Development and Validation of an Interpretable Machine Learning Model for Predicting Early Fusion After Anterior Cervical Discectomy and Fusion.

STUDY

DESIGN

Retrospective case-control study.

OBJECTIVES

This study aimed to develop and preliminarily validate a machine learning (ML) model for predicting the likelihood of early fusion (EF) after anterior cervical discectomy and fusion (ACDF) and to explore the influential factors.

SUMMARY OF BACKGROUND DATA: ACDF is a commonly performed procedure, where EF plays an important role in achieving favorable outcomes. However, EF varies substantially among patients, and reliable predictive approaches remain limited.

METHODS

We retrospectively analyzed 1039 surgical segments from 840 patients who underwent ACDF between 2013 and 2020. EF, defined as radiographic fusion within three months, was assessed using standard imaging criteria. Basic information, laboratory indicators, perioperative data, and radiologic parameters were collected. After multiple imputation and dimensionality reduction, nine ML algorithms were trained, evaluated, and compared. SHapley Additive exPlanations (SHAP) were applied for model interpretation.

RESULTS

Among the nine algorithms, stochastic gradient boosting (SGB) exhibited the highest predictive ability, with an AUC of 0.884 in the training set and 0.830 in the testing set. SHAP analysis indicated that preoperative functional spinal unit (FSU) range of motion (ROM), ΔFSU height, fasting plasma glucose (FPG), calcium (Ca), low-density lipoprotein cholesterol (LDL-C), surgical type, age, and femoral bone mineral density (BMD) were the most influential factors. Higher preoperative FSU ROM, FPG, LDL-C, age, and two-level surgery were associated with a lower probability of EF, whereas optimal ΔFSU height and higher Ca and femoral BMD were favored EF.

CONCLUSIONS

This exploratory study established an ML-based approach for predicting EF after ACDF, with the SGB algorithm showing relatively strong predictive performance. The identified influential factors may provide preliminary insights for individualized clinical assessment and perioperative management, warranting further validation in multicenter settings.