Related Experiment Video
Updated: Jul 31, 2025
An Experimental Paradigm for the Prediction of Post-Operative Pain PPOP
Published on: January 27, 2010
Construction and Validation of Machine Learning Algorithms to Predict Chronic Post-Surgical Pain Among Patients
Zeping Yan1, Mengqi Liu2, Xiaoli Wang2
1School of Nursing and Rehabilitation, Cheeloo College of Medicine, Shandong University, Jinan, China; University of Health and Rehabilitation Sciences, Qingdao, China; Beijing Bo'ai Hospital, China Rehabilitation Research Center, Beijing, China.
Background:
Chronic post-surgical pain (CPSP) is a common but undertreated condition with a high prevalence among patients undergoing total knee arthroplasty (TKA). An effective model for CPSP prediction has not been established yet.
Aims:
To construct and validate machine learning models for the early prediction of CPSP among patients undergoing TKA.
Design:
A prospective cohort study.
Participants/Subjects:
A total of 320 patients in the modeling group and 150 patients in the validation group were recruited from two independent hospitals between December 2021 and July 2022. They were followed up for 6 months to determine the outcomes of CPSP through telephone interviews.
Methods:
Four machine learning algorithms were developed through 10-fold cross-validation for five times. In the validation group, the discrimination and calibration of the machine learning algorithms were compared by the logistic regression model. The importance of the variables in the best model identified was ranked.
Results:
The incidence of CPSP in the modeling group was 25.3%, and that in the validation group was 27.6%. Compared with other models, the random forest model achieved the best performance with the highest C-statistic of 0.897 and the lowest Brier score of 0.119 in the validation group. The top three important factors for predicting CPSP were knee joint function, fear of movement, and pain at rest in the baseline.
Conclusions:
The random forest model demonstrated good discrimination and calibration capacity for identifying patients undergoing TKA at high risk for CPSP. Clinical nurses would screen out high-risk patients for CPSP by using the risk factors identified in the random forest model, and efficiently distribute preventive strategy.
