Scholarlink

Purpose <p>Accurate prediction of postoperative pain can improve recovery quality and guide personalized pain management. This study aimed to develop and validate a machine learning (ML) model that predicts the presence of postoperative pain using biosignals recorded in the post-anesthesia care unit (PACU).</p> Methods <p>Adult patients who underwent surgery between January 2021 and December 2022 at Chungnam National University Hospital (CNUH) (<i>n</i> = 21,855) and Chungnam National University Sejong Hospital (CNUSH) (<i>n</i> = 2,356) in South Korea were included. Electrocardiography (ECG) and photoplethysmography (PPG) signals were continuously recorded during the postoperative period in the PACU. From these signals, heart rate variability (HRV) and surgical pleth index (SPI) features were extracted. These biosignal features, together with demographic variables, were used as input features for training the ML models, including logistic regression, support vector machine, multilayer perceptron, random forest, and XGBoost. The model was developed and internally validated using the CNUH dataset, while external validation was performed using the independent CNUSH dataset to assess generalizability.</p> Results <p>In the test dataset of 1068 patients, 961 reported postoperative pain. The best-performing model achieved an accuracy of 85.2%, an AUROC of 0.77, and average precision of 0.95, while external validation yielded an accuracy of 83.2% and an AUROC of 0.78. Feature importance analysis using SHapley Additive exPlanations (SHAP) indicated that the most influential predictors were the proportion of SPI values exceeding 50, the mean SPI, and the low-to-high frequency power ratio of HRV. Incorporating demographic features, such as age and sex, improved prediction accuracy by up to 5.91%. This improvement may be attributed to the mitigation of inter- and intra-individual variability inherent in biosignals, thereby enhancing the model’s stability and generalizability.</p> Conclusion <p>ML models incorporating ECG- and PPG-derived features demonstrated reliable prediction of postoperative pain across independent hospital cohorts, highlighting the potential of biosignal-based approaches for objective pain assessment in clinical practice.</p> Clinical trial number <p>Not applicable.</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

Predictive machine learning for postoperative pain using biosignals: a retrospective observational study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月15日

查看全文
  • Jieun Oh,
  • Dongheon Lee,
  • Minwoong Kang,
  • Chahyun Oh,
  • Seyeon Park,
  • Jiho Park,
  • Kyungsang Kim,
  • Boohwi Hong

摘要

Purpose

Accurate prediction of postoperative pain can improve recovery quality and guide personalized pain management. This study aimed to develop and validate a machine learning (ML) model that predicts the presence of postoperative pain using biosignals recorded in the post-anesthesia care unit (PACU).

Methods

Adult patients who underwent surgery between January 2021 and December 2022 at Chungnam National University Hospital (CNUH) (n = 21,855) and Chungnam National University Sejong Hospital (CNUSH) (n = 2,356) in South Korea were included. Electrocardiography (ECG) and photoplethysmography (PPG) signals were continuously recorded during the postoperative period in the PACU. From these signals, heart rate variability (HRV) and surgical pleth index (SPI) features were extracted. These biosignal features, together with demographic variables, were used as input features for training the ML models, including logistic regression, support vector machine, multilayer perceptron, random forest, and XGBoost. The model was developed and internally validated using the CNUH dataset, while external validation was performed using the independent CNUSH dataset to assess generalizability.

Results

In the test dataset of 1068 patients, 961 reported postoperative pain. The best-performing model achieved an accuracy of 85.2%, an AUROC of 0.77, and average precision of 0.95, while external validation yielded an accuracy of 83.2% and an AUROC of 0.78. Feature importance analysis using SHapley Additive exPlanations (SHAP) indicated that the most influential predictors were the proportion of SPI values exceeding 50, the mean SPI, and the low-to-high frequency power ratio of HRV. Incorporating demographic features, such as age and sex, improved prediction accuracy by up to 5.91%. This improvement may be attributed to the mitigation of inter- and intra-individual variability inherent in biosignals, thereby enhancing the model’s stability and generalizability.

Conclusion

ML models incorporating ECG- and PPG-derived features demonstrated reliable prediction of postoperative pain across independent hospital cohorts, highlighting the potential of biosignal-based approaches for objective pain assessment in clinical practice.

Clinical trial number

Not applicable.