Development of machine learning models for predicting patient-perceived benefit following lumbar disc herniation or...

Spinal surgery outcomes vary significantly among patients, underscoring the need for objective tools to guide clinical decision-making. While previous prediction models have focused on functional and pain-specific outcomes, global patient-reported measures such as the Global Perceived Effect (GPE) remain underexplored. This study aimed to develop and temporally validate machine learning models to predict patient's GPE 12 months after lumbar disc herniation and spinal stenosis surgery.

This registry-based study used data from the Norwegian Registry for Spine Surgery. The dataset was temporally split for model development (2007-2017) and validation (2018-2021). Six supervised machine learning models, including XGBoost, Gradient Boosting, Random Forest, Multilayer Perceptron (MLP), Decision Tree, and K-Nearest Neighbors, were used to predict dichotomized GPE outcomes (success vs. non-success). Model performance was evaluated through discrimination, calibration, and decision curve analysis.

Analyses included 13,029 patients operated for disc herniation and 18,058 for spinal stenosis. For disc herniation, the MLP model achieved the highest performance at temporal validation, with a C-statistic of 0.72 (95% CI 0.71-0.74) and good calibration. For spinal stenosis, XGBoost performed best, with C-statistic 0.67 (95% CI 0.66-0.68) and good calibration. Key predictors included pain duration, prior surgery, anxiety/depression, education level, and healthcare setting. The models demonstrated consistent clinical utility through decision curve analysis.

Using nationwide registry data, we developed and temporally validated machine learning models to predict patient-perceived benefit one year after lumbar spine surgery. The disc herniation model showed possibly useful discrimination and good calibration, supporting its potential to inform clinical decision-making.