Complementary frailty and mortality prediction models on older patients as a tool for assessing palliative care needs

Introduction Palliative care (PC) has demonstrated benefits for life-limiting illnesses. Nowadays, there is a growing consensus about giving access these care services to non-cancer older patients. Bad survival prognosis and patients’ decline are working criterions to guide PC decision making. Objective The main aim of this work is to propose complementary models based on machine learning approaches to predict frailty and mortality in older patients in the context of supporting PC decision making. Methods The dataset used in this study is composed by 39,310 hospital admissions for 19,753 older patients (age >= 65) from January 1st, 2011 to December 30th, 2018. Predictive models based on Gradient Boosting Machines and Deep Neural Networks were implemented for binary one-year mortality classification, survival estimation and binary one-year frailty classification. Besides, we tested the similarity between mortality and frailty distributions. Results The one-year mortality classifier achieved an AUC ROC of 0.87 [0.86, 0.87]; whereas the mortality regression model achieved an MAE of 333.13 [323.10, 342.49] days. Moreover, the one-year frailty classifier obtained an AUC ROC of 0.89 [0.88, 0.90]. Conclusions The performance of our one-year mortality model is competitive with the current state-of-the-art. Besides, to our knowledge, this is the first study predicting one-year frailty status based on a frailty index. We found mortality and frailty criteria are weakly correlated and have different distributions; therefore, we interpreted them as complementary assessment measurements for palliative care decision making. Predictive models are accessible as an online tool at http://demoiapc.upv.es . The models presented here may be part of decision support systems for care services in non-cancer older patients after their external validation.