TY - JOUR
T1 - P1.04-44 Radiomics for Predicting Response to First-Line Anti-PD1 Therapy in Advanced NSCLC
AU - Ackermann, Christoph J.
AU - Fornacon-Wood, I.
AU - Tay, R.
AU - Manoharan, P.
AU - Price, G.
AU - Lindsay, C.
AU - Faivre-Finn, C.
AU - Blackhall, F.
AU - Cobben, D.
PY - 2019/10
Y1 - 2019/10
N2 - Background: Radiomics is the high-throughput extraction of quantitative imaging features from medical images that can reflect underlying tumour pathophysiology. Imaging biomarkers have the potential to improve disease characterisation and predict patient outcomes. In this study, the utility of radiomic features to predict response and survival to first-line immune check-point inhibition with pembrolizumab in advanced non-small cell lung cancer (NSCLC) was explored. Method: Patients with Stage IIIB/IV NSCLC treated with first-line pembrolizumab and PD-L1 ≥50% were retrospectively identified and stratified by Best Overall Response (BOR) by RECIST 1.1. Patients with the primary tumour in situ and a contrast-enhanced CT thorax/abdomen (minimum 5mm CT slice thickness) at baseline were included. The single largest thoracic lesion was segmented in the diagnostic image using the Pinnacle radiotherapy treatment planning system. All tumour delineations were supervised by a highly experienced certified senior radiologist. Lesions 0.85) were removed from further analysis. Least Absolute Shrinkage and Selection Operator (LASSO) feature selection was performed to find informative features that could predict either best overall response or overall survival. Univariate logistic regression and cox proportional hazard models were then used for an initial assessment of the potential of these features in predicting response and survival respectively. Result: Sixteen patients with evaluable best overall response (partial response n=9, progressive disease n=7) were selected for the initial discovery-cohort. Mean age was 68 years with 63% adenocarcinoma histology. From the 47 features extracted, 32 were highly correlated to each other and were removed from further analysis. For predicting best overall response, LASSO selected 5 features with univariate logistic regression suggesting that tumour surface area to volume ratio might be informative (p=0.057, AUC of 0.83 (95% CI 0.61-1.0)). With respect to overall survival, LASSO selected 3 features with univariate cox regression suggesting the first-order feature skewness might be predictive (HR = 0.27, 95% CI 0.08-0.88, p=0.03). When split on the median skewness value the Kaplan-Meier plot showed a significant survival difference between high and low risk patients (p=0.007). Conclusion: Radiomic features extracted from baseline contrast-enhanced CT scans may have the potential to predict response and survival in patients treated with first-line pembrolizumab in advanced NSCLC. We emphasize the exploratory nature of these results given the very limited number of patients in the study. We are expanding this discovery cohort to further investigate and validate these results. Updated results will be presented at the meeting. Keywords: NSCLC, Radiomics, Immunotherapy
AB - Background: Radiomics is the high-throughput extraction of quantitative imaging features from medical images that can reflect underlying tumour pathophysiology. Imaging biomarkers have the potential to improve disease characterisation and predict patient outcomes. In this study, the utility of radiomic features to predict response and survival to first-line immune check-point inhibition with pembrolizumab in advanced non-small cell lung cancer (NSCLC) was explored. Method: Patients with Stage IIIB/IV NSCLC treated with first-line pembrolizumab and PD-L1 ≥50% were retrospectively identified and stratified by Best Overall Response (BOR) by RECIST 1.1. Patients with the primary tumour in situ and a contrast-enhanced CT thorax/abdomen (minimum 5mm CT slice thickness) at baseline were included. The single largest thoracic lesion was segmented in the diagnostic image using the Pinnacle radiotherapy treatment planning system. All tumour delineations were supervised by a highly experienced certified senior radiologist. Lesions 0.85) were removed from further analysis. Least Absolute Shrinkage and Selection Operator (LASSO) feature selection was performed to find informative features that could predict either best overall response or overall survival. Univariate logistic regression and cox proportional hazard models were then used for an initial assessment of the potential of these features in predicting response and survival respectively. Result: Sixteen patients with evaluable best overall response (partial response n=9, progressive disease n=7) were selected for the initial discovery-cohort. Mean age was 68 years with 63% adenocarcinoma histology. From the 47 features extracted, 32 were highly correlated to each other and were removed from further analysis. For predicting best overall response, LASSO selected 5 features with univariate logistic regression suggesting that tumour surface area to volume ratio might be informative (p=0.057, AUC of 0.83 (95% CI 0.61-1.0)). With respect to overall survival, LASSO selected 3 features with univariate cox regression suggesting the first-order feature skewness might be predictive (HR = 0.27, 95% CI 0.08-0.88, p=0.03). When split on the median skewness value the Kaplan-Meier plot showed a significant survival difference between high and low risk patients (p=0.007). Conclusion: Radiomic features extracted from baseline contrast-enhanced CT scans may have the potential to predict response and survival in patients treated with first-line pembrolizumab in advanced NSCLC. We emphasize the exploratory nature of these results given the very limited number of patients in the study. We are expanding this discovery cohort to further investigate and validate these results. Updated results will be presented at the meeting. Keywords: NSCLC, Radiomics, Immunotherapy
UR - https://www.mendeley.com/catalogue/6d28d581-c1d7-3e67-ad01-9e8af9b6a942/
U2 - 10.1016/j.jtho.2019.08.947
DO - 10.1016/j.jtho.2019.08.947
M3 - Meeting Abstract
SN - 1556-0864
VL - 14
SP - S457-S458
JO - Journal of Thoracic Oncology
JF - Journal of Thoracic Oncology
IS - 10
ER -