TY - JOUR
T1 - The role of cell-free DNA biomarkers and patient data in the early prediction of preeclampsia
T2 - an artificial intelligence model
AU - Khalil, Asma
AU - Bellesia, Giovanni
AU - Norton, Mary E.
AU - Jacobsson, Bo
AU - Haeri, Sina
AU - Egbert, Melissa
AU - Malone, Fergal D.
AU - Wapner, Ronald J.
AU - Roman, Ashley
AU - Faro, Revital
AU - Madankumar, Rajeevi
AU - Strong, Noel
AU - Silver, Robert M.
AU - Vohra, Nidhi
AU - Hyett, Jon
AU - MacPherson, Cora
AU - Prigmore, Brittany
AU - Ahmed, Ebad
AU - Demko, Zachary
AU - Ortiz, J. Bryce
AU - Souter, Vivienne
AU - Dar, Pe'er
N1 - Publisher Copyright:
© 2024
PY - 2024/11
Y1 - 2024/11
N2 - Background: Accurate individualized assessment of preeclampsia risk enables the identification of patients most likely to benefit from initiation of low-dose aspirin at 12 to 16 weeks of gestation when there is evidence for its effectiveness, and enables the guidance of appropriate pregnancy care pathways and surveillance. Objective: The primary objective of this study was to evaluate the performance of artificial neural network models for the prediction of preterm preeclampsia (<37 weeks’ gestation) using patient characteristics available at the first antenatal visit and data from prenatal cell-free DNA screening. Secondary outcomes were prediction of early-onset preeclampsia (<34 weeks’ gestation) and term preeclampsia (≥37 weeks’ gestation). Methods: This secondary analysis of a prospective, multicenter, observational prenatal cell-free DNA screening study (SMART) included singleton pregnancies with known pregnancy outcomes. Thirteen patient characteristics that are routinely collected at the first prenatal visit and 2 characteristics of cell-free DNA (total cell-free DNA and fetal fraction) were used to develop predictive models for early-onset (<34 weeks), preterm (<37 weeks), and term (≥37 weeks) preeclampsia. For the models, the “reference” classifier was a shallow logistic regression model. We also explored several feedforward (nonlinear) neural network architectures with ≥1 hidden layers, and compared their performance with the logistic regression model. We selected a simple neural network model built with 1 hidden layer and made up of 15 units. Results: Of the 17,520 participants included in the final analysis, 72 (0.4%) developed early-onset, 251 (1.4%) preterm, and 420 (2.4%) term preeclampsia. Median gestational age at cell-free DNA measurement was 12.6 weeks, and 2155 (12.3%) had their cell-free DNA measurement at ≥16 weeks’ gestation. Preeclampsia was associated with higher total cell-free DNA (median, 362.3 vs 339.0 copies/mL cell-free DNA; P<.001) and lower fetal fraction (median, 7.5% vs 9.4%; P<.001). The expected, cross-validated area under the curve scores for early-onset, preterm, and term preeclampsia were 0.782, 0.801, and 0.712, respectively, for the logistic regression model, and 0.797, 0.800, and 0.713, respectively, for the neural network model. At a screen-positive rate of 15%, sensitivity for preterm preeclampsia was 58.4% (95% confidence interval, 0.569–0.599) for the logistic regression model and 59.3% (95% confidence interval, 0.578–0.608) for the neural network model. The contribution of both total cell-free DNA and fetal fraction to the prediction of term and preterm preeclampsia was negligible. For early-onset preeclampsia, removal of the total cell-free DNA and fetal fraction features from the neural network model was associated with a 6.9% decrease in sensitivity at a 15% screen-positive rate, from 54.9% (95% confidence interval, 52.9–56.9) to 48.0% (95% confidence interval, 45.0–51.0). Conclusion: Routinely available patient characteristics and cell-free DNA markers can be used to predict preeclampsia with performance comparable to that of other patient characteristic models for the prediction of preterm preeclampsia. Logistic regression and neural network models showed similar performance.
AB - Background: Accurate individualized assessment of preeclampsia risk enables the identification of patients most likely to benefit from initiation of low-dose aspirin at 12 to 16 weeks of gestation when there is evidence for its effectiveness, and enables the guidance of appropriate pregnancy care pathways and surveillance. Objective: The primary objective of this study was to evaluate the performance of artificial neural network models for the prediction of preterm preeclampsia (<37 weeks’ gestation) using patient characteristics available at the first antenatal visit and data from prenatal cell-free DNA screening. Secondary outcomes were prediction of early-onset preeclampsia (<34 weeks’ gestation) and term preeclampsia (≥37 weeks’ gestation). Methods: This secondary analysis of a prospective, multicenter, observational prenatal cell-free DNA screening study (SMART) included singleton pregnancies with known pregnancy outcomes. Thirteen patient characteristics that are routinely collected at the first prenatal visit and 2 characteristics of cell-free DNA (total cell-free DNA and fetal fraction) were used to develop predictive models for early-onset (<34 weeks), preterm (<37 weeks), and term (≥37 weeks) preeclampsia. For the models, the “reference” classifier was a shallow logistic regression model. We also explored several feedforward (nonlinear) neural network architectures with ≥1 hidden layers, and compared their performance with the logistic regression model. We selected a simple neural network model built with 1 hidden layer and made up of 15 units. Results: Of the 17,520 participants included in the final analysis, 72 (0.4%) developed early-onset, 251 (1.4%) preterm, and 420 (2.4%) term preeclampsia. Median gestational age at cell-free DNA measurement was 12.6 weeks, and 2155 (12.3%) had their cell-free DNA measurement at ≥16 weeks’ gestation. Preeclampsia was associated with higher total cell-free DNA (median, 362.3 vs 339.0 copies/mL cell-free DNA; P<.001) and lower fetal fraction (median, 7.5% vs 9.4%; P<.001). The expected, cross-validated area under the curve scores for early-onset, preterm, and term preeclampsia were 0.782, 0.801, and 0.712, respectively, for the logistic regression model, and 0.797, 0.800, and 0.713, respectively, for the neural network model. At a screen-positive rate of 15%, sensitivity for preterm preeclampsia was 58.4% (95% confidence interval, 0.569–0.599) for the logistic regression model and 59.3% (95% confidence interval, 0.578–0.608) for the neural network model. The contribution of both total cell-free DNA and fetal fraction to the prediction of term and preterm preeclampsia was negligible. For early-onset preeclampsia, removal of the total cell-free DNA and fetal fraction features from the neural network model was associated with a 6.9% decrease in sensitivity at a 15% screen-positive rate, from 54.9% (95% confidence interval, 52.9–56.9) to 48.0% (95% confidence interval, 45.0–51.0). Conclusion: Routinely available patient characteristics and cell-free DNA markers can be used to predict preeclampsia with performance comparable to that of other patient characteristic models for the prediction of preterm preeclampsia. Logistic regression and neural network models showed similar performance.
KW - cell-free DNA
KW - early-onset preeclampsia
KW - fetal fraction
KW - linear neural network
KW - noninvasive prenatal screening
KW - nonlinear neural network
KW - pregnancy
KW - preterm preeclampsia
KW - term preeclampsia
UR - http://www.scopus.com/inward/record.url?scp=85199054726&partnerID=8YFLogxK
U2 - 10.1016/j.ajog.2024.02.299
DO - 10.1016/j.ajog.2024.02.299
M3 - Article
C2 - 38432413
AN - SCOPUS:85199054726
SN - 0002-9378
VL - 231
SP - 554.e1-554.e18
JO - American Journal of Obstetrics and Gynecology
JF - American Journal of Obstetrics and Gynecology
IS - 5
ER -