Non-small Cell Lung Cancer Datasets Research Articles

SU‐E‐T‐244: Interfractional Motion Simulation of Head and Neck and Lung Patients Correlated to Treatment Outcome

Purpose: To quantify the correlation between gross tumor volume (GTV) margins and local control (LC) outcome in head and neck (H&N) and non‐small cell lung cancer (NSCLC) patients. Methods: Interfractional motion simulation was done using the robustness analysis tool in CERR. This tool applies shifts to individual patients fraction by fraction within an entire course of treatment obtaining a final integral dose distribution. The shifts are defined assuming a Gaussian distribution of possible positioning for every fraction. Because of their fractionation schemes, 35 shifts were applied to the H&N, and 30 to the NSCLC dataset. We applied isotropic random shifts sampled from three independent normal distributions all with standard deviations of 3mm. Systematic errors were not included in this study, nor roll rotations. We repeated the process 10 times and averaged the integral doses. Ring structures around the GTV extending isotropically were created. The ring's width ranged from 2 mm to 15mm in H&N, and from 2mm to 20mm in NSCLC. A cell‐kill based equivalent uniform dose (cEUD) metric was calculated for each structure. A logistic regression model was built using cEUD for the margins and for the GTV. LC prediction was evaluated using Spearman correlation coefficient (Rs) on the entire datasets and also on cross‐validation. Results: Adding a margin to GTV did not improve LC correlation in H&N after cross‐validation (Rs= 0.398 p=0.0003 for the GTV alone, and Rs= 0.371 p=0.0007 for the next best added margin). For NSCLC we improved from Rs=0.393 (p=0.003) to Rs=0.413 (p=0.0015) at 5mm margin. Conclusion: Our analysis suggests that in H&N the dose given to the tissue surrounding the GTV is not correlated with LC, but it is for NSCLC. This may be because immobilization devices and setup verification systems are more accurate in H&N and also because the differences in prescription doses.

TH‐C‐213AB‐02: Improved Predictive Modeling of Radiation Pneumonitis in Lung Cancer Patients Using Machine Learning Techniques

Purpose: Early prediction of radiation pneumonitis (RP) could help physicians provide locally advanced non‐small cell lung cancer (NSCLC) patients at high risk of RP with targeted radiotherapy. To date, several RP predictive models have been developed using statistical methods and machine learning techniques. We designed an improved risk model for predicting RP using a recently introduced machine learning method. Methods: We retrospectively analyzed a single NSCLC dataset collected at Washington University School of Medicine, which consisted of 123 NSCLC patients who received 3D conformal radiation therapy with a median prescription dose of 66 Gy and had a median follow‐up of 17 months. In this analysis, 59 patients with grade=2 were considered to have RP. Using clinical variables and lung and heart dose‐volume variables extracted from this dataset, we built a novel RP predictive model based on multivariate logistic regression and supervised principal component analysis (SPCA). The performance of our proposed method was compared with that of multivariate logistic regression alone. Results: Using lung‐related dosimetric variables, SPCA‐based logistic regression model had a Spearman correlation coefficient (Rs) of 0.33 (p=0.0001), whereas logistic regression model had Rs=0.20 (p=0.014). Using heart‐related dosimetric variables, Rs=0.42 (p < 0.0001) and Rs=0.40 (p < 0.0001) were obtained in the SPCA‐based logistic regression and logistic regression models, respectively. Interestingly, when clinical variables alone were used, logistic model showed better performance with Rs=0.34 (p < 0.0001) than SPCA‐based logistic regression model with Rs=0.32 (p < 0.0001). Integrating all variables, SPCA‐based logistic regression model obtained the best performance with Rs=0.44 (p< 0.0001) while logistic model had Rs=0.37 (p< 0.0001). Conclusions: Incorporating multivariate logistic regression with a new machine learning method allowed us to produce a better predictive RP model that should be further tested for potential clinical use.

Prognostic Gene Expression Signature for Squamous Cell Carcinoma of Lung

This study aimed to identify and validate a gene expression signature for squamous cell carcinoma of the lung (SQCC). A published microarray dataset from 129 SQCC patients was used as a training set to identify the minimal gene set prognostic signature. This was selected using the MAximizing R Square Algorithm (MARSA), a novel heuristic signature optimization procedure based on goodness-of-fit (R square). The signature was tested internally by leave-one-out-cross-validation (LOOCV), and then externally in three independent public lung cancer microarray datasets: two datasets of non-small cell lung cancer (NSCLC) and one of adenocarcinoma (ADC) only. Quantitative-PCR (qPCR) was used to validate the signature in a fourth independent SQCC cohort. A 12-gene signature that passed the internal LOOCV validation was identified. The signature was independently prognostic for SQCC in two NSCLC datasets (total n = 223) but not in ADC. The lack of prognostic significance in ADC was confirmed in the Director's Challenge ADC dataset (n = 442). The prognostic significance of the signature was validated further by qPCR in another independent cohort containing 62 SQCC samples (hazard ratio, 3.76; 95% confidence interval, 1.10-12.87; P = 0.035). We identified a novel 12-gene prognostic signature specific for SQCC and showed the effectiveness of MARSA to identify prognostic gene expression signatures.

Identification of Gene Signatures and Molecular Markers for Human Lung Cancer Prognosis using anIn vitroLung Carcinogenesis System

Lung cancer continues to be a major deadly malignancy. The mortality of this disease could be reduced by improving the ability to predict cancer patients' survival. We hypothesized that genes differentially expressed among cells constituting an in vitro human lung carcinogenesis model consisting of normal, immortalized, transformed, and tumorigenic bronchial epithelial cells are relevant to the clinical outcome of non-small cell lung cancer (NSCLC). Multidimensional scaling, microarray, and functional pathways analyses of the transcriptomes of the above cells were done and combined with integrative genomics to incorporate the microarray data with published NSCLC data sets. Up-regulated (n = 301) and down-regulated genes (n = 358) displayed expression level variation across the in vitro model with progressive changes in cancer-related molecular functions. A subset of these genes (n = 584) separated lung adenocarcinoma clinical samples (n = 361) into two clusters with significant survival differences. Six genes, UBE2C, TPX2, MCM2, MCM6, FEN1, and SFN, selected by functional array analysis, were also effective in prognosis. The mRNA and protein levels of one these genes-UBE2C-were significantly up-regulated in NSCLC tissue relative to normal lung and increased progressively in lung lesions. Moreover, stage I NSCLC patients with positive UBE2C expression exhibited significantly poorer overall and progression-free survival than patients with negative expression. Our studies with this in vitro model have lead to the identification of a robust six-gene signature, which may be valuable for predicting the survival of lung adenocarcinoma patients. Moreover, one of those genes, UBE2C, seems to be a powerful biomarker for NSCLC survival prediction.

Recursive partitioning analysis of 1592 patients on four radiation therapy oncology group studies in inoperable non-small cell lung cancer

Survival outcome of 1592 analyzable patients on four Radiation Therapy Oncology Group (RTOG) studies in inoperable non-small cell lung cancer were studied utilizing a recursive partitioning analysis (RPA). This approach creates a regression tree according to prognostic variables which partitions into homogenous subsets according to survival. Four protocols, RTOG 83-11, 83-21, 84-03 and 84-07 were analyzed. 83-11 and 84-07 were studies utilizing irradiation with alterfractionation; 83-21 and 84-03 were studies evaluating thymocin with irradiation and prophylactic cranial irradiation with thoracic irradiation respectively. Nine pretreatment variables and one treatment variable were analyzed. Adjustment for radiotherapy effect was made in the accelerated treatment protocol (84-07). Overall, median survival for the entire group was 9.0 months with 17% alive at 2 years. Univariate analysis suggests that KPS, ≤ 70 vs. 80–100, pleural effusion, weight loss, ≤ 5% vs. 5%, age, 60 + vs. < 60, T stage (T1 and T2 vs. T3 and T4) and N stage (N- vs. N+) were important prognastic factors. Radiation dose, sex, race and histology were not univariate prognastic factors. RPA identified KPS as the most significant covariate (median survival 5.9 mos. ≤ 70 vs. 9.0 mos. 80–100). Within KPS 80–100 other splits occurred for N stage, age, weight loss and radiation therapy dose. KPS ≤ 70 split at pleural effusion only. The best overall RPA tree has four distinct classes with median survival times ranging from 3.3 to 12.6 months. The RPA classes were validated in an independent non-small cell lung cancer dataset. This analysis may allow more intelligent stratification and study-design for future RTOG trials in inoperable non-small cell lung cancer.