AUTHOR=Zhang Rui , Guo Yuanbing , Zhai Xiaonan , Wang Juan , Hao Xiaoyan , Yang Liu , Zhou Lei , Gao Jiawei , Liu Jiayun 

TITLE=Machine learning algorithm based on combined clinical indicators for the prediction of infertility and pregnancy loss

JOURNAL=Frontiers in Endocrinology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2025.1544724

DOI=10.3389/fendo.2025.1544724

ISSN=1664-2392

ABSTRACT=Background and objectivesDiagnosis and treatment of infertility and pregnancy loss are complicated by various factors. We aimed to develop a simpler, more efficient system for diagnosing infertility and pregnancy loss.MethodsThis study included 333 female patients with infertility and 319 female patients with pregnancy loss, as well as 327 healthy individuals for modeling; 1264 female patients with infertility and 1030 female patients with pregnancy loss, as well as 1059 healthy individuals for validating the models. The average age and basic information were matched between the groups. Three methods were used for screening 100+ clinical indicators, and five machine learning algorithms were used to develop and evaluate diagnostic models based on the most relevant indicators.ResultsMultivariate analysis revealed significant differences in several factors between the patients and the control group. 25-hydroxy vitamin D3 (25OHVD3) was the factor exhibiting the most prominent difference, and most patients presented deficiency in the levels of this vitamin. 25OHVD3 is associated with blood lipids, hormones, thyroid function, human papillomavirus infection, hepatitis B infection, sedimentation rate, renal function, coagulation function, and amino acids in patients with infertility. The model for infertility diagnosis included eleven factors and exhibited area under the curve (AUC), sensitivity, and specificity values higher than 0.958, 86.52%, and 91.23%, respectively. The model for potential pregnancy loss was also developed using five machine learning algorithms and was based on 7 indicators. According to the results obtained from the testing set, the sensitivity was higher than 92.02%, the specificity was higher than 95.18%, the accuracy was higher than 94.34%, and the AUC was higher than 0.972.ConclusionThe simplicity, good diagnostic performance, and high sensitivity of the models presented here may facilitate early detection, treatment, and prevention of infertility and pregnancy loss.