Abstract
Introduction
Patient reported outcome measures (PROM) are now a mandatory part of the assessment of clinical results in comparative effectiveness research. The 10-item neck disability index (NDI) is the most commonly used disease specific PROM in studies of patients with cervical spinal disorders. It is well known, easy to administer and score (10 items—answers added together and multiplied by 2, to provide a score). Recent studies have shown significant deficiencies in the psychometrics of the standard NDI, including poor coverage (large floor effect), multivariance, and most concerning, extremely poor raw score to measure correlation—which means using the additive score is concerning, particularly with standard statistical testing. By excluding two questions, or five questions, an NDI-8 and NDI-5 can be constructed with much-improved psychometrics and decreased clinical burden, yet this has not been evaluated in a formal spine medical practice and is not used in clinical research. The goal of this study is to clarify the psychometric properties of the standard NDI-10, in comparison with the new NDI-8 and NDI-5, in search of an improved PROM for neck patients.
Material and Methods
Patient demographics and the NDI were collected on 714 patients with neck ± upper extremity pain. Using Rasch analysis, the NDI-10, -8, and -5 were assessed for psychometric properties, including overall item fit, variance, coverage, and raw score to measure correlation.
Results
Data were assessed from 714 consecutive patients (53% males, 86% whites) with average age being 55 years, seen at a university-based spine center. The NDI-5 and NDI-8 were highly correlated at 0.95, had excellent fit, similar to the NDI-10, and less variance than the NDI-10. Like the NDI-10, coverage for the NDI-8 and -5 were poor, with very high floor effects of 39.36 and 39.22% (compared with 35.5% in the NDI-10). Ceiling effects were acceptable for both measures. Most importantly, the raw score to measure correlation of the NDI-8 and NDI-5 are very high at 0.94 each, while the NDI-10 correlation was poor, at 0.02.
Conclusion
While the standard 10-item NDI is of questioned value, the NDI-5 and NDI-8 show improved psychometrics, in particular the raw score to measure correlation. With fewer items the NDI-5 decreases patient and clinic burden without significantly effecting coverage, and is therefore the preferred measure. Future work should be concerned with improving coverage with the addition of items at the extremes of disability, particularly low disability items.