Machine learning can aid in the differential diagnosis of neurogenic thoracic outlet syndrome and carpal tunnel syndrome

Introduction

Symptoms related to neurogenic thoracic outlet syndrome (nTOS) and carpal tunnel syndrome (CTS) may overlap, leading to diagnostic uncertainty. In this study, we used a machine learning model to identify key predictors of nTOS by comparing it with CTS.

Methods

We reviewed records of patients who underwent surgical intervention for nTOS (n = 68) or CTS (n = 65). The machine learning model was developed using the scikit-learn library in Python, and a binary logistic regression model incorporating patient history and physical exam findings was developed to differentiate nTOS from CTS. Positivity rates of Tinel’s sign and the scratch collapse test (SCT) were compared using Agresti-Coull confidence intervals, chi-squared goodness-of-fit, and binomial tests.

Results

For diagnosis of nTOS, the baseline random forest model achieved 80.0% accuracy (F1-score: 0.76, area under the receiver operating characteristic curve: 0.91). After hyperparameter tuning, accuracy improved to 85.0% and precision reached 1.0, yielding a 7.7% gain in overall performance. Both Tinel’s sign and SCT in isolation were diagnostic of nTOS and CTS but could not differentiate between the 2 conditions. In both the baseline and optimized random forest model, the Roos/Elevated Arm Stress Test, body mass index, and duration of symptoms prior to surgery emerged as the most influential predictors of nTOS.

Conclusions

The random forest model predicted nTOS with up to 85% accuracy. SCT and Tinel’s tests in isolation could not distinguish between nTOS and CTS. Combining multiple clinical and demographic variables within a machine learning model yielded superior diagnostic accuracy for distinguishing nTOS from CTS.