Scholarlink

Purpose <p>C5 palsy (C5P) is one of the main postoperative complications of ossification of the posterior longitudinal ligament (OPLL). However, an accurate predictive tool for C5P remains to be developed. Forecasting this potentially avoidable consequence preoperatively can improve patient management, shared decision-making, and counseling.</p> Methods <p>Electronic medical records and radiographic measurements of patients with OPLL who underwent posterior cervical spine surgery between January 1, 2012 and May 1, 2022 were retrospectively collected and analyzed. The data were split into training (80%) and testing (20%) datasets. Two machine learning algorithms and a logistic regression model were trained and optimized using the training dataset, and their performances were assessed using the testing dataset. Important feature screening was conducted to identify the influential factors associated with the dependent variable, and model-based threshold effect analysis was further performed to derive data-driven cutoff values for continuous predictors, and category-specific effects were summarized for discrete variables. Additionally, three advanced neural network models were also preliminarily applied and explored in this study.</p> Results <p>In terms of predicting the postoperative C5P condition, the model with the best performance was the random forest model, with an AUC value of 0.9273, an accuracy rate of 0.8983, a precision rate of 0.8, a recall rate of 0.6667, and an F1 score of 0.7273. The results of the feature importance of the logistic regression, random forest, and XGBoost models indicated that the spinal canal occupancy rate (SCOR) was the most important predictive factor for postoperative C5P. The variable threshold effect analysis of the random forest model showed that when the SCOR was greater than 46%, there would be a significant risk of C5P.</p> Conclusion <p>Machine learning models showed good predictive power and provided information about the phenotypes of patients with OPLL most likely to develop C5P after surgical intervention. In general, the utilization of machine learning holds promises as a valuable tool for the management of patients with OPLL.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning and neural network algorithms for prediction of C5 palsy after posterior surgery for ossification of posterior longitudinal ligament

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Dingyu Jiang,
  • Ben Liu,
  • Yuxing Tan,
  • Binbin Ma,
  • Yunqi Wu,
  • Xiyang Wang,
  • Xiongjie Shen,
  • Yilu Zhang,
  • Zheng Liu

摘要

Purpose

C5 palsy (C5P) is one of the main postoperative complications of ossification of the posterior longitudinal ligament (OPLL). However, an accurate predictive tool for C5P remains to be developed. Forecasting this potentially avoidable consequence preoperatively can improve patient management, shared decision-making, and counseling.

Methods

Electronic medical records and radiographic measurements of patients with OPLL who underwent posterior cervical spine surgery between January 1, 2012 and May 1, 2022 were retrospectively collected and analyzed. The data were split into training (80%) and testing (20%) datasets. Two machine learning algorithms and a logistic regression model were trained and optimized using the training dataset, and their performances were assessed using the testing dataset. Important feature screening was conducted to identify the influential factors associated with the dependent variable, and model-based threshold effect analysis was further performed to derive data-driven cutoff values for continuous predictors, and category-specific effects were summarized for discrete variables. Additionally, three advanced neural network models were also preliminarily applied and explored in this study.

Results

In terms of predicting the postoperative C5P condition, the model with the best performance was the random forest model, with an AUC value of 0.9273, an accuracy rate of 0.8983, a precision rate of 0.8, a recall rate of 0.6667, and an F1 score of 0.7273. The results of the feature importance of the logistic regression, random forest, and XGBoost models indicated that the spinal canal occupancy rate (SCOR) was the most important predictive factor for postoperative C5P. The variable threshold effect analysis of the random forest model showed that when the SCOR was greater than 46%, there would be a significant risk of C5P.

Conclusion

Machine learning models showed good predictive power and provided information about the phenotypes of patients with OPLL most likely to develop C5P after surgical intervention. In general, the utilization of machine learning holds promises as a valuable tool for the management of patients with OPLL.