Related Experiment Video
Updated: Sep 17, 2025
Author Spotlight: Scope of LE-ULBD as a Safe, Effective, and Minimally Invasive Approach to Treat Lumbar Spinal Stenosis
Published on: February 9, 2024
Development of a machine-learning model for patient satisfaction prediction in lumbar spinal stenosis surgery: A
Soya Kawabata1, Gen Miura2, Yuki Akaike3
1Department of Orthopedic Surgery, School of Medicine, Fujita Health University, Aichi, Japan.
Background:
Patient satisfaction is an essential metric for evaluating treatment outcomes for LSS, both for patients and for their primary physicians. However, the Zurich Claudication Questionnaire (ZCQ) is the only representative patient-reported outcome measure that evaluates satisfaction. To develop a model using machine learning to predict postoperative satisfaction among older patients with lumbar spinal stenosis (LSS) based on preoperative and postoperative scores of the Japanese Orthopedic Association Back Pain Evaluation Questionnaire (JOABPEQ).
Methods:
The training dataset was composed of time-course data of ZCQ and JOABPEQ scores from patients aged ≥65 years who underwent LSS surgery at a university hospital. The validation dataset included data from patients with LSS treated at a private orthopedic clinic. A linear support vector machine classifier was trained to predict achievement of a "Satisfied" state from preoperative and postoperative JOABPEQ scores. Internal validation was carried out via leave-one-out cross-validation, and external validation using a separate dataset to assess the accuracy, sensitivity, specificity, F1 score, and area under the receiver operating characteristics curve (AUROC). Variable importance was analyzed using model class reliance.
Results:
A total of 232 and 66 individuals were included in the training and validation datasets, respectively. The machine-learning model exhibited an accuracy of 0.72, sensitivity of 0.75, specificity of 0.69, and AUROC of 0.82. Psychological disorder and walking ability were identified through permutation importance analysis as key factors for satisfaction. External validation on an independent dataset demonstrated comparable accuracy (0.76), sensitivity (0.83), and AUROC (0.75), although the specificity decreased (0.42).
Conclusions:
The machine learning model presented here can predict the postoperative satisfaction score on the ZCQ from preoperative and postoperative JOABPEQ scores, highlighting its potential for broader application in clinical settings.
