A crosswalk for the conversion of scores between the Neck Disability Index (NDI) and the neck version of the Core Outcome Measures Index (COMI-neck).

introductionThe Neck Disability Index (NDI) and the neck version of the Core Outcome Measures Index (COMI-neck) are two commonly used self-rating outcome instruments in patients with cervical spinal disorders. The present study aimed to create a crosswalk between them to allow the scores of one to be interpreted in terms of the other.

methodsWe performed a secondary analysis of matched pairs of NDI and COMI-neck data collected in an international spine surgery registry from 624 patients (51 ± 10y; 57.2% female) at baseline, 625 patients at follow-up (FU) and 396 with both baseline and follow-up. In a random 80% of cases at each time-point (training sets), Pearson correlations between scores as well as Cohen’s kappa for agreement (κ) regarding achievement of literature-based minimal clinically important change (MCIC) scores on each instrument (NDI, ≥ 12 points; COMI, ≥ 2 points) were calculated, and linear regression models were built to predict the score on one instrument derived from that on the other; the latter were tested for accuracy in the remaining 20% of the corresponding samples (test sets).

resultsAll pairs of measures were significantly positively correlated (80% training set: baseline, r = 0.67; FU, r = 0.84; change-scores, r = 0.73). MCIC was achieved in 59.8% patients for NDI and 65.5% for COMI-neck, with 79.1% agreement on an individual basis (κ = 0.56, 95% CI 0.46–0.65). NDI was predicted as COMI-neck × 6.545 – 2.852 (at baseline), COMI-neck × 6.134 + 2.157 (at FU) and COMI-neck × 5.343 + 1.330 (for change score); COMI-neck was predicted as NDI × 0.069 + 4.219 (baseline), NDI × 0.115 + 0.946 (FU), and NDI × 0.098 + 1.276 (change score). Robust standard errors for the regression slopes and intercepts (group error) were low, but root mean square (RMS) residuals (reflecting individual error) were relatively high. Intraclass correlation coefficients between derived and actual scores were 0.50–0.84.

conclusionThe crosswalk was validated for use at the group level, with low errors, meeting standards justifying the pooling of data for use in multi-centre studies and meta-analyses. However, as previously reported for other outcome instruments, the crosswalk lacked the necessary precision for reliable individual-level assessment, e.g. in the monitoring of patients over time using the alternate instrument.