Validación de un modelo predictivo de supervivencia en cáncer renal localizado. Valor de los índices inflamatorios neutrófilo/linfocito y plaqueta/linfocito

Abstract

BACKGROUND: Renal cell carcinoma (RCC) is a heterogeneous disease. Identification of prognostic factors is necessary for patients counselling and treatment decisions. Several anatomical, histological, molecular, and clinical prognostic factors have been described; despite this, its incorporation into clinical practice has been irregular. Integration of these prognostic factors into predictive models constitutes a tool to estimate the probability of an event in a specific patient and to predict a specific risk. For these models to be accurate, they must be applicable to a different population from their generation, what is called external validation. However, the lack of validation of the existing models as well as the absence of robust markers have made the implementation of these models in RCC scarce. HYPOTHESIS: Determination of prognostic factors and identifying patients at high risk of relapse is essential for the management of RCC. Despite the existence of multiple preoperative and postoperative nomograms, none have been consolidated in clinical practice. One of the main reasons is the lack of external validation. It is essential that its predictive performance be empirically evaluated in datasets that were not used to develop the model. The external validation, together with the incorporation of inflammatory markers , would allow the applicability of the nomogram as well as the improvement of its accuracy. OBJECTIVES: Carry out the external validation of the postoperative models of overall survival (OS), cancer-specific survival (CSS) and progression-free survival (PFS) generated from the original cohort of the Hospital Universitario Príncipe de Asturias (HUPA) in a cohort of patients with localized RCC at Ramón y Cajal University Hospital (HURyC). To determine the impact of the incorporation of the neutrophil lymphocyte ratio (NLR) and the platelet lymphocyte ratio (PLR) to the postoperative models of OS, CSS and PFS generated from the original HUPA cohort. METHODS: A retrospective cohort study was conducted including 173 patients who underwent surgery for localized RCC at HURyC in the period between May 2009 and December 2014 with a minimum follow-up of three years. Patients were selected following the same inclusion criteria used in the development of the generation model. The accuracy of the model was evaluated through calibration and discrimination analysis. The calibration analysis was performed graphically, by comparing survival curves predicted by the model and real survival curves estimated by Kaplan-Meier. The discrimination capacity of the model was analyzed using Harrell's Concordance Index (c-index). RESULTS: The OS model generated from the original HUPA cohort was validated in the HURyC cohort ,showing good calibration and good discrimination capacity with a c-Index of 0.770. The CSS model was not validated in the HURyC cohort of patients with localized RCC by not demonstrating a good calibration. The generation model systematically overestimates the probability of cancer-specific mortality (CSM) in the validation cohort. By the other hand, discrimination capacity remained high for the validation cohort with a c-Index of 0.927. The PFS model was not validated in the HURyC cohort as a good calibration could not be demonstrated. The generation model systematically overestimates the probability of recurrence in the validation cohort. On the contrary, the discrimination capacity remained high for the validation cohort with a c-Index of 0.810. The incorporation of the NLR and PLR to the OS, CSS and PFS models did not increase the discrimination capacity of the original generation cohort at any case. CONCLUSIONS: Only the predictive model of OS generated in the HUPA has been able to be validated in the HURyC population. The CSS and PFS models could not be validated in our cohort, since despite a good discrimination capacity, they are not properly calibrated and overestimate CSM and relapse for the HURyC population. The incorporation of the NLR and PLR inflammation markers failed to increase the predictive ability of the initial model.