Predicting survival in primary resected retroperitoneal soft tissue sarcoma: A specific nomogram built on three major sarcoma center data sets.

Abstract

10000 Background: To build a specific nomogram for predicting postoperative overall survival (OS) in primary retroperitoneal soft tissue sarcoma (RPS). A few single institutions attempts are already available, but due to the rarity of the disease they are based on limited number of cases with inherent limitations. In an attempt to improve the predictive capability of such a prognostic model we utilized series from 3 major sarcoma centers. Methods: We included patients with primary localized RPS resected with curative intent between 1999 and 2009 at 3 institutions (1 European and 2 North Americans). Univariable (Kaplan Meier plots) and multivariable (Cox model) analyses were carried out. Prognostic covariates were: Age (modeled as continuous covariate using 3 knot restricted cubic spline transformation); Tumor size (modeled as continuous covariate using 3 knot restricted cubic spline transformation); Grade (I,II,III); Histological subtype (Leiomyosarcoma, DD Liposarcoma, WD Liposarcoma, MPNST, Pleomorphic Sa, SFT, Other); Multifocality (no,yes); Quality of surgery (macroscopically complete, incomplete); radiation therapy (RT, done/not done). A backward selection procedure, based on the Akaike Information Criterion (AIC), was applied to select the Cox model covariates. The model discriminative ability was estimated by means of bootstrap-corrected Harrel C statistic. Results: 526 patients were identified. At a median follow-up of 45 months (interquartile range: 22‑72 months) 174 deaths were recorded. 5-yr and 7yr OS (95% confidence interval) were 56.4% (51.1,62.2%) and 50.1% (44.1,57.0%). The backward selection procedure by AIC criterion lead to exclude RT from the covariates set. All other covariates proved to be statistically significant in the final multivariable Cox model. The Bootstrap-corrected Harrell C statistic was 0.74. Of interest, size of the tumor and age of the patient had a bimodal contribution to the risk of death and the combination of histological subtype and grade proved significant. Conclusions: A new multi-institution RPS specific nomogram is now available for prognostication in the clinic and patient selection in clinical trials.