Development and Validation of Diagnostic Models for Cervical Nerve Root Involvement Based on Items From the Patient Interview and Clinical Examination

Abstract

OBJECTIVE: To develop and internally validate diagnostic models for cervical nerve root involvement based on patient interview and clinical examination items. DESIGN: Diagnostic predictive modelling study. METHODS: People with a suspicion of cervical nerve root involvement (i.e., radicular pain and/or radiculopathy) (N=134) were included. Three diagnostic models (i.e., patient interview items alone, clinical examination alone, and combined patient interview plus clinical examination) were developed using multivariable logistic regression analyses. For internal validation, we performed bootstrapping techniques (250 repetitions). The diagnostic accuracy (Area Under the Curve (AUC)) and explained variance (Nagelkerke’s r-squared) of the models were assessed. An AUC of 0.7 or higher was considered adequate. RESULTS: The patient interview model consisted of two items and showed an explained variance of 0.23 and an AUC of 0.74 (95%CI: 0.66-0.81) after bootstrapping. The clinical examination model consisted of 2 items and had an explained variance of 0.29, and an AUC of 0.77 (95%CI: 0.69-0.85) after internal validation. The combined model had an explained variance of 0.38 and an AUC of 0.82 (95%CI: 0.75-0.89) after bootstrapping and consisted of the Spurling test (odds ratio (OR): 8.0 (95%CI: 3.1-20.4)), ‘Arm pain worse than neck pain’ (OR: 4.8 (95%CI: 1.9-11.8)) and the patient-reported ‘Presence of paraesthesia and/or numbness’ (OR: 2.8 (95%CI: 1.0-7.8)). CONCLUSIONS: The combined model showed the best diagnostic accuracy to determine the likelihood of cervical nerve root involvement. External validation is required before implementing any diagnostic model.