Researchers are using AI to predict rehabilitation outcomes in children after spinal fusion surgery for adolescent scoliosis (AIS).
This could reshape how surgeons approach patient counselling and decision-making, paving the way for more personalised surgical planning.
AIS, the most prevalent form of scoliosis, often requires surgical intervention.
While the US Scoliosis Research Society-22R (SRS-22R) tool has long provided insights into patient-reported outcomes (PROs), its lack of correlation with radiographic correction rates has made it difficult for surgeons to integrate subjective outcomes into planning.
This gap has prompted researchers to explore the predictive power of AI.
The study analysed a unique multi-site cohort of 455 paediatric AIS patients treated at Shriners Children’s Hospitals.
Leveraging 171 pre-operative clinical features, researchers developed and tested six machine-learning models to predict postoperative outcomes, including responses to the SRS-22R questionnaire.
The predictions were evaluated at three key post-operative milestones: six months, one year and two years.
The team also incorporated ‘responsible AI’ practices, including model confidence calibration and measures to mitigate health disparities, to ensure clinical applicability and uphold ethical standards.
The top-performing AI model achieved an area under the receiver operating curve (AUROC) of 0.86, 0.85, and 0.83 for predicting SRS-22R outcomes at six months, one year, and two years, respectively. This demonstrates strong predictive capabilities for short- and long-term recovery trajectories.
Explainability analysis revealed that parameters such as sagittal alignment (lordosis, sagittal vertical axis) and patient-reported self-image played a more significant role in predicting outcomes than traditional radiographic metrics like Cobb angles.
The AI system successfully predicted outcomes based on minimal clinically significant differences (MCID), offering a nuanced view of patient recovery that aligns with clinical priorities.
This AI-enabled tool represents a significant step forward in personalised medicine for AIS patients.
The system supports pre-operative counselling, shared decision-making, and tailored surgical planning by providing robust predictions of post-operative rehabilitation outcomes. Its fairness and reliability make it a promising addition to real-world clinical workflows.
Researchers suggest that greater emphasis on sagittal parameters and self-image in decision-making may enhance outcomes for AIS patients.
The study highlights AI’s increasing role in improving patient outcomes through data-driven insights.
