Related Experiment Videos
Explainable Machine Learning for Perioperative Risk Stratification of Radiographic Adjacent Segment Degeneration
Ruizhang Yao1,2, Dongfan Wang1,2, Peng Cui1,2
1Department of Orthopedics & Elderly Spinal Surgery, Xuanwu Hospital of Capital Medical University, National Clinical Research Center for Geriatric Diseases, Beijing, China.
Spine
|August 4, 2026
Summary
This study developed an explainable machine-learning model to predict radiographic adjacent segment degeneration (ASDeg) after lumbar fusion. The random forest model accurately stratified risk using key perioperative variables.
Area of Science:
- Spine Surgery
- Machine Learning in Healthcare
- Radiographic Degeneration
Background:
- Adjacent segment degeneration (ASDeg) is common after lumbar fusion, potentially preceding symptomatic adjacent segment disease (ASDis).
- Current risk assessment for ASDeg is limited by heterogeneous factors, poor interpretability, and lack of external validation.
- Machine learning offers potential for improved risk stratification by integrating diverse patient data.
Purpose of the Study:
- To develop and externally validate an explainable machine-learning framework for perioperative risk stratification of radiographic ASDeg.
- To identify key predictors of ASDeg following short-segment lumbar fusion.
Main Methods:
- Retrospective cohort study with internal (n=570) and external (n=150) validation cohorts.
- Feature selection using LASSO, RE-RFE, and Boruta; comparison of logistic regression, RF, XGBoost, LightGBM, and MLP algorithms.
- Model performance evaluated using discrimination, calibration, P-R analysis, and DCA; SHAP for interpretability.
Main Results:
- Radiographic ASDeg occurred in 37.2% of the internal cohort.
- Five key perioperative variables identified: preoperative intervertebral space height (ISH), postoperative pelvic incidence-lumbar lordosis (PI-LL) mismatch, frailty, Coflex implantation, and preoperative WOMAC.
- The Random Forest (RF) model demonstrated superior performance (AUROC 0.782 internal, 0.749 external); SHAP highlighted preoperative ISH as the strongest predictor.
Conclusions:
- An externally validated, interpretable RF model effectively stratifies radiographic ASDeg risk after short-segment lumbar fusion.
- The model integrates accessible perioperative variables, aiding in individualized imaging follow-up.
- This framework provides a basis for future research on symptomatic ASDis and revision surgery.