Prognostic Model Using Clinical Parameters In Non-Small-Cell Lung Cancer Patients

Abstract

RATIONALE : Clinical parameters have been individually studied to know their prognostic roles in non-small-cell lung cancer (NSCLC) patients. In this study, we tried to develop prognostic model using clinical parameters in the patients. METHOD : Eight hundred NSCLC patients who were diagnosed from Jan. 2001 to Dec. 2009 and took pulmonary function testing at diagnosis were enrolled from prospective lung cancer cohort. The patients were treated with platinum doublets or supportive care only. Patients with stage II were 103 (12.9%), stage III of 240 (30.0%), and stage IV of 457 (57.1%). In addition to forced expiratory volume per 1 sec. (% of pred.) (FEV %), demographic data including smoking habits, and weight loss, treatment status (chemotherapy vs. palliative), 1 histological types, stages, and laboratory estimates (C-reactive protein, hemoglobin, albumin, calcium, lactic dehydrogenase, and alkaline phosphatase) at the diagnosis were evaluated. RESULT : We constructed the final model from initial full model using Cox proportional hazard analysis. With the final model, the hazard ratio (95% CI) for the effect of chemotherapy was 0.42 (0.36, 0.49), adenocarcinoma (referent squamous cell carcinoma) was 1.13 (0.95, 1.34), and other cancer was 1.29 (0.98, 1.70), stage III was 1.11 (0.86, 1.44), and stage IV was 1.56 (1.22, 1.99), log C-reactive protein was 1.13 (1.07, 1.20), weight loss was 1.29 (1.10, 1.51), FEV % was 0.94 (0.91, 0.97), albumin was 0.78 (0.66, 0.91). Gonen and Heller's K index was 1 0.66 which measuring the discrimination performance of the model. We created the prognostic model on a derivation dataset and evaluate its predictions on a validation set. We split the dataset at random into equal, each of size 400 patients, and fit the Royston-Parmar model to the first half. We predicted the prognostic index for all patients, and made four prognostic groups for all patients, using cutpoints at the 15 , 50 , and 85 centiles of the prognostic index in the first half. We compute Kaplan-Meier survival curves in the four th th th prognostic groups in the second half, and finally compare predicted and Kaplan-Meier curves in the second half graphically (Figure 1). We compared survival estimates predicted on the first and second halves from the model fit on the first half. The predicted survival curves agree well with the nonparametric estimates in 15th-50th centiles and 50th-85 centiles groups. CONCLUSION : The model using the clinical parameters appears to be validated well subjectively. Therefore, this would be helpful in classification of NSCLC patients. Figure 1. Graphical validation of a prognostic model in non-small-cell lung cancer patients This abstract is funded by: None Am J Respir Crit Care Med 185;2012:A2527 Internet address: www.atsjournals.org Online Abstracts Issue