Validation And Improvement Of The Prognostic Models For Nonmetastatic Renal Cell Carcinoma In Chinese Population

Renal cell carcinoma(RCC) is the most common malignancy in adult kidney,represent over 85%of renal cell carcinomas,and accounts for 3%of all human malignancy. Furthermore,approximately 25%of patients will present with metastatic disease.Currently, there are several factors can be used as predictors of survival for patients with RCC, including TNM stage,Fuhrman grade,histologic type,tumor size,tumor necrosis,Eastern Cooperative Oncology Group(ECOG) performance status(PS) and some molecular markers. However,single variable often provide low accuracy in predicting prognoses of RCC,and new prognostic models have been recently proposed by some researchers.These models are able to classify patients in groups with different prognoses or calculate with a punctual precision the survival probabilities of each single patient.Furthermore,they are helpful tools in the planning of postoperative follow-up and the design and interpretation of the results of RCTs.The prediction system of RCC is mathematical models developed based on the statistical method.The models were projected with the objective to calculate the survival probability containing all available clinical and pathological information.With a higher accuracy and easier manipuility,their use in clinical practice and research trials is widespread. According to their structure,the models are classified into formula,algorithm and nomogram. More recently,with the development of gene arrays and proteomics,the model including both the clinical variables and molecular markers was projected,which might provide more accurate prognostic information.Till now,13 prognostic models were proposed in the literature and 9 models were proposed for nonmetastatic renal cell carcinoma.However,these models have varying degrees of accuracy in different patient samples.Furthermore,the research of developing and validating the models were entirely based on Caucasian population,no application of the models in Asian population has been published in the literature to date.The aim of this study was to better define the general applicability of the currently used prognostic models for nonmetastatic RCC in Chinese population based on 15-year experience in a large single center in China.And we also evaluated molecular prognostic markers selected based on a review of the scientific literature by a tissue microarray.Finally,a nomogram including molecular and clinical predictors has been developed based on the Chinese population.The accuracy of the models was also compared with currently used models for predicting survival.Part 1.Validation of the current clinical prognostic models for Nonmetastatic Renal Cell Carcinoma after Nephrectomy in Chinese populationObjective:To explore the general applicability of the current clinical prognostic models for nonmetastatic renal cell carcinoma in Chinese population.Methods:Clinical and pathological variables of 653 nonmetastatic renal cell carcinoma patients in our hospital from 1993 to 2004 were retrospectively reviewed.7 models were used to predict the prognosis,including the Yaycioglu model,the Cindolo model,the UISS model, the SSIGN model,the Kattan nomogram,Sorbellini nomogram and Karakiewicz nomogram. 3 different endpoints were used for validation,including overall survival(OS), cancer-specific survival(CSS),and recurrence-free survival(RFS).Survival was estimated by the Kaplan-Meier method.Discriminating ability was assessed by the Harrell's c-index. All statistical tests were two-sided,with significance defined as P＜0.05.Analyses were performed using SPSS version 13.0 and S-Plus 6 software packages with the Design and Hmisc libraries.Results:(1) At last follow-up,159 patients had died of any causes,123 patients died of cancer progression,and disease recurrence occurred in 156 patients.Overall median follow-up is 65 months.(2) The discriminating ability of all models was confirmed in the Chinese population.Different groups in the same model had significant differences in survival analysis.(3) The Kattan nomogram was the most accurate,with the highest C-indexes of 0.752,0.793 and 0.841 for OS,CSS,and RFS,respectively.Sorbellini nomogram and Karakiewicz nomogram also presented high accuracy,though a little lower than Kattan nomogram,with no significant difference.SSIGN model is the most accurate model in algorithm models,with the C-indexes of 0.712,0.751 and 0.777 for OS,CSS,and RFS,respectively.And the Cindolo model performed as well as the SSIGN model,though only including clinical presentation and size of tumor.In all models,Yaycioglu model showed lowest accuracy,with the C-indexes of 0.616,0.649 and 0.661 for OS,CSS,and RFS, respectively.Conclusions:Mathematical models have a prognostic accuracy higher than the one of the single clinical and/or pathological variables.The results showed that nomograms discriminate better than other models,regardless of endpoints.The Kattan model was found to be the most accurate.This study defines a better applicability of the nomograms for Chinese patients with nonmetastatic RCC treated with nephrectomy.Though with a lower accuracy,algorithms could be a useful tool for patient counseling.Therefore,models should be chosen according to different environments and purposes.Part 2.Validation of the prognostic models