Abstract 1283P
Background
ICI-induced pneumonitis is a severe side effect occurring in 4-10% of the patients, requiring an appropriate diagnosis for management with steroids. The differential diagnosis between ICI-related (ICI-R) and other causes of pneumonitis (OP) remains challenging. We therefore developed a prediction model including clinical and radiomic parameters for the differential diagnosis of pneumonitis in stage IV NSCLC patients.
Methods
Patients receiving ICI were prospectively recruited from six centers. CT-based radiomic features were extracted from the segmented lungs using PyRadiomics. Multinomial regression was used for model development (reference=no pneumonitis). One center (205 patients) was used to externally validate the models.
Results
A total of 556 patients, mean age of 64.6±9.4 years and 54.3% males, were included. 64.7% and 31.5% of patients were treated with nivolumab and pembrolizumab, respectively. The incidence of ICI-R and OP was 8.1% and 9.7%, respectively. The following variables were found to be significant predictors for ICI-R pneumonitis in the combined model: number of comorbidities, type of anti-PD(L)1 medication, dyspnea, and Gray Level Co-occurrence Matrix Inverse Difference Moment Normalized (GLCM IDMN). The area under the ROC curve of the clinical, radiomic, and combined models for predicting ICI-R pneumonitis were 0.99, 0.65, and 0.99, respectively (Table). Hosmer-Lemeshow test showed fair calibration for three models. Table: 1283P
Predictive performance of the combined (clinical + radiomic) model
Predicted outcome | ||||
No pneumonitis | ICI-R pneumonitis | Other pneumonitis | ||
Observed outcome | No pneumonitis | 174 | 2 | 2 |
ICI-R pneumonitis | 0 | 5 | 0 | |
Other pneumonitis | 0 | 2 | 0 | |
Sensitivity (%) | 97.8 | 100 | 0 | |
Specificity (%) | 100 | 97.8 | 98.9 | |
PPV (%) | 100 | 55.6 | 0 | |
NPV (%) | 63.6 | 100 | 98.9 | |
Accuracy (95% CI) | 96.8 (93.1, 98.8) | |||
AUC | 0.994 | 0.99 | 96.5 | |
Discrimination slope | 0.917 | 0.699 | 0.183 | |
H-L P value | <0.001 | 0.999 | 0.999 | |
Brier score | 0.022 | 0.015 | 0.016 |
Conclusions
Clinical and radiomic features have the potency to help in the differential diagnosis of pneumonitis in patients receiving ICI.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Bristol Myers Squibb.
Disclosure
All authors have declared no conflicts of interest.