Machine learning model for prediction of palliative care phases in patients with advanced cancer: a retrospective study.

Abstract

Background: Developing an accurate predictive model for palliative care phases is crucial for improving cancer patient management, enabling healthcare providers to identify those in need of specific care plans and streamlining decision-making process for patients and caregivers. This study aims to identify symptom and functional indicators from Palliative Care Outcomes Collaboration (PCOC) data and develop a predictive model capable of accurately categorizing palliative care phases in advanced cancer patients.

Methods: A retrospective cohort study design was adopted in this study. Data on PCOC information were collected and analyzed from patients admitted to a palliative care unit at a cancer hospital in China between April 2023 and December 2024. The Gradient Boosting Decision Tree in the machine learning algorithm to establish a palliative care phase prediction model and evaluated the prediction performance of this model.

Results: A total of 9,787 assessments from 793 patients were included in the analysis of this study. Significant differences were identified among the four PCOC phases of care in terms of the symptom distress, palliative care problem severity, functional status and daily living activities. The machine learning model developed in this study achieved areas under the curve (AUCs) of 0.997, 0.996, 0.999, and 0.999 for predicting the stable, unstable, deteriorating, and terminal phases in the training group, respectively. In the testing group, the corresponding AUCs were 0.976, 0.965, 0.971, and 0.998.

Conclusions: The prediction model developed in this study based on the machine learning algorithm showed good performance, offering significant potential for facilitating timely interventions, enhancing symptom management, and optimizing palliative care resource allocation in advanced cancer patients in mainland China.