Artificial intelligence models predicting abnormal uterine bleeding after COVID-19 vaccination

Abstract

The rapid deployment of COVID-19 vaccines has necessitated the ongoing surveillance of adverse events, with abnormal uterine bleeding (AUB) emerging as a reported concern in vaccinated females. We aimed to develop a machine learning (ML) model to predict post-vaccination AUB in women aged less than 50 years. A large-scale national cohort, the Korean Nationwide Cohort (K-COV-N cohort), was utilized, comprising over 7 million participants. The study employed advanced ML techniques, including ensemble models combining gradient boosting machine and logistic regression, and conducted feature importance analysis. The dataset was meticulously curated, focusing on relevant demographics and variables, and balanced using Synthetic Minority Over-sampling Technique. Using a national cohort of over 2 million COVID-19 vaccinated cases in South Korea, we developed a ML model for AUB prediction. Our study is the first to develop a predictive model for post-vaccination AUB, employing feature importance analysis to identify the key contributing factors. The analysis revealed three primary predictive features: COVID-19 vaccination frequency, NVX-CoV2373 (Novavax) COVID-19 vaccination count, and hemoglobin levels. These findings provide valuable insights into predicting the risk AUB following COVID-19 vaccination, potentially enhancing post-vaccination monitoring strategies.

Keywords: Abnormal uterine bleeding; COVID-19 vaccination; Ensemble models; Machine learning.

Fig. 1

Study population and data selection process from K-COV-N.

Fig. 2

Results of the ranked normalized feature importance from the ensemble model combining GBM and LR. Abbreviations: AUB, Abnormal uterine bleeding; COPD, chronic obstructive pulmonary disease; GBM, gradient boosting machine; LR, logistic regression T2DM, type 2 diabetes mellitus.