Development and validation of a nomogram for differentiating immune checkpoint inhibitor-related pneumonitis from pneumonia in patients undergoing immunochemotherapy: a multicenter, real-world, retrospective study

Abstract

Background

Immune Checkpoint Inhibitor-related Pneumonitis (CIP) exhibits high morbidity and mortality rates in the real world, often coexisting with pneumonia, particularly after immunochemotherapy. We aimed to develop and validate a non-invasive nomogram for differentiating CIP from pneumonia in patients undergoing immunochemotherapy.

Methods

This study encompassed 237 patients from three hospitals. A multivariate logistic regression analysis was conducted to identify risk factors for CIP. Utilizing the random forest machine learning method, optimal development and validation cohort allocation ratios (in a ratio of 8:2) were determined for the predictive model. The performance of the nomogram was evaluated using calibration, the area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA). Subsequently respiratory pathogens, management, and outcomes were compared between CIP and No CIP cases.

Results

Among the 237 patients, 104 were diagnosed with CIP, and 133 were no CIP but pneumonia(No CIP). Smoking status, prior chronic obstructive pulmonary disease (COPD), ground glass opacities, non-specific interstitial pneumonitis, Neutrophil to Lymphocyte Ratio (NLR), pleural effusions, and Oxygen Partial Pressure (PaO2) emerged as non-invasive independent predictors of CIP. The nomogram exhibited good discrimination for both the development and validation cohorts, with AUC values of 0.817 (95% CI, 0.754–0.879) and 0.913 (95% CI, 0.826–0.999), respectively. The calibration curves demonstrated good fit for both the development and validation cohort, as evidenced by the Hosmer-Lemeshow tests (χ² = 3.939, p = 0.863 and χ² = 8.117, p = 0.422, respectively). DCA further highlighted their clinical utility. In CIP patients, the use of gamma globulin/albumin and glucocorticoids was significantly higher than in No CIP patients (39.4% vs 23.3%, p = 0.007; 79.8% vs 12.8%, p < 0.0001, respectively). The proportion of patients requiring mechanical ventilation was also significantly higher in the CIP compared to the No CIP group (21.2% vs 11.3%, p = 0.038).

Conclusion

The nomogram offers a non-invasive approach to differentiate CIP from pneumonia associated with immunochemotherapy, potentially facilitating early intervention and informed treatment decisions.