Is checkpoint inhibitor pneumonitis underreported in patients with advanced non-small cell lung cancer (NSCLC) on PD-1 inhibitor monotherapy?

Abstract

9579 Background: For patients with advanced non-small cell lung cancer (NSCLC), immunotherapy (ImT) has led to improvements in survival and quality of life. Checkpoint inhibitor pneumonitis (CIP) is an uncommon but sometimes life-threatening adverse event. While CIP is a diagnosis of exclusion, many oncologists believe the incidence of CIP is underreported. Radiomics, an image analysis technique that can extract imperceptible information from radiographic images, has been incorporated into predictive models for many cancers. Recent studies suggest that radiomic analysis of pre-ImT imaging can predict CIP. We hypothesized that for patients with advanced NSCLC treated with Nivolumab monotherapy, the rate of CIP is underreported and radiomics features can identify CIP that was clinically misclassified. Methods: From an IRB-approved database (DB) of 159 patients with advanced NSCLC treated with Nivolumab, chart review identified 8 patients diagnosed with CIP of any grade (5%). 42 additional patients from the same DB without diagnosis of CIP were randomly selected for analysis. For all 50 patients, uninvolved lung in the last pre-ImT CT imaging study was segmented, delineated, and analyzed for radiomics features associated with CIP. A logistic regression model incorporating radiomics assigned a CIP probability score to every patient. Results: Six radiomics features correlated with CIP ( p-values range from 0.02 to 0.03). Each feature had an AUC of ~0.79 (range 0.789 to 0.794) showing large effect size, with odds ratios greater than 3.50 (4 features) or less than 0.3 (2 features). The radiomics-based probability model assigned 7/42 patients (17.5%) without clinical diagnosis of CIP a greater than 50% probability of CIP. Chart review revealed that 6/7 “misclassified” patients exhibited symptoms or radiographic features highly suggestive of CIP within 5 months of initial immunotherapy treatment. These indications originally had been attributed to disease progression, overshadowed by more severe symptoms or simply mislabeled (e.g. a case of recall pneumonitis was described as "radiation pneumonitis"). Conclusions: For patients with advanced NSCLC treated with nivolumab, the incidence of checkpoint-inhibitor pneumonitis (CIP) is underreported and radiomics features can help identify CIP that has been clinically misclassified. Future directions include expansion of this study across the full database, correlation of radiomics features with blood biomarkers, and the inclusion of tumor burden as an additional covariate in the analysis.