Correlation of radiomic features with PD-L1 expression in early stage non-small cell lung cancer (ES-NSCLC) to predict recurrence and overall survival (OS).

e24247 Background: Outcomes in patients with ES-NSCLC (stages 1 and 2) remain dismal despite curative intent surgical resection and adjuvant chemotherapy. With low dose chest CT screening for lung cancer, we are likely to see an increase in patients diagnosed with ES-NSCLC. There is a need for clinically validated biomarkers to predict risk of recurrence. PD-L1 is a transmembrane protein which has been shown to significantly modulate immune responses in carcinogenesis. PD-L1 expression has been associated with poor prognosis in ES-NSCLC. Radiomics has been used to successfully differentiate between patients with high and low risk of recurrence from pre-surgical CT scans. In this study we examined whether radiomic features which were correlated with PD-L1 status could also predict risk of recurrence and OS in ES-NSCLC. Methods: A total of 166 ES-NSCLC pts who had curative surgery with or without adjuvant chemotherapy were chosen for the study. PD-L1 expression on the tumor cells was evaluated with E1L3N anti-PDL1 antibody. The cohort was divided into a training (n = 116) and independent validation set (n = 50). A total of 248 intratumoral and peritumoral radiomic textural features were extracted. Results: A statistical feature selection algorithm was used to select the top three features that separated presence of PD-L1 expression from its absence. These features included two peritumoral features (0-6 mm outside the tumor nodule) and one intratumoral feature representing textural heterogeneity inside and outside the nodule. The machine learning classifier correlated with PD-L1 expression predicted recurrence on the independent validation set (n = 50; AUC = 0.73) and was also prognostic of overall survival (p < 0.001) and recurrence free survival (p < 0.001). Conclusions: Radiomic texture features both inside and outside the tumor were found to correlate with PD-L1 expression and were prognostic of overall and recurrence free survival. Future work will involve evaluating whether these radiomic features are potentially also predictive of PDL1 expression in advanced NSCLC and hence potentially predictive of response to immune checkpoint inhibition.