Abstract
INTRODUCTION: Achondroplasia, the most common skeletal dysplasia, is frequently associated with life-threatening craniospinal complications, resulting in a substantial neurosurgical burden. Nearly 30% of affected patients require neurosurgical intervention. However, the disorder’s clinical heterogeneity and the absence of standardized decision-making criteria make it difficult to predict surgical need, often resulting in reliance on subjective clinical judgment. METHODS: A multicenter retrospective cohort of 150 pediatric patients with achondroplasia (age 1.1 ± 2.2 years) from four major U.S. hospitals (2006–2022) was analyzed. Thirty-three demographic, clinical, and imaging variables were included in a multivariate regression analysis. in parallel, machine learning and deep learning models were applied to assess predictor importance and enhance classification accuracy across three categories: no surgery, cranial surgery, and spinal surgery. Additionally, Kaplan-Meier analyses were used to evaluate temporal trends and identify predictors associated with earlier surgical intervention. RESULTS: The multilayer perceptron model outperformed other classifiers (accuracy: 77%; AUC: 0.76-0.78). SHAP analysis identified FM stenosis, hydrocephalus, frontal bossing, macrocephaly, sleep disturbances, and family history as key predictors for cranial surgery, while spinal stenosis, Chiari malformation, back pain, FM stenosis, and family history were predictive of spinal surgery. Several of these associations were not fully captured by multinomial regression. Regression confirmed FM stenosis, hydrocephalus, and family history as significant predictors for cranial surgery, and spinal stenosis, Chiari malformation, and back pain for spinal surgery. Kaplan-Meier analysis showed earlier cranial surgeries in patients with FM stenosis, frontal bossing, and hydrocephalus, and earlier spinal surgeries in those with spinal stenosis and Chiari malformation. CONCLUSIONS: This multicenter study demonstrates the feasibility of identifying key predictors that inform craniospinal surgical decision-making, supporting standardized, data-driven risk assessment and timely intervention across varied clinical settings.