KEY POINTS
- This secondary analysis of the Palliative Radiotherapy and Inflammation Study included 573 adults with metastatic cancer undergoing palliative radiotherapy for bone metastases.
- Overall survival was classified into 4 groups: ≤ 3 weeks, 3–24 weeks, 24–52 weeks, and ≥ 52 weeks after treatment.
- Candidate predictors included 65 clinical, tumor-related, dosimetric, treatment-related, and laboratory variables; after feature reduction, 8 variables were used in decision tree modeling.
- Interleukin-8, lymphocyte count, and hemoglobin were the top predictors and accounted for more than 80% of model contribution in SHapley Additive exPlanations analysis.
- The Gini-based model showed best discrimination for very short and long survival: area under the curve 0.80 for survival ≤ 3 weeks and 0.78 for survival ≥ 52 weeks, but performance for 24–52 weeks was close to random.
CLINICAL TAKEAWAY
This model is relevant to palliative radiotherapy decision-making because it targets the hardest practical question: whether expected survival supports radiotherapy or favors best supportive care. The key signal is that objective laboratory markers, especially interleukin-8, lymphocyte count, and hemoglobin, carried more prognostic weight than many traditional clinical variables. But without external validation and with weak discrimination for intermediate survival, this remains technically relevant prognostic research, not a tool for routine treatment selection.