399
ResearchBib IF - 11.01, ISSN: 3030-3753, Volume 2 Issue 11
УДК:618.3-07:004.8
MODELING THE RISK OF PREECLAMPSIA DEVELOPMENT USING
ARTIFICIAL INTELLIGENCE METHODS.
Нематова Маржона Зикриллаевна
Ассистент кафедры акушерства и гинекологии №1
Бухарский государственный медицинский институт, Узбекистан, Бухара.
https://orcid.org/0009-0000-4105-1064
https://doi.org/10.5281/zenodo.17612067
Summary
. Preeclampsia is a multifactorial obstetric complication that occurs in the
second half of pregnancy and is characterized by hypertension and dysfunction of various
organs, significantly increasing the risk of maternal and perinatal morbidity and mortality.
Despite considerable progress in understanding the pathophysiological mechanisms,
effective methods for early detection and prediction of preeclampsia remain a pressing issue in
clinical practice. This article presents an overview of modern approaches to predicting
preeclampsia using artificial intelligence (AI) technologies. Machine learning algorithms,
including random forests, gradient boosting, and deep neural networks, are discussed in the
context of analyzing clinical, biochemical, and ultrasound data of pregnant women. Particular
attention is given to the integration of multimodal data to improve prediction accuracy.
Keywords:
preeclampsia, artificial intelligence, hypertension, pathophysiological
mechanisms.
Currently, there are three most commonly used preeclampsia risk assessment systems
(RAS):
1.
the NICE risk assessment system (National Institute for Health and Clinical Excellence,
UK);
2.
the ACOG risk assessment system (American College of Obstetricians and
Gynecologists, USA);
3.
the FMF risk assessment system (Fetal Medicine Foundation, UK).
The NICE RAS identifies moderate- and high-risk groups based on the following factors:
age ≥ 40 years; BMI ≥ 35 kg/m²; nulliparity; family history of preeclampsia; interpregnancy
interval > 10 years; hypertensive disorders in previous pregnancies; chronic hypertension;
chronic kidney disease; diabetes mellitus; and autoimmune diseases. [1].
The ACOG RAS identifies moderate- and high-risk groups based on the following
factors: age ≥ 35 years; BMI > 30 kg/m²; first pregnancy; family history of preeclampsia;
chronic hypertension; kidney disease; autoimmune diseases; and type 1 or type 2 diabetes
mellitus. [2].
According to FIGO recommendations, the prediction of preeclampsia risk should be
based on the patient’s medical history and risk factors identified through first-trimester prenatal
screening. [3].
In a study by N. O’Gorman et al., the predictive accuracy of the NICE and ACOG [4].
risk assessment systems was evaluated. The detection rates of preterm and term preeclampsia
using the NICE scale were 39% and 34%, respectively. Corresponding detection rates using the
ACOG scale were 90% and 89%, with a false-positive rate of 64.3%.
Predicting preeclampsia using artificial intelligence algorithms
400
ResearchBib IF - 11.01, ISSN: 3030-3753, Volume 2 Issue 11
Due to the insufficient effectiveness of current preeclampsia prediction methods proposed
by major international obstetrics and gynecology organizations—NICE and ACOG [4]. —
research laboratories worldwide are continuing to search for reliable methods of assessing
preeclampsia risk based on biophysical factors, biochemical markers, and AI algorithms. [5].
This approach aims not only to improve predictive accuracy but also to provide an
individualized risk estimate.
Currently, the list of FIGO-recommended parameters for first-trimester screening in
singleton pregnancies includes:
maternal characteristics, medical history, and comorbidities;
placental growth factor (PlGF), secreted by trophoblast cells;
pregnancy-associated plasma protein-A (PAPP-A), secreted by the syncytiotrophoblast;
uterine artery pulsatility index (PI);
mean arterial pressure (MAP).
One of the first attempts to combine the analysis of biochemical markers with machine
learning algorithms to develop a predictive model for preeclampsia (PE) was made by L.C.
Kenny et al. [6]. The researchers used genetic programming to identify patterns in the changes of
plasma metabolite levels in patients with PE. The analysis included 87 plasma samples from
pregnant women with PE and 87 control samples from healthy pregnant women. Gas
chromatography was performed, and the resulting data were processed using genetic
programming algorithms. As a result, a model was developed with sensitivity and specificity of
100% and 98%, respectively.
Various combinations of maternal history data, instrumental studies, or laboratory
markers can be used to generate predictions. Each research team independently determines
which features to include in the model. In their study, S.M. van Kuijk et al. limited the number of
variables used to predict recurrent early-onset PE to five: [7]. div mass index (BMI),
gestational age at previous delivery, fasting blood glucose level, presence of hypertension, and
birth weight of the previous baby being small for gestational age. According to the authors, this
limited set of features helps prevent model overfitting due to the small dataset. Their database
included 407 pregnant women with a history of early-onset PE leading to preterm delivery. After
statistical processing, a predictive model was developed that can identify women at low risk of
recurrent PE using the five features listed above. The model achieved an AUC of 0.65. Given the
small sample size, the authors emphasized the need for external validation on a larger dataset.
R.M. Villa et al. conducted a study to identify factors that influence both the risk and
severity of PE [8]. Their database consisted of 903 cases of pregnant women with known PE risk
factors. The researchers used a Bayesian algorithm for cluster analysis, classifying patients into
groups based on observed combinations of risk factors. A total of 25 risk factor combinations
were identified. For each group, the risk of different PE subtypes was calculated, and a heatmap
was created for visualization. The study found that the risk of developing PE increases
exponentially with the number of risk factors. It was also noted that the risk profiles of women
with severe PE and preterm delivery differ from those with term pregnancies.
