Back to 2026 Abstracts
Designing A Decision-making Tool For Patients With Carpel Tunnel Syndrome Using Neural Networks
Moaath Saggaf, MD, PhD, Jana Dengler, MD, Heather Baltzer, MD, Dimitri Anastakis, MD.
University of Toronto, Toronto, ON, Canada.
PURPOSE:This study aimed to design a decision aid for carpal tunnel syndrome (CTS) using artificial intelligence (AI). We hypothesized excellent prediction of patient-reported improvement in CTS, measured by the area under the receiver operating characteristic curve.
METHODS:We built a neural network from a recent prospective comparative study to estimate the outcomes of surgical and nonsurgical interventions for CTS. We measured CTS severity using validated tools. The outcome was classified based on improvement above or below the minimal clinically important difference (MCID) in CTS severity. To improve the interpretability of the model, we added hypothetical observations to understand its behaviour. We used the Bayesian approximation method to account for uncertainty. We obtained a distribution for each prediction and constructed 95% confidence intervals for the calculated odds ratio (OR).
RESULTS: There were 85 patients who completed the study: surgery (n=49); splinting (n=36). The mean follow-up period was 10 months. Following the internal validation, the model accurately predicted the outcomes in (17/18) 94.4% of the cases. The area under the curve was 94%, indicating excellent performance. The only misprediction was explained by a floor effect. Surgery (OR=5.56, 95% CI: 5.37-5.75), high abductor pollicis brevis Medical Research Council grade (OR=2.36, 95% CI: 2.29-2.43), and high baseline CTS severity scores (OR=4.49, 95% CI: 4.32-4.69) predicted improvement following CTS treatment. The model provided individualized predictions for future patients.
CONCLUSION:This AI model can predict personalized patient-reported outcomes in CTS with excellent performance and will be used as a decision aid for CTS patients.
Back to 2026 Abstracts
