AUTHOR=Zhang Xiao , Fei Ningbo , Zhang Xinxin , Wang Qun , Fang Zongping TITLE=Machine Learning Prediction Models for Postoperative Stroke in Elderly Patients: Analyses of the MIMIC Database JOURNAL=Frontiers in Aging Neuroscience VOLUME=Volume 14 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/aging-neuroscience/articles/10.3389/fnagi.2022.897611 DOI=10.3389/fnagi.2022.897611 ISSN=1663-4365 ABSTRACT=Abstract Objective With the aging population growing up and high prevalence, post operative stroke becomes more and more concerned. Thus, this study aimed to establish prediction model and assess the risk factors of stroke in elderly patients during the postoperative period. Methods Machine Learning Prediction Models was conducted on the elderly patients from MIMIC-Ⅵand MIMIC-Ⅲ database. The SMOTENC balancing technique and iterative SVD data imputation method was used to deal with problem of category imbalance and missing values respectively. We analyzed the possible predictive factors of the stroke in the elderly patients, using 7 modeling approaches to train model. The ROC curve (receiver operating characteristic curve) methods were conducted to evaluate the diagnostic value of the model documented from machine learning. Results We analyzed 7129 and 661 cases of patients with MIMIC-Ⅵ and MIMIC-Ⅲ respectively, the XGB model had the highest AUC (area under curve) 0.78, outperforming the other six models as well as the XGB model without data balancing. The top 5 predictor variables included hypertension, cancer, congestive heart failure, chronic pulmonary disease and peripheral vascular disease. Conclusion The stroke in elderly patients during the postoperative period can be correctly predicted. To prevent post-operative stroke in elderly patients, the history of hypertension should be taken more seriously than other results of laboratory tests.