Risk stratification for immune-related pneumonitis in esophageal cancer: differential roles of systemic inflammation across severity grades and additive effect of radiation

Background Immune checkpoint inhibition is increasingly integrated with thoracic radiotherapy in esophageal squamous cell carcinoma (ESCC), but checkpoint inhibitor pneumonitis (CIP) remains a clinically important toxicity. Whether baseline systemic inflammation modifies the association between lung radiation dose and clinically significant CIP is unclear. Methods This single-center retrospective cohort included 268 patients with ESCC treated with PD-1/PD-L1 inhibitors; 154 subsequently received thoracic radiotherapy. Baseline systemic inflammation was assessed using the systemic immune-inflammation index (SII) from a complete blood count obtained within 7 days before the first immunotherapy dose. The primary endpoint was grade ≥2 CIP according to CTCAE v5.0; grade ≥1 and grade ≥3 CIP were secondary endpoints. Multivariable Cox models evaluated clinical risk factors, and SII-by-V20 interaction analyses were performed in the radiotherapy subgroup. The grade ≥2 nomogram underwent 10-fold cross-validation, 1,000-resample bootstrap validation/calibration, and decision-curve analysis. Results In the overall cohort, 77/268 (28.7%), 29/268 (10.8%), and 19/268 (7.1%) patients developed grade ≥1, grade ≥2, and grade ≥3 CIP, respectively. In the radiotherapy subgroup, the corresponding event counts were 54/154 (35.1%), 20/154 (13.0%), and 13/154 (8.4%). High SII, thoracic radiotherapy, and COPD were independently associated with CIP, while higher BMI was protective. The association of V20 with CIP was concentrated in the high-SII subgroup, with significant SII×V20 interactions across severity thresholds. The nomogram had a C-index of 0.758 (95% CI, 0.680–0.836); the mean 10-fold cross-validated C-index was 0.74 (95% CI, 0.69–0.79), and the bootstrap optimism-corrected C-index was 0.73. Calibration and decision-curve analyses supported acceptable internal performance and potential clinical utility. Conclusions Baseline systemic inflammation was associated with heterogeneity in the relationship between lung V20 and subsequent CIP risk. These findings support a hypothesis-generating host–dosimetry framework for risk stratification but do not establish a validated V20 threshold for treatment modification. Independent external validation and mechanistic studies are required before clinical implementation.