The association between different outcome measures and prognostic factors in patients with neck pain: a cohort study

Abstract

Background Health domains like pain, disability, and health-related quality of life are commonly used outcomes for musculoskeletal disorders. Most prognostic studies include only one outcome, and it is unknown if prognostic factors and models may be generic across different outcomes. The objectives of this study were to examine the correlation among commonly used outcomes for neck pain (pain intensity, disability, and health-related quality of life) and to explore how the predictive performance of a prognostic model differs across commonly used outcomes.

Methods We conducted an observational prospective cