A novel tool for predicting the survival of endoprosthesis used for reconstruction of the knee following tumor resection: a retrospective cohort study

Abstract

Abstract Background Prosthesis-related complications, after knee reconstruction with endoprosthesis during operation for tumors around the knee, remain an unresolved problem which necessitate a revision or even an amputational surgery. The purpose of the current study was to identify significant risk factors associated with implant failure, and establish a novel model to predict survival of the prosthesis in patients operated with endoprostheses for tumor around knee. Methods We retrospectively reviewed the clinical database of our institution for patients who underwent knee reconstruction due to tumors. A total of 203 patients were included, including 123 males (60.6%) and 80 (39.4%) females, ranging in age from 14 to 77 years (mean: 34.3 ± 17.3 years). The cohort was randomly divided into training (n = 156) and validation (n = 47) samples. Univariable COX analysis was used for initially identifying potential independent predictors of prosthesis survival with the training group (p < 0.150). Multivariate COX proportional hazard model was selected to identify final significant prognostic factors. Using these significant predictors, a graphic nomogram, and an online dynamic nomogram were generated for predicting the prosthetic survival. C-index and calibration curve were used for evaluate the discrimination ability and accuracy of the novel model, both in the training and validation groups. Results The 1-, 5-, and 10-year prosthetic survival rates were 94.0, 90.8, and 83.0% in training sample, and 96.7, 85.8, and 76.9% in validation sample, respectively. Anatomic sites, length of resection and length of prosthetic stem were independently associated with the prosthetic failure according to multivariate COX regression model (p<0.05). Using these three significant predictors, a graphical nomogram and an online dynamic nomogram model were generated. The C-indexes in training and validation groups were 0.717 and 0.726 respectively, demonstrating favourable discrimination ability of the novel model. And the calibration curve at each time point showed favorable consistency between the predicted and actual survival rates in training and validation samples. Conclusions The length of resection, anatomical location of tumor, and length of prosthetic stem were significantly associated with prosthetic survival in patients operated for tumor around knee. A user-friendly novel online model model, with favorable discrimination ability and accuracy, was generated to help surgeons predict the survival of the prosthesis.