External Validation and Optimization of the SPRING Model for Prediction of Survival After Surgical Treatment of Bone Metastases of the Extremities.

BACKGROUND: Survival predictions before surgery for metastatic bone disease in the extremities (based on statistical models and data of previous patients) are important for choosing an implant that will function for the remainder of the patient's life. The 2008-SPRING model, presented in 2016, enables the clinician to predict expected survival before surgery for metastatic bone disease in the extremities. However, to maximize the model's accuracy, it is necessary to maintain and update the patient database to refit the prediction models achieving more accurate calibration.

QUESTIONS/PURPOSES: The purposes of this study were (1) to refit the 2008-SPRING model for prediction of survival before surgery for metastatic bone disease in the extremities with a more modern cohort; and (2) to evaluate the performance of the refitted SPRING model in a population-based cohort of patients having surgery for metastatic bone disease in the extremities.

METHODS: We produced the 2013-SPRING model by adding to the 2008-SPRING model (n = 130) a cohort of patients from a consecutive institutional database of patients who underwent surgery for bone metastases in the extremities with bone resection and reconstruction between 2009 and 2013 at a highly specialized surgical center in Denmark (n = 140). Currently the model is only available as the nomogram fully available in the current article, which is sufficient to use in daily clinical work, but we are working on making the tool available online. As such, the 2013-SPRING model was produced using a consecutive cohort of patients (n = 270) treated during an 11-year period (2003-2013) called the training cohort, all treated with bone resection and reconstruction. We externally validated the 2008-SPRING and the 2013-SPRING models in a prospective cohort (n = 164) of patients who underwent surgery for metastatic bone disease in the extremities from May 2014 to May 2016, called the validation cohort. The validation cohort was identified from a cross-section of the Danish population who were treated for metastatic lesions (using endoprostheses and internal fixation) in the extremities at five secondary surgical centers and one highly specialized surgical center. This cross-section is representative of the Danish population and no patients were treated outside the included centers as a result of public healthcare settings. The indications for surgery for training and the validation cohort were pathologic fracture, impending fracture, or intractable pain despite radiation. Exact date of death was known for all patients as a result of the Danish Civil Registration System and no loss to followup existed. In the training cohort, 150 patients (out of 270 [56%]) and in the validation cohort 97 patients (out of 164 [59%]) died of disease within 1 year postoperatively. The 2013 model did not differ from the 2008 model and included hemoglobin, complete fracture/impending fracture, visceral and multiple bone metastases, Karnofsky Performance Status, and the American Society of Anesthesiologists score and primary cancer. The models were evaluated by area under the receiver operating characteristic curve (AUC ROC) and Brier score (the lower the better).

RESULTS: The 2013-SPRING model was successfully refitted with a cohort using more patients than the 2008-SPRING model. Comparison of performance in external validation between the 2008 and 2013-SPRING models showed the AUC ROC was increased by 3% (95% confidence interval [CI], 0%-5%; p = 0.027) and 2% (95% CI, 0%-4%; p = 0.013) at 3-month and 6-month survival predictions, respectively, but not at 12 months at 1% (95% CI, 0%-3%; p = 0.112). Brier score was improved by -0.018 (95% CI, -0.032 to -0.004; p = 0.011) for 3-month, -0.028 (95% CI, -0.043 to -0.0123; p < 0.001) for 6-month, and -0.014 (95% CI, -0.025 to -0.002; p = 0.017) for 12-month survival prediction.

CONCLUSIONS: We improved the SPRING model's ability to predict survival after surgery for metastatic bone disease in the extremities. As such, the refitted 2013-SPRING model gives the surgeon a tool to assist in the decision-making of a surgical implant that will serve the patient for the remainder of their life. The 2013-SPRING model may provide increased quality of life for patients with bone metastasis because potential implant failures can be minimized by precise survival prediction preoperatively and the model is freely available and ready to use from the current article.

LEVEL OF EVIDENCE: Level I, diagnostic study.