A Novel Machine Learning Model for Predicting Traumatic Brain Injury Rehabilitation Outcomes with Social and Clinical Considerations

Background and/or Objectives: Traumatic brain injury (TBI) recovery is influenced by injury severity, rehabilitation access, and socioeconomic status (SES). AI-driven predictive models often overestimate functional gains by failing to account for real-world barriers. Based on these findings, we developed a machine learning (ML) model to improve TBI outcome predictions by integrating clinical, functional, and socioeconomic factors to create individualized rehabilitation forecasts.

Design: Development and validation of an ML-based predictive model for TBI rehabilitation outcomes.

Setting : Multi-institutional retrospective dataset of TBI patients in inpatient and outpatient rehabilitation settings.

Participants : TBI patients with varying injury severity and SES backgrounds.

Interventions: A supervised ML model trained on demographic data, injury characteristics, rehabilitation intensity, functional scores (Section GG Self-Care scale), and SES-related variables (insurance status, geographic access, caregiver support).

Main Outcome Measures: Prediction accuracy for functional independence compared to actual patient outcomes, assessed using mean absolute error (MAE) and paired t-tests.

Results: Our ML model demonstrated improved predictive accuracy over conventional AI forecasts, reducing MAE in functional outcome predictions by 32% (p < 0.01). Unlike standard models, our tool dynamically adjusted projections based on SES constraints, refining estimates for dressing, transfers, and ambulation. The model also identified key SES variables that significantly impacted recovery trajectories.

Conclusions: By incorporating SES-adjusted predictive modeling, our ML tool enhances individualized TBI rehabilitation planning. Future applications include integrating real-time rehabilitation adjustments, optimizing resource allocation, and improving equity in TBI recovery for underserved populations.