Model-Based Prediction of Motor Scores From Sensory Scores in the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI): Implications on Motor Levels in Segments Without Clinically Testable Key Muscles.

BACKGROUND

In the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI), motor levels are inferred from sensory levels for high cervical, thoracic, and low sacral injuries, as key muscles are only assessed in upper and lower extremities. This is known as the "motor follows sensory level" rule.

OBJECTIVES

To develop regression models for estimating motor scores from sensory scores in segments without clinically testable key muscles and to validate the consensus-based "motor follows sensory level" approach.

METHODS

A total of 6940 ISNCSCI examinations from the European Multicenter Study about Spinal Cord Injury were reviewed. Multiple linear and random forest regression models were trained on scores in clinically testable segments to predict motor from sensory scores of the same spinal segment and side. Models based on ipsilateral light touch or pinprick scores alone, as well as all bilateral sensory scores, were also evaluated. Predicted motor scores were used to recalculate motor levels for the segments without clinically testable key muscles and compared to the true motor levels.

RESULTS

The ipsilateral regression models showed minimal differences (R 2 0.64-0.65; RMSE 1.34). Normal motor scores were predicted only for normal sensory function; in the linear model, this was captured by the equation: motor score = 0.18 + 1.22 * light touch score + 0.96 * pinprick score. Model-based motor levels were shifted caudally 0.18 segments (linear regression) and 0.32 segments (random forest regression).

CONCLUSION

As models predict normal motor function only for normal sensory scores, predicted motor levels deviate only marginally, supporting the "motor follows sensory level" rule.