Predictive Model for the Radiotherapy Induced Rib Fracture (RIRF) after Stereotactic Body Radiotherapy

Purpose/Objective(s) To identify clinical and dosimetric factors associated with the radiation-induced rib fracture (RIRF) after stereotactic body radiotherapy (SBRT), and to provide more appropriate ribs dose constraints. Materials/Methods Patients treated with SBRT for primary or metastatic lung tumors within 1 cm from Ribs, and had no previous overlapping radiation exposure were eligible to this study. Clinical factors, rib bone mineral density (BMD) and dosimetric parameters were evaluated for their associations with RIRFs. All cases were reviewed by experienced radiologists to detect and grade RIRF using the radiological as well as clinical toxicity data, as per CTCAE v5.0. Variable filtering was done through the least absolute shrinkage and selection operator (Lasso). A dose-event curve and predictive model were created. Calibration Curves were used to evaluate model prediction accuracy. Results A total of 160 patients with 165 lesions were eligible. After a median follow-up of 26.2 months (range: 10.5 – 62.6), 46 RIRFs were detected (27.9%) , including 20(42.6%) Grade 1 RIRFs, 25(53.2%) Grade 2 RIRFs and only 1 Grade 3 RIRF. The median interval from SBRT to RIRF detection was 16 months (range: 6.33-45.30). Of the 68 fractured ribs, 52 (76.5%) developed in ribs 2-5. 3 optimal dosimetric parameters and 1 optimal rib BMD parameter were filtered from 82 dosimetric parameters and 12 rib BMD parameters. As these three dosimetric parameters were cross-correlated, we selected Rib-BED3D1.4cc as the representative dosimetric predictor, who's cut-off value (α/β=3, LQ-model, Rib-BED3D1.4cc=212.16Gy) was extraordinarily higher than the AAPM Task Group 101 dose constrains for ribs (Rib-BED3D1cc<120.96Gy for 3 fractions, 116.67Gy for 5 fractions). Multivariate analysis showed that the rib location of the maximum dose point (Dmax Rib, odds ratio:21.491, p=0.018), mean density of the rib cortex (Cortex mean, odds ratio: 0.995, p = 0.018) and Rib-BED3D1.4cc (odds ratio: 1.013, p = 0.004) were all significantly associated with the RIRF. The final predictive logistic model for RIRFs included Dmax Rib, Cortex mean and Rib-BED3D1.4cc had significantly better predictive ability than any of the dosimetric predictor alone (AUC 0.791 vs 0.701, p = 0.0336). A nomogram was constructed based on the logistic model. Conclusion This study developed and validated a comprehensive model integrating rib BMD parameters with dosimetric factors to predict RIRFs. A predictive nomogram of AUC of 0.79 may provide an opportunity to guide clinicians. For the dosimetric predictor alone, we suggest that Rib-BED3D1.4cc less than 212.16Gy would be more appropriate.