Validation of a prognostic model to predict survival after non-small-cell lung cancer surgery
Objective: Surgery is the first choice of treatment for localised non-small-cell lung cancer (NSCLC). When making decisions regarding resection, physicians must balance the potential long-term benefits of surgery with the risk of surgery-related death, particularly among elderly patients with multiple co-morbid conditions. In 2005, a predictive model with a preoperative and a postoperative mode to predict survival of an individual patient after NSCLC surgery was created. This model combines the patient-, tumour- and treatment characteristics and can be used to assist in clinical decision making. Till present, this model has not been validated. The purpose of this study was to validate this model in patients operated on for primary NSCLC. Methods: A total of 126 patients underwent surgery for primary NSCLC between January 2002 and December 2006. Required model variables were collected for all patients and inserted into the model. To evaluate the performance of the two models, we assessed these models in terms of both discrimination (resolution) and calibration (reliability). The discriminative ability was measured using the c-index and calibration was evaluated by the Hosmer-Lemeshow goodness-of-fit test. Results: The median follow-up time was 3.4 years. Hospital mortality was 2.4%. One-, 2- and 3-year survival was 86%, 75% and 72%, respectively. The discriminative ability of the preoperative mode showed a c-statistic for 1-year survival of 0.68, for 2-year of 0.68 and for 3-year of 0.66. The postoperative model showed a discriminative ability for 1-year survival of 0.72, for 2-year of 0.76 and for 3-year of 0.77. Calibration was adequate for the first 2 years. The preoperative mode showed a p-value of 0.62 for 1-year survival and 0.14 for 2-year survival. Calibration was poor for 3-year survival (p= 0.0027). For the postoperative mode, calibration was quite similar with p-values of 0.4 for 1-year survival, 0.14 for 2-year survival and 0.003 for 3-year survival. Conclusions: The model adequately estimates the 1- and 2-year survival. Discrimination was good for 3-year survival. Inclusion of more factors with additional prognostic value could potentially further improve the accuracy of the model.