External validation of prognostic models for overall survival (OS) in patients (pts) with advanced cancer (UC) treated with cisplatin-based chemotherapy.

Abstract

4592 Background: The most commonly used model predicting OS for UC pts treated with cisplatin-based chemotherapy is based on 2-variables (visceral metastases and performance status), developed at MSKCC in 1999, and validated in a phase III study (DeSantis JCO 2011). A prognostic model of OS for advanced UC pts based on 4 variables (visceral metastases, albumin, performance status, and hemoglobin) was developed using 308 pts from MSKCC (ASCO 2007 abstr 5055). We report the discriminative ability of the 4- and 2- variable models for advanced UC pts using an independent dataset from CALGB 90102. Methods: The analysis was performed using an external multi-institutional dataset from CALGB 90102. The primary measurement of predictive discrimination was Harrell’s c-index which was computed with 95% confidence interval (CI). To assess whether there was a statistically significant difference in discrimination between the two models, the U statistic was used to test whether the predictions of the 4-variable model in all possible pairs were more concordant with actual observations than the 2-variable model in the same pairs. Results: CALGB 90102 included 74 UC pts (58 males, 16 females), median age 64 years, treated with cisplatin, gemcitabine and gefitinib, enrolled from 7/02 to 4/05 with a median follow-up of 72.5 months. Visceral metastases were present in 64% (bone, 18%, liver, 31%, lung, 43%), median KPS 90%. The MSKCC 2-variable risk group distribution was 30% =0, 65%= 1 and 5%=2. The median OS =12.7 months (95% CI=10.4-20.5) with 68 deaths observed. When applied to the CALGB cohort, the predictive accuracy for the 4- and 2-variable models were 0.63 (95 CI= 0.56- 0.69) and 0.58 (95% CI= 0.52-0.65), respectively. There was a statistically significant difference in discrimination between the two models (p =0.019), with superiority of the 4-variable model compared to the 2-variable model. Conclusions: A 4-variable prognostic nomogram for survival in pts with advanced UC was superior to a 2-variable risk-group model. The 4-variable prognostic model may replace the widely used 2-variable model and can be used in the design and conduct of future phase II and III trials in advanced UC.