The age, lung volume, fractional dose, V 10 , and Hb levels were found to be independent factors for grade >2 RP by analyzing 186 patients with primary or metastatic lung tumor treated with SBRT. The age and Hb levels are novel predictors reported for the first time. A prediction model combining clinical, dosimetric factors, and laboratory biomarkers was proposed and achieved an AUC value up to 0.830, higher than models reported in the literature. Background: The study aims to identify the risk factors and develop a model for predicting grade >2 radiation pneumoni-tis (RP) for lung cancer patients treated with stereotactic body radiation therapy (SBRT). Materials and Methods: Clini-cal data, dosimetric data, and laboratory biomarkers from 186 patients treated with lung SBRT were collected. Univariate and multivariate logistic regression were performed to determine the predictive factors for grade >= 2 RP. Three models were developed by using the clinical, dosimetric, and combined factors, respectively. Results: With a median follow-up of 36 months, grade >= 2 RP was recorded in 13.4% of patients. On univariate logistic regression analysis, clinical factors of age and lung volume, dosimetric factors of treatment durations, fractional dose and V 10 , and laboratory biomarkers of neutrophil, PLT, PLR, and Hb levels were significantly associated with grade >= 2 RP. However, on multivariate analy-sis, only age, lung volume, fractional dose, V 10 , and Hb levels were independent factors. AUC values for the clinical, dosimetric, and combined models were 0.730 (95% CI, 0.660-0.793), 0.711 (95% CI, 0.641-0.775) and 0.830 (95% CI, 0.768-0.881), respectively. The combined model provided superior discriminative ability than the clinical and dosimetric models ( P < .05). Conclusion: Age, lung volume, fractional dose, V 10 , and Hb levels were demonstrated to be significant factors associated with grade >2 RP for lung cancer patients after SBRT. A novel model combining clinical, dosimetric factors, and laboratory biomarkers improved predictive performance compared with the clinical and dosimetric model alone.