1253 Machine Learning Approaches for Imputation and Validation of Missing AIS Grades in the Spinal Cord Injury Data Model Systems (SCIMS) Database

INTRODUCTION: The American Spinal Injury Association Impairment Scale (AIS) at admission is an important predictor of outcomes following spinal cord injury (tSCI). Nearly 80% of records in the Spinal Cord Injury Model Systems (SCIMS) database—a multicenter, prospective database of SCI patients—lack admission AIS grades. Accurate imputation of this missing data could enable more robust analyses and insights into SCI recovery. METHODS: We included 14,110 SCI patients from the SCIMS database (1988-2020). Five machine learning algorithms—Random Forest (RF), Linear Discriminant Analysis, K-Nearest Neighbors, Naive Bayes, and Support Vector Machines—were compared for imputation accuracy using 5-fold cross-validation. The highest accuracy model (RF) was applied to the entire dataset. We validated the imputed AIS grades by predicting discharge Functional Independence Measure (FIM) scores (range 13-91) with simple and multiple linear regression models on a 1:1 propensity-score-matched cohort (n=5,828). Model performance was compared using differences in RMSE with bootstrapped confidence intervals. RESULTS: The full cohort contained a representative distribution of AIS grades (46.0% A, 14.7% B, 19.4% C, 19.8% D), and the propensity-matched cohort characteristics were well-balanced. RF demonstrated the highest validation accuracy, 81.7%. Predictive models showed no significant differences between models using true vs. imputed AIS grades, with the 95%CI for RMSE being [-0.18,0.85] for simple and [-0.28,0.77] for multiple regression models. The coefficients of AIS grades were not significantly different between models with true vs. imputed values. CONCLUSIONS: A data-driven approach to imputation resulted in a robust method for imputing admission AIS that demonstrates clinical validity in the SCIMS database. This approach extends the utility of this longitudinal database and, with validation, may provide a framework for other SCI databases.