Machine learning models to predict 6-month mortality risk in home-based hospice patients with advanced cancer

Abstract

Objective

This study aimed to construct predictive models using five different machine learning algorithms for predicting 6-month mortality risk among home-based hospice patients with advanced cancer.

Methods

This population-based retrospective prognostic study examined data from 7023 patients in a home-based hospice center. Various algorithms including logistic regression, random forest, XGBoost, support vector machine, and neural network were implemented in this study. The model performance and effectiveness were assessed using sensitivity, specificity, accuracy, the area under the curve (AUC), and F1 Score. Additionally, a nomogram was developed to calculate individualized risk probabilities, enhancing clinical utility.

Results

Among the five types of predictive models, the logistic regression model achieved an AUC of 0.754 (95% CI: 0.721–0.786) in the test dataset, outperforming other machine learning algorithms. The nomogram developed from the logistic regression model included 10 independent risk factors for 6-month mortality. The Hosmer–Lemeshow test showed no significant difference between the predicted and observed outcomes (training set: 12.646, P ​= ​0.13; testing set: 3.807, P ​= ​0.87). Clinical decision curve analysis indicated that the model provided substantial net benefits across a wide range of thresholds.

Conclusions

Our study demonstrated that routinely collected healthcare data on the first home visit have the potential to help screen high-risk patients, which may provide evidence for targeted hospice care.