Forecasting Future Unwell-being in Hospice Patients using Real-World Patient-Reported Outcomes

Background/aims: In palliative care, patient-reported outcome measures (PROMs) assess symptom severity and concerns. Prediction models can aid proactive care. We aim to identify which multidimensional symptoms and concerns (MDSC) forecast future unwell-being in hospice patients.
Methods: A prospective cohort study conducted in 15 hospices in the Netherlands. Patients were admitted between August 2015 and May 2023.
Primary Outcome: Future unwell-being.
Covariates: Demographics and multidimensional symptoms and concerns (MDSC) measured using the Utrecht Symptom Diary-four dimensional (USD-4D) collected twice weekly. Components included socio-spiritual concerns, psychological symptoms, physical symptoms, unwell-being, and a value of life measure. All items were rated on a 0-10 scale.
Outcome Creation: Lagging back the unwell-being measure from time ‘T’ to ‘T-1’.
Analysis:
•
Linear mixed-effect model (LME) constructed via backward selection.
•
Joint model created using LME combined with time-to-event analysis.
•
Bayesian approach employed with Markov Chain Monte Carlo settings set at 100 adaptations, 1000 iterations, and three chains.
Results: 3167 USD-4D were used of 739 patients, 55% women, mean age 74.
A combination of physical, psychological and socio-spiritual MDSC are identified to predict future unwell-being (Table 1).
Conclusions: Symptoms and concerns together predict future unwell-being of hospice patients. Joint models can support proactive hospice care, but require further implementation in statistical software.