BackgroundThe management of potential pre-eclamptic patients using the soluble FMS-like tyrosine kinase 1 (sFlt-1)/ placental growth factor (PlGF) ratio is characterised by frequent false-positive results.MethodsA retrospective cohort study was conducted to identify and validate cut-off values, obtained using a machine learning model, for the sFlt-1/PlGF ratio and NT-proBNP that would be predictive of the absence or presence of early-onset pre-eclampsia (PE) in singleton pregnancies presenting at 24 to 33 + 6 weeks of gestation.ResultsFor the development cohort, we defined two sFlt-1/PlGF ratio cut-off values of 23 and 45 to rule out and rule in early-onset PE at any time between 24 and 33 + 6 weeks of gestation. Using an sFlt-1/PlGF ratio cut-off value of 23, the negative predictive value (NPV) for the development of early-onset PE was 100% (95% confidence interval [CI]: 99.5–100). The positive predictive value (PPV) of an sFlt-1/PlGF ratio >45 for a diagnosis of early-onset PE was 49.5% (95% CI: 45.8–55.6). When an NT-proBNP value >174 was combined with an sFlt-1/PlGF ratio >45, the PPV was 86% (95% CI: 79.2–92.6). In the validation cohort, the negative and positive values were very similar to those found for the development cohort.ConclusionsAn sFlt-1/PlGF ratio <23 rules out early-onset PE between 24 and 33 + 6 weeks of gestation at any time, with an NPV of 100%. An sFlt-1/PlGF ratio >45 with an NT-proBNP value >174 significantly enhances the probability of developing early-onset PE.

N-terminal pro-brain natriuretic peptide (NT-proBNP) and uric acid are elevated in pregnancies with preeclampsia (PE). Short-term prediction of PE using angiogenic factors has many false-positive results. Our objective was to validate a machine-learning model (MLM) to predict PE in patients with clinical suspicion, and evaluate if the model performed better than the sFlt-1/PlGF ratio alone. A multicentric cohort study of pregnancies with suspected PE between 24+0 and 36+6 weeks was used. The MLM included six predictors: gestational age, chronic hypertension, sFlt-1, PlGF, NT-proBNP, and uric acid. A total of 936 serum samples from 597 women were included. The PPV of the MLM for PE following 6 weeks was 83.1% (95% CI 78.5–88.2) compared to 72.8% (95% CI 67.4–78.4) for the sFlt-1/PlGF ratio. The specificity of the model was better; 94.9% vs. 91%, respectively. The AUC was significantly improved compared to the ratio alone [0.941 (95% CI 0.926–0.956) vs. 0.901 (95% CI 0.880–0.921), p < 0.05]. For prediction of preterm PE within 1 week, the AUC of the MLM was 0.954 (95% CI 0.937–0.968); significantly greater than the ratio alone [0.914 (95% CI 0.890–0.934), p < 0.01]. To conclude, an MLM combining the sFlt-1/PlGF ratio, NT-proBNP, and uric acid performs better to predict preterm PE compared to the sFlt-1/PlGF ratio alone, potentially increasing clinical precision.