Longitudinal prognostication in retroperitoneal sarcoma survivors: Development and external validation of two dynamic nomograms.

Abstract Purpose The aim of this study was to create and validate dynamic nomograms to predict overall survival (OS) and disease-free survival (DFS) at different time points during follow-up in patients who had undergone resection of primary retroperitoneal sarcoma (RPS). Methods Patients with primary RPS operated upon between 2002 and 2017 at four and six referral centres comprised the development and external validation cohorts, respectively. Landmark analysis and multivariable Cox models were used to develop dynamic nomograms. Variables were selected using two backward procedures based on the Akaike information criterion. The prediction window was fixed at 5 years. Nomogram performances were tested in terms of calibration and discrimination on the development and validation cohorts. Results Development and validation cohorts totalled 1357 and 487 patients (OS analysis), and 1309 and 452 patients (DFS analysis), respectively. The final OS model included age, landmark time (TLM), tumour grade, completeness of resection and occurrence of local/distant recurrence. The final DFS model included TLM, histologic subtype, tumour size, tumour grade, multifocality and the interaction terms between TLM and size, grade and multifocality. For OS, Harrell C indices were higher than 0.7 in both cohorts, indicating very good discriminative capability. For DFS, Harrell C indices were between 0.64 and 0.72 in the development cohort and 0.62 and 0.68 in the validation cohort. Calibration plots showed good agreement between predicted and observed outcomes. Conclusion Validated nomograms are available to predict the 5-year OS and DFS probability at different time points throughout the first 5 years of follow-up in RPS survivors.