Lower urinary tract symptoms (LUTS) in women affect patients’ quality of life and are highly prevalent globally. Our previous study reported an increased subsequent risk of hospitalization for acute cardiovascular disease in the LUTS group.  Therefore, LUTS may be a precursor condition predisposing patients to acute cardiovascular disease. Recently, artificial intelligence (AI)-supervised machine learning-based techniques have been applied in medicine and healthcare systems to improve care allocation and disease risk prediction. This study aimed to develop an AI-based prediction model for women with LUTS.

This retrospective study collected data from the Medical Center hospital information system (HIS) database. We included on patients aged 20 years or older who had at least three out-patient visits or one hospitalization with symptom-related International Classification of Diseases, Ninth Revision, Clinical Modification/International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-9-CM)/(ICD-10-CM) codes between January 1, 2001 and December, 31, 2018. Data was accessed for research purposes after IRB approval (April 27, 2023). Study outcomes included acute coronary syndrome and stroke. Based on the time interval between LUTS diagnosis and cardiovascular disease happened, the patients were divided into three groups, 3 years, 5 years and 8 years.
    We selected 20 features based on their clinical importance and expert opinion. Data from the HIS dataset were used as inputs to the machine learning algorithms to select significant factors. Features with cases number ≤10 or low correlation coefficients with the outcome were excluded. We also performed SHapley Additive exPlanations (SHAP) feature importance summary plots to estimate the significance of each feature within a model, which based on mean SHAP value. It depicted the impact for model output magnitude (Fig. 1).
    We used six machine learning algorithms, consisting of (1) Logistic regression, (2) Random forest, (3) Support vector machine (SVM), (4) Light Gradient Boosting Machine (LightGBM), (5) Multilayer perceptron (MLP), and (6) eXtreme Gradient Boosting (XGBoost). The performance of the models were evaluated based on accuracy, sensitivity, specificity, area under the receiver operating characteristic curve (AUC), and receiver operating characteristic (ROC) curve. The model with the highest AUC value was used as a future clinical risk assessment tool.

We included 1,799 women with LUTS and Table 1 shows the correlations between the 8 features and outcomes. The most relevant 5 features, including age, hypertension, hyperlipidemia, HbA1c and DM, were selected for modeling. In addition, based on their clinical importance and previous study results, SBP, DBP and creatinine were included in the model because of their high association with cardiovascular risk.
    In the machine learning prediction of cardiovasculardisease within 3 years, the sensitivity of logistic regression, SVM and MLP reached 0.765. The MLP had the highest specificity (0.765), followed by logistic regression (0.750). Half of the models (Logistic regression, SVM and MLP), achieved high AUC values above 0.750. 
    We designed a web-based prediction application using our best model (the MLP model) and integrated it into an existing HIS. The predictive models were built in Python and the web service application was implemented in Microsoft Visual Studio. NET 2017. As shown in Fig. 2, the AI predicts that patient tended to experience acute cardiovascular disease as the risk value exceeds 50%.

In this retrospective study, we used medical big data and successfully established an AI-based prediction model to evaluate the subsequent risks of acute cardiovascular disease among women with LUTS. To the best of our knowledge, this is the first study to use AI techniques to predict the risk of acute cardiovascular disease for women with LUTS. The accuracy was approximately 80% in predicting of acute cardiovascular disease within 3 years, highlighting its importance for physicians caring of women with LUTS the possibilities of acute cardiovascular disease. 
    LUTS share several risk factors with cardiovascular diseases, such as obesity, hypertension, DM, hyperlipidemia, and nicotine use. Some mechanisms have been proposed to explain the association between LUTS and cardiovascular disease. First, LUTS shares some common etiology with cardiovascular diseases, such as endothelial dysfunction. Second, LUTS may be a precursor condition that predisposes patients to develop cardiovascular disease.

Our AI-based prediction model had a significantly high AUC (0.803) in prediction of cardiovascular disease within 3 years, which can assist physicians in related medical fields in performing cardiovascular disease risk assessments for women with LUTS. Researchers can include more potential variables, such as CRP and albumin, for follow-up study to improve the quality of the models.