The multinomial mixed-effect regression model for predicting PCOC phases in hospice patients

Abstract

Purpose: The Palliative Care Outcomes Collaboration (PCOC) aims to enhance patient outcomes systematically. However, identifying crucial items and accurately determining PCOC phases remain challenging. This study aims to identify essential PCOC data items and construct a prediction model to accurately classify PCOC phases in terminal patients. Methods: A retrospective cohort study assessed PCOC data items across four PCOC phases: stable, unstable, deteriorating, and terminal. From July 2020 to March 2023, terminal patients were enrolled. A multinomial mixed-effect regression model was used for the analysis of multivariate PCOC repeated measurement data. Results: The dataset comprised 1933 terminally ill patients from 4 different hospice service settings. A total of 13,219 phases of care were analyzed. There were significant differences in the symptom assessment scale, palliative care problem severity score, Australia-modified Karnofsky performance status, and resource utilization groups-activities of daily living among the four PCOC phases of care. Clinical needs, including pain and other symptoms, declined from unstable to terminal phases, while psychological/spiritual and functional status for bed mobility, eating, and transfers increased. A robust prediction model achieved areas under the curves (AUCs) of 0.94, 0.94, 0.920, and 0.96 for stable, unstable, deteriorating, and terminal phases, respectively. Conclusions: Critical PCOC items distinguishing between PCOC phases were identified, enabling the development of an accurate prediction model. This model enhances hospice care quality by facilitating timely interventions and adjustments based on patients' PCOC phases.