Leveraging Artificial Intelligence to Predict Craniospinal Surgery Needs in Achondroplasia Patients: A Multicenter Cohort Study

Introduction: Achondroplasia is the most prevalent form of skeletal dysplasia, with about 30% of affected individuals requiring surgical intervention in their lifetime. Despite established clinical guidelines, predicting the timing and necessity for craniospinal surgery in these patients remains challenging. This study employs artificial intelligence (AI) to predict craniospinal surgical needs in achondroplasia patients.

Methods: We conducted a multicenter cohort study analyzing data from pediatric patients diagnosed with achondroplasia at four major U.S. hospitals. The study utilized 39 clinical, imaging, and neurological variables to predict the necessity for craniospinal surgical intervention. Nine machine learning and deep learning models were applied alongside traditional statistical techniques. SHapley Additive exPlanations (SHAP) were also used to identify key predictive variables and interpret the AI models' results.

Results: Multivariate logistic regression analysis of 150 patients, diagnosed at an average age of 1.1 ± 2.2 years, identified significant predictors for surgical necessity including foramen magnum stenosis (FM stenosis), elevated intracranial pressure (ICP), and follow-up time. AI analysis demonstrated that the multilayer perceptron (MLP) model was the best-performing model, achieving an accuracy of 83% and an area under the curve of 86%. SHAP analysis, used to interpret the MLP model’s results, highlighted the important roles of FM stenosis, elevated ICP, hydrocephalus, and spinal stenosis as critical variables in increasing the likelihood of craniospinal surgery. Although follow-up time was also identified as an important variable, its impact on surgical necessity remained unclear. Consequently, Kaplan-Meier analysis, which treated follow-up time as the primary metric, indicated that achondroplasia patients with frontal bossing, Chiari malformation, FM stenosis, elevated ICP, and hydrocephalus are likely to require surgery earlier than others.

Conclusion : Integrating AI-based prediction models into clinical workflows can standardize and enhance decision-making processes for craniospinal surgery in patients with achondroplasia, potentially improving surgical management.