Two subsequent studies on PE prediction were based on neural network technology. E.
Tejera et al. proposed using heart rate variability (HRV) indices to classify patients into groups
with normal blood pressure, hypertension, and PE [9]. A total of 568 short ECG recordings were
collected from 217 pregnant women, with varying gestational ages.
401
ResearchBib IF - 11.01, ISSN: 3030-3753, Volume 2 Issue 11
Maternal history and blood pressure data were also included. The best performance was
achieved by a predictive model based on an artificial neural network, with an AUC of 0.95.
S.K. Neocleous et al. employed a multi-slab neural network structure [10]. Their database
included 6,838 birth records. The model used 15 input features, including mean arterial pressure
(MAP), uterine artery pulsatility index (PI), racial and ethnic background, and more. The
resulting model showed high predictive accuracy: 83.6% on the training set and 93.8% on the
test set.
Conclusion
Every year, approximately half a million women around the world die from causes
related to hypertensive disorders of pregnancy (HDP), and about 12% of maternal deaths result
from severe complications associated with these conditions. Despite the availability of effective
preventive measures, preeclampsia (PE) and eclampsia remain among the leading causes of
maternal morbidity and mortality, primarily due to the lack of accurate risk assessment and
prediction methods.
Therefore, the implementation of predictive analytics and clinical decision support
systems in obstetrics and gynecology—based on artificial intelligence algorithms and the
development of highly accurate, iterative predictive models that continuously refine risk
assessments as new diagnostic data become available—has the potential to build a consistent
chain of effective preventive measures, starting from the preconception stage through the second
trimester. This approach could significantly reduce the incidence of PE and other pregnancy
complications related to HDP, positively impacting both reproductive health and the
demographic situation in the Republic of Uzbekistan.
References:
1.
National Collaborating Centre for Women's and Children's Health (UK).
2.
Hypertension in Pregnancy: The Management of Hypertensive Disorders During
Pregnancy. London: RCOG Press, 2010.
3.
LeFevre M.L., U.S. Preventive Services Task Force. Low-dose aspirin use for the
prevention of morbidity and mortality from preeclampsia: U.S.
4.
Preventive Services Task Force recommendation statement. Ann Intern Med.
2014;161(11):819–26. https://doi.org/10.7326/M14-1884.
5.
Poon L.C., Shennan A., Hyett J.A. et al. The International Federation of Gynecology and
Obstetrics (FIGO) initiative on pre-eclampsia: A pragmatic guide for first-trimester
screening and prevention [published correction appears in Int J Gynaecol Obstet.
2019;146(3):390–1].
Int
J
Gynaecol
Obstet.
2019;145(Suppl
1):1–33.
https://doi.org/10.1002/ijgo.12802.
6.
O’Gorman N., Wright D., Poon L.C. et al. Multicenter screening for pre-eclampsia by
maternal factors and biomarkers at 11-13 weeks’ gestation: comparison with NICE
guidelines and ACOG recommendations.
7.
Ultrasound Obstet Gynecol. 2017;49(6):756–60.
8.
Kleinrouweler C.E., Cheong-See F.M., Collins G.S. et al. Prognostic models in
obstetrics: available, but far from applicable. Am J Obstet
9.
Gynecol. 2016;214(1):79–90.e36.
. ajog.2015.06.013. 6. Kenny
L.C., Dunn W.B., Ellis D.I. et al. Novel biomarkers for pre-eclampsia detected using
metabolomics and machine learning. Metabolomics.
402
ResearchBib IF - 11.01, ISSN: 3030-3753, Volume 2 Issue 11
10.
2005;1(3):227–34.
https://doi.org/10.1007/s11306-005-0003-1.
van
Kuijk
S.M.,
Delahaije D.H., Dirksen C.D. et al. External validation of a model for periconceptional
prediction of recurrent early-onset preeclampsia. Hypertens Pregnancy. 2014;33(3):265–
76. https://doi.org/ 10.3109/10641955.013.872253.
11.
Villa P.M., Marttinen P., Gillberg J. et al. Cluster analysis to estimate the risk of
preeclampsia in the high-risk Prediction and Prevention of
12.
Preeclampsia and Intrauterine Growth Restriction (PREDO) study. PLoS One.
2017;12(3):e0174399.
https://doi.org/10.1371/journal
. pone.0174399.
13.
Tejera E., Areias J.M., Rodrigues A. et al. Artificial neural network for normal,
hypertensive, and preeclamptic pregnancy classification using maternal heart rate
variability indexes. J Matern Fetal Neonatal Med.
14.
2011;24(9):1147–51. https://doi.org/10.3109/14767058.2010.545916.
15.
Neocleous C.K., Anastasopoulos P., Nikolaides K.H. et al. Neural networks to estimate
the risk for preeclampsia occurrence. International Joint
16.
Conference on Neural Networks. Atlanta, Georgia: USA. 14–19 June 2009.
17.
2221–5. https://doi.org/10.1109/IJCNN.2009.5178820.
18.
Marić I., Tsur A., Aghaeepour N. et al. Early prediction of preeclampsia via machine
learning. Am J Obstet Gynecol MFM. 2020;2(2):100100.
19.
https://doi.org/10.1016/j.ajogmf.2020.100100.
