Big Data/Machine Learning/AI
Machine learning for predicting hospital mortality in cervical cancer patients Audêncio Victor* Audêncio Victor Fabiano Barcellos Filho Sancho Xavier Pedro Alexandre Dias Porto Chiavegatto Filho
Introduction: Cervical cancer (CC) is a leading cause of cancer mortality among women in low- and middle-income countries, especially in regions with limited healthcare infrastructure. This study utilizes machine learning (ML) models to predict hospital mortality in patients hospitalized with CC from 2011 to 2023 in Mato Grosso, Brazil.
Methods: Hospital data on cervical cancer patients were obtained from the Hospital Information System (SIH) of the SUS, covering 2011–2023. Five ML algorithms Logistic Regression, Random Forest, CatBoost, LightGBM, and XGBoost were used to predict hospital mortality based on demographic, clinical, and hospital variables. Performance metrics included AUC-ROC, accuracy, sensitivity, specificity, precision, F1, and Matthew’s correlation coefficient (MCC).
Results: All models achieved AUC-ROC above 0.87, indicating high predictive performance. XGBoost demonstrated the best overall performance, with an AUC-ROC of 0.89, 88% accuracy, and 95% specificity, with an MCC of 0.28. Tree-based models such as Random Forest and XGBoost showed high specificity with 94% and 95%, respectively, while CatBoost also performed well with an accuracy of 88% and specificity of 94%. SHAP value analysis identified medical procedure type, hospitalization cost, and service complexity as the main mortality predictors.
Conclusion: ML was found effective for predicting hospital mortality in CC patients in Mato Grosso, providing valuable insights into risk factors and facilitating clinical decision-making. These models can be integrated into healthcare systems to optimize resource allocation and improve survival in areas with limited infrastructure.
Keywords: Cervical cancer, hospital mortality, machine learning, oncology, Brazil
