Domínguez Del Olmo Paula, Herraiz Ignacio, Villalaín Cecilia, Galindo Alberto, Ayala Jose Luis
Department of Computer Architecture and Automation, Faculty of Informatics, Complutense University of Madrid, Spain.
Maternal and Child Health and Development Research Network (RICORS-SAMID Network), Madrid, Spain.
Womens Health (Lond). 2025 Jan-Dec;21:17455057251348978. doi: 10.1177/17455057251348978. Epub 2025 Jul 21.
Preeclampsia, a complex and unpredictable pregnancy complication, poses significant challenges in predicting maternal outcomes, often leading to heightened anxiety among patients, families, and clinicians. This study introduces an innovative approach to enhance the prediction of complications in early-onset preeclampsia, leveraging advanced machine learning techniques inspired by bio-algorithms.
Our goal is to enhance the clinical management of preeclampsia by improving risk stratification and offering a more personalized approach to patient care.
A single-center, observational, retrospective cohort study with 246 singleton pregnancies diagnosed with early-onset preeclampsia between January 2007 and December 2020 was conducted at 12 de Octubre Hospital. Exclusions included pregnancies with congenital anomalies, lack of angiogenesis biomarker determination or loss of follow-up, resulting in a cohort of 234 patients.
We employed innovative genetic algorithm strategies, integrating two distinct supervised machine learning models. These aim to accurately forecast key maternal risks associated with preeclampsia and determine the optimal timing for delivery. This approach culminates in a unique ensemble framework, comprising a primary model for assessing the risk of adverse outcomes and two specialized sub-models focusing on Hemolysis, Elevated Liver enzymes, and Low Platelets-abruption and temporal factors.
Our findings are promising. The mono-objective genetic algorithm strategy yielded predictive -scores of 68.3%, 83.1% ± 7.2%, and 71.5% ± 3.5% in the "Risk of Adverse Outcomes," "Hemolysis, Elevated Liver enzymes, and Low Platelets-Abruption," and "Time to Delivery" models, respectively. The multi-objective strategy, utilizing minimal yet powerful variable combinations, achieved predictive accuracies of 61.5%, 80.0% ± 6.2%, and 69.3% ± 7.2% with just five, four, and six features in the respective models. These results highlight the potential of our approach in enhancing clinical decision-making.
This study introduces a novel approach to risk stratification in early-onset preeclampsia, integrating baseline and delivery data within a machine learning framework. Our results demonstrate that refined risk prediction with a minimal number of variables can complement existing clinical tools. Further validation in larger cohorts is needed to confirm its potential impact on decision-making and maternal outcomes.
子痫前期是一种复杂且难以预测的妊娠并发症，在预测孕产妇结局方面存在重大挑战，常常导致患者、家庭和临床医生的焦虑加剧。本研究引入了一种创新方法，利用受生物算法启发的先进机器学习技术，以加强早发型子痫前期并发症的预测。
我们的目标是通过改善风险分层并为患者护理提供更个性化的方法，来加强子痫前期的临床管理。
在12月10日医院进行了一项单中心、观察性、回顾性队列研究，纳入了2007年1月至2020年12月期间诊断为早发型子痫前期的246例单胎妊娠。排除标准包括伴有先天性异常的妊娠、未测定血管生成生物标志物或失访的情况，最终得到234例患者的队列。
我们采用了创新的遗传算法策略，整合了两种不同的监督机器学习模型。这些模型旨在准确预测与子痫前期相关的关键孕产妇风险，并确定最佳分娩时机。这种方法最终形成了一个独特的集成框架，包括一个用于评估不良结局风险的主模型以及两个专注于溶血、肝酶升高和血小板减少 - 胎盘早剥及时间因素的专门子模型。
我们的研究结果很有前景。单目标遗传算法策略在“不良结局风险”“溶血、肝酶升高和血小板减少 - 胎盘早剥”以及“分娩时机”模型中的预测得分分别为68.3%、83.1%± 7.2%和71.5%± 3.5%。多目标策略利用最少但强大的变量组合，在各自模型中仅使用五个、四个和六个特征时，预测准确率分别达到61.5%、80.0%± 6.2%和69.3%± 7.2%。这些结果凸显了我们的方法在加强临床决策方面的潜力。
本研究引入了一种早发型子痫前期风险分层的新方法，在机器学习框架内整合了基线和分娩数据。我们的结果表明，用最少数量的变量进行精确风险预测可以补充现有的临床工具。需要在更大的队列中进一步验证，以确认其对决策和孕产妇结局的潜在影响。
