Key points are not available for this paper at this time.
Purpose: This study aimed to develop and validate a CT-based radiomics model for predicting pain relief after palliative radiotherapy in patients with bone metastases, and to compare the performance of 11 machine learning algorithms. Methods: We retrospectively enrolled patients with bone metastases who received palliative radiotherapy at Xuzhou Central Hospital (Center 1) and Xuzhou First People's Hospital (Center 2) between January 2022 and December 2024. All patients completed a prescribed dose of 40 Gy in 20 fractions or 30 Gy in 10 fractions. Clinical variables-including age, sex, primary tumor type, pattern of bone destruction, and metastatic site-were collected alongside CT images. Pain response was assessed per the International Consensus on Endpoints for Palliative Radiotherapy in Bone Metastases: complete response (CR) and partial response (PR) were grouped as the relief group, while progressive disease (PD) and stable disease (SD) constituted the non-relief group. ROIs were delineated over the tumor areas on bone-window CT images, and radiomic features were extracted, normalized, and screened to construct a radiomics signature. Eleven machine learning classifiers were trained and compared; the optimal model was selected for predictive performance evaluation and clinical applicability analysis. Results: A total of 134 eligible patients were included (pain relief group: n = 53; non-relief group: n = 81). Center 1 patients were randomly split approximately 8:2 into training (n = 91) and internal validation (n = 26) sets; Center 2 served as the external test set (n = 17). No significant differences existed between the two centers in baseline demographics, tumor-related variables, or treatment parameters, except for bone-protective drug use and bone metastasis site. After feature selection, 7 radiomic features remained for modeling. Among 11 tested machine learning models, the k-nearest neighbors (KNN) model demonstrated the best performance: area under the receiver operating characteristic curve (AUC) was 0.823 (95% confidence interval (CI): 0.743-0.903) in the training set, 0.812 (95% CI: 0.661-0.964) in the internal validation set, and 0.818 (95% CI: 0.556-1.000) in the external test set. Decision curve analysis (DCA) indicated favorable net clinical benefit. Conclusion: The KNN model based on CT radiomics can effectively predict pain relief outcomes after palliative radiotherapy in patients with bone metastases, showing potential clinical utility, and may help identify patients likely to achieve pain relief from radiotherapy.
Wan et al. (Tue,) studied this question.