Development and Validation of an Algorithm for Item Reduction of the International Standards for Neurological Classification of Spinal Cord Injury Examination to Determine Level and Severity of SCI.

Background: In 2020, a first, expedited version of the International Standards for Neurological Classification of Spinal Cord Injury (E-ISNCSCI-V1) was proposed for determination of neurological level of injury (NLI) and American Spinal Injury Association Impairment Scale (AIS) classifications.

Objectives: This work describes assessment of E-ISNCSCI-V1 classification accuracy and the development and data-based validation of an ISNCSCI Item Reduction Algorithm (IIRA).

Methods: Classification accuracy for E-ISNCSCI-V1 examination shortcut options was assessed with automated analysis of 7026 full ISNCSCI examinations. Rules for the IIRA were iteratively adjusted to optimize the balance between omitting exam items and minimizing misclassification errors, and then it was validated through classification of 100 full ISNCSCI exams.

Results: If S1 findings are substituted for anorectal exam findings as proposed for E-ISNCSCI-V1, the error rate for AIS is 10%, with a high error rate (45%) for classifying true AIS B. The IIRA, which begins with full motor testing, followed by limited sensory testing required an average of 31% (42/134) of the full ISNCSCI exam items, with a 2% error rate for NLI and no AIS errors.

Conclusion: The previously proposed E-ISNCSCI-V1, which included an option to substitute S1 findings for anorectal exam findings, is not recommended due to AIS error rate. The IIRA provides a standardized option for a shortened examination classifying NLI and AIS with high accuracy. It will serve as a basis for version 2 of the E-ISNCSCI.