External validation of the SORG machine learning for 90-day and 1-year mortality in patients suffering from extremity metastatic disease in an European cohort of 174 patients.

Abstract

Accurate survival prediction of patients with long-bone metastases is challenging, but important for optimizing treatment. The Skeletal Oncology Research Group (SORG) machine learning algorithm (MLA) has been previously developed and internally validated to predict 90-day and 1-year survival. External validation showed promise in the United States and Taiwan. To ensure global generalizability, the algorithm remains to be validated in Europe. We therefore asked: does the SORG-MLA for long-bone metastases accurately predict 90-day and 1-year survival in a European cohort? One-hundred seventy-four patients undergoing surgery for long-bone metastases between 1997-2019 were included at a tertiary referral Orthopaedic Oncology Center in the Netherlands. Model performance measures included discrimination, calibration, overall performance, and decision curve analysis. The SORG-MLA retained reasonable discriminative ability, showing an area under the curve of 0.73 for 90-day mortality and 0.77 for 1-year mortality. However, the calibration analysis demonstrated overestimation of European patients' 90- day mortality (calibration intercept -0.54, slope 0.60). For 1-year mortality (calibration intercept 0.01, slope 0.60) this was not the case. The Brier score predictions were lower than their respective null model (0.13 versus 0.14 for 90-day; 0.20 versus 0.25 for 1-year), suggesting good overall performance of the SORG-MLA for both timepoints. The SORG-MLA showed promise in predicting survival of patients with extremity metastatic disease. However, clinicians should keep in mind that due to differences in patient population, the model tends to underestimate survival in this Dutch cohort. The SORG model can be accessed freely at https://sorg-apps.shinyapps.io/extremitymetssurvival/.