IntroductionSeveral studies have examined the risk factors for post-percutaneous kyphoplasty (PKP) refractures and developed many clinical prognostic models. However, no prior research exists using the Random Forest (RF) model, a favored tool for model development, to predict the occurrence of new vertebral compression fractures (NVCFs). Therefore, this study aimed to investigate the risk factors for the occurrence of post-PKP fractures, compare the predictive performance of logistic regression and RF models in forecasting post-PKP fractures, and visualize the logistic regression model.MethodsWe collected clinical data from 349 patients who underwent PKP treatment at our institution from January 2018 to December 2021. Lasso regression was employed to select risk factors associated with the occurrence of NVCFs. Subsequently, logistic regression and RF models were established, and their predictive capabilities were compared. Finally, a nomogram was created.ResultsThe variables selected using Lasso regression, including bone density, cement distribution, vertebral fracture location, preoperative vertebral height, and vertebral height restoration rate, were included in both the logistic regression and RF models. The area under the curves of the logistic regression and RF models were 0.868 and 0.786, respectively, in the training set and 0.786 and 0.599, respectively, in the validation set. Furthermore, the calibration curve of the logistic regression model also outperformed that of the RF model.ConclusionThe logistic regression model provided better predictive capabilities for identifying patients at risk for post-PKP vertebral fractures than the RF model.

中文翻译：

---

通过逻辑回归建立列线图预测经皮椎体后凸成形术后再骨折

简介多项研究探讨了经皮椎体后凸成形术（PKP）术后再骨折的危险因素，并开发了许多临床预后模型。然而，之前没有研究使用随机森林 (RF) 模型（模型开发的常用工具）来预测新的椎体压缩性骨折 (NVCF) 的发生。因此，本研究旨在探讨PKP术后骨折发生的危险因素，比较Logistic回归和RF模型在预测PKP术后骨折中的预测性能，并对Logistic回归模型进行可视化。方法收集349例患者的临床数据2018年1月至2021年12月在我机构接受PKP治疗的患者。采用Lasso回归筛选与NVCF发生相关的危险因素。随后建立了Logistic回归和RF模型，并比较了它们的预测能力。最后创建列线图。结果使用Lasso回归选择的变量，包括骨密度、骨水泥分布、椎体骨折位置、术前椎体高度和椎体高度恢复率，均包含在逻辑回归和RF模型中。逻辑回归和 RF 模型的曲线下面积在训练集中分别为 0.868 和 0.786，在验证集中分别为 0.786 和 0.599。此外，逻辑回归模型的校准曲线也优于 RF 模型。 结论 逻辑回归模型比 RF 模型在识别 PKP 后椎体骨折风险患者方面提供了更好的预测能力。