Virtual Library

Start Your Search

J. Wang



Author of

  • +

    P1.19 - Poster Session 1 - Imaging (ID 179)

    • Event: WCLC 2013
    • Type: Poster Session
    • Track: Imaging, Staging & Screening
    • Presentations: 1
    • +

      P1.19-010 - Prognostic Value of FDG-PET-based Metabolic Tumor Volume<br /> in Conjunction with Mathematical Classifiers<br /> in Non-small Cell Lung CancerPatients (ID 2983)

      09:30 - 09:30  |  Author(s): J. Wang

      • Abstract

      Background
      The TNM staging system for prognostication in non-small cell lung cancer (NSCLC) is less than satisfactorily accurate. Previous research suggests that FDG-PET-based metabolic tumor volume (MTV) may be an independent prognostic marker in NSCLC when used in Cox-regression models. This study was to determine the potential prognostic utility of FDG-PET-based MTV used in mathematical classifiers for prognostication of NSCLC.

      Methods
      A consecutive cohort of 328 NSCLC patients (156 males, 172 females) with histologic confirmation and FDG-PET scans was identified for retrospective analysis. MTV measurements were made on baseline 18F-FDG PET/CT scans of the primary tumor (MTV~T~), metastatic lymph nodes (MTV~N~), and metastatic tumors (MTV~M~). Whole-body MTV (MTV~WB~) was defined as the sum of MTVs of primary tumor, metastatic lymph nodes, and metastatic tumors (MTV~WB~ = MTV~T~ + MTV~N~ + MTV~M~). A semi-automated 3D contouring program was used for obtaining the MTVs. Patient survival was determined from cancer registry data, and known survival length of living and lost-to-follow-up (i.e., censored) patients was recorded. Patient survival was determined at one year (189 survived, 129 did not) and two years (127 survived, 186 did not) after baseline FDG-PET scan, and survival status was known for most censored patients except for ten at one year and 15 at two years. Linear discriminant analysis classifiers were constructed on data of patient age, gender, histology, TNM stage, and MTVs. Cross-validation was done as follows: to predict a patient's survival status, a classifier was trained on data of all other patients of known survival status and not on the patient in question. Areas under receiver operating characteristic (ROC) curve (AUC), analogous to C concordance statistics, were compared statistically.

      Results
      The median patient age was 68.3 years and patient distribution by stage was 46 stage IA, 43 stage IB, 19 stage IIA, 18 stage IIB, 52 stage IIIA, 39 stage IIIB, and 111 stage IV. Median follow-up of living and lost-to-follow-up patients was 58 months and death confirmation was known in 249 (88.4%) patients. For one-year survival, the AUC (± standard error) value of the classifier based on age, gender, histology, and stage was 0.77±0.026 and that of the classifier based on age, gender, histology, stage, and MTVs was 0.82±0.023 (p < 0.01). For two-year survival, the corresponding AUC values were 0.75±0.028 and 0.79±0.026, respectively (p < 0.02). The AUC values were similar for classifiers based on age, gender, histology and MTVs and for classifiers based on age, gender, histology, and stage (p > 0.5). These results were consistent with previous results of statistically significantly better prognostic ability of multivariate Cox-regression models of MTV~WB~ compared with models not including MTV~WB~, after adjusting for age, gender, histology, treatment options, and whole-body tumor SUV~max~.

      Conclusion
      FDG-PET-based metabolic tumor volume can potentially increase prognostic accuracy for NSCLC patients beyond that of the TNM staging system at one and two years, and mathematical linear classifiers can potentially be used as an alternative to Cox-regression models for estimating survival probability.