including molecular markers for Nonmetastatic Renal Cell Carcinoma after Nephrectomy in Chinese populationObjective:To explore the applicability of the models including molecular markers for nonmetastatic renal cell carcinoma in Chinese population,based on a tissue microarray.Methods:A custom tissue array was constructed from 482 RCC patients who underwent nephrectomy.Clinical and pathological variables of all patients were retrospectively reviewed. Immunohistochemistry was performed for protein markers in the 2 prognostic models both developed by Kim et al,including CA9,Vimentin,P53,ki-67,Gelsolin,VEGFR-1 and VEGF-D.3 different endpoints were used for validation,including overall survival(OS), cancer-specific survival(CSS),and recurrence-free survival(RFS).Survival was estimated by the Kaplan- Meier method.Discriminating ability was assessed by the Harrell's c-index. All statistical tests were two-sided,with significance defined as P＜0.05.Analyses were performed using SPSS version 13.0 and S-Plus 6 software packages with the Design and Hmisc libraries. Results:(1) At last follow-up,167 patients had died of any causes,131 patients died of cancer progression,and disease recurrence occurred in 128 patients.Overall median follow-up is 65 months.(2) The expression of CA9,Vimentin,P53,Ki-67,Gelsolin, VEGFR-1 and VEGF-D were identified in 85.3%(411/482),65.1%(314/482),31.7% (153/482),61.6%(297/482),41.3%(199/482),70.1%(338/482),25.9%(125/482) of renal cell carcinomas,respectively.(3) For descriptive purposes only,individual probability values from the nomogram were arbitrarily categorized in 5 classes(＜0.6,0.6-0.7,0.7-0.8,0.8-0.9, and 0.9-1.0).Different groups in the same model had significant differences in survival analysis.The discriminating ability of all models was confirmed in the Chinese population.(4) The C-indexes of the model developed for nonmetastatic renal cell carcinoma only was 0.812, 0.833 and 0.872 for OS,CSS,and RFS,respectively,which was highest in all models.While the number of the model developed for both the metastatic and nonmetastatic renal cell carcinoma was 0.778,0.782 and 0.799 for OS,CSS,and RFS,respectively.Conclusions:In patients with RCC,a prognostic model for survival that includes molecular and clinical predictors is significantly more accurate than a standard clinical model. The model developed for nonmetastatic renal cell carcinoma showed well applicability in our samples.But the other one did not perform well in the nonmetastatic RCCs because it was developed based on both the metastatic and nonmetastatic population.Furthermore,the different accuracy of the 2 models implied that prognostic implication of tumor markers might differ in various populations,due to the heterogeneity tumorigeness.Therefore,if we want to develop the Chinese version of prognostic model,clinical and molecular markers should be re-examination.Part 3.Development and validation of a new prognostic models for renal cell carcinoma based on the Chinese populationObjective:To propose a prognostic model for renal cell carcinoma that includes molecular and clinical predictors in Chinese population.Methods:A systematic literature review was performed to search for molecular markers influence prognosis in nonmetastatic RCC.Immunohistochemical analysis was done on the tissue microarray of all searched RCC related markers.Associations between predictors and survival time were evaluated with Cox models.According to the result of Cox model, prognostic models were developed for OS,CSS,and RFS,respectively,using markers and clinical predictors significantly affect prognosis.And the prognostic accuracy was assessed by the Harrell's c-index.Results:(1) From the initial search,we selected 10 new makers which might influence the prognosis of RCC,including Ki67,p27,SKP-2,COX-2,HIF-1α,VEGF,Cyclin D1, CXCR3,EpCAM,Survivino(2) On multivariate Cox regression analysis that included all markers and clinical variables,VEGF(P=0.027),p27(P=0.016),p53(P=0.006),T category (P＜0.001),tumor size(P＜0.001) and ECOG-PS(P=0.001) were significant independent predictors of disease specific survival and they were used to construct a combined molecular and clinical prognostic model.(3) The constructed nomogram combined the clinical and molecular factors and approached the concordance index of 0.852,0.883 and 0.901 for OS, CSS and RFS,respectively.The C-index of the model was significantly higher than that of previous models.Conclusions:A nomogram consisting of 6 predictors(VEGF,p27,p53,T category, tumor size and ECOG-PS) was constructed based on the Chinese population.To our knowledge,this is the first prognostic model in Chinese RCC patients.The prognostic ability of the nomogram may be superior to clinical factors alone and even previous models. However,independent extemal validation of the nomogram is required.In our research,we validated the prognostic models for RCC based on the population other than Caucasian for the first time.Furthermore,we also evaluated individual candidate molecular markers for prognostic information.Based on the above research,we finally proposed the first prognostic model in Chinese population.By using the model,we could precisely predict the prognosis of RCC patients.With validation in independent patient samples,the model will be a useful tool for patient counseling,clinical trial design and patient follow-up planning.