictive performance than previously reported models relying solely on the sFlt-1/PlGF ratio.

This study externally validates a machine-learning model incorporating sFlt-1/PlGF ratio, NT-proBNP, and uric acid, showing superior performance in predicting preterm preeclampsia compared to the sFlt-1/PlGF ratio alone. The model's enhanced PPV, specificity, and performance within shorter timeframes to delivery could lead to more precise clinical decisions, avoiding unnecessary interventions. Further research should focus on developing user-friendly software incorporating this algorithm and evaluating its cost-effectiveness in clinical practice.

While the study provides robust external validation, its nested design and strict inclusion criteria might limit the generalizability of findings to diverse populations. The specific cut-off values used are platform-dependent, applicable only to Elecsys immunoassays. Further randomized controlled trials are needed to assess the impact of this algorithm on clinical outcomes and resource utilization.