Design of 1-year mortality forecast at hospital admission: A machine learning approach

Background: Palliative care is referred to a set of programs for patients that suffer life-limiting illnesses. These aim guarantee minimum level quality life (QoL) the last stage life. They are currently based on clinical evaluation risk one-year mortality. Objectives: The main objective this work develop and validate machine-learning models predict exitus patient within next year using data gathered at hospital admission. Methods: Five machine learning techniques were applied in our study predictive models: Support Vector Machines, K-neighbors Classifier, Gradient Boosting Random Forest Multilayer Perceptron. All trained evaluated retrospective dataset. was performed with five metrics computed by resampling strategy: Accuracy, area under ROC curve, Specificity, Sensitivity, Balanced Error Rate. Results: forecasting mortality achieved an AUC from 0.858 0.911. Specifically, Classifier best model, producing 0.911 (CI 95%, 0.912), sensitivity 0.856 0.86) specificity 0.807 0.806 0808) BER 0.168 0.167 0.169). Conclusions: analysis common information admission combined produced competitive discriminative power. Our reach results reported state art. demonstrate they can be used as accurate data-driven palliative criteria inclusion.