BACKGROUND
Accurate prediction of survival in patients with advanced cancer is essential for optimizing end-of-life care and allocating palliative resources. However, fragmented clinical data and limited prognostic tools make reliable survival estimation challenging in real-world clinical practice.
MATERIAL AND METHODS
This retrospective study analyzed data from 842 patients with advanced cancer admitted to the palliative care department of a tertiary hospital in China between 2018 and 2020. Multiple clinical and laboratory indicators were collected, including electrolyte levels, hematological parameters, and biochemical markers. Survival prediction models were developed using traditional Cox regression, LASSO-Cox regression, and forward likelihood ratio (Forward-LR) Cox regression, with 80% of the dataset used for training and 20% for validation. Model performance was evaluated using receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) values for predicting survival at 1 to 4 weeks.
RESULTS
The median survival time of the cohort was 9.7 days, with most patients dying within 4 weeks. Age significantly influenced survival risk (HR=1.010). Several clinical indicators were independently associated with mortality, including potassium, chloride, iron, magnesium, platelet count, and inflammatory markers. Elevated potassium and iron levels increased mortality risk, whereas higher chloride and platelet levels showed protective effects. In model comparisons, LASSO-Cox demonstrated better performance in short-term survival prediction (1-2 weeks), while Forward-LR Cox regression showed superior accuracy for longer-term prediction (3-4 weeks).
CONCLUSIONS
A multi-model prognostic framework based on routine clinical indicators can effectively predict short-term survival in advanced cancer patients. These models may support clinical decision-making in palliative care and improve end-of-life resource allocation pending external validation.

Keywords: Survival; Prognosis; Cancer Care Facilities