JMIR Publications

ABSTRACT

Background:

Preeclampsia is a severe hypertensive disorder that typically manifests after 20 weeks of gestation, characterized by hypertension and proteinuria. The global incidence of preeclampsia ranges from approximately 3% to 9%. In China, the prevalence has notably increased from 5.79% in 2005 to 9.5% in 2019, presenting a significant public health challenge. Given its complex etiology, traditional statistical methods face limitations in processing relevant data, whereas machine learning technology has demonstrated considerable potential in predicting preeclampsia.

Objective:

This study aims to evaluate the performance of machine learning models in predicting preeclampsia through a systematic review and meta-analysis, while also exploring their potential clinical application value, in order to specifically enhance the quality of future research and the predictive capability of the models.

Methods:

Following the PRISMA guidelines, an extensive literature review was performed using multiple databases, such as PubMed, Web of Science, IEEE Xplore, and CNKI. Our search yielded 27 studies that encompassed a total of 32 machine learning prediction models. The researchers systematically extracted data regarding the characteristics and performance metrics of each model and evaluated the risk of bias and applicability utilizing the PROBAST tool. The pooled estimates of the models were calculated using Meta-DiSc software, and analyses, including subgroup analysis and meta-regression, were conducted to explore the sources of heterogeneity.

Results:

The total sample size of the included studies varied significantly, and the number of predictors incorporated into the models differed among them. Twenty-two studies performed internal validation, while five undertook external validation. The overall pooled area under the receiver operating characteristic curve of the machine learning models was 0.9202, indicating an excellent discriminative ability. Sensitivity analysis demonstrated that the studies were not influenced by extreme values. Analysis of subgroups indicated that the models demonstrated enhanced predictive performance with smaller sample sizes, showed improvement when employing laboratory test indicators, and that neural network models surpassed conventional machine learning techniques. Meta-regression analysis indicated that the source of heterogeneity among studies was related to the design of the studies.

Conclusions:

Machine learning models have demonstrated remarkable performance in predicting preeclampsia, indicating their potential for clinical application. However, substantial heterogeneity exists among the studies, and most models lack external validation. Future efforts should concentrate on conducting high-quality, multicenter studies to improve the reliability and generalizability of these models. Clinical Trial: CRD420251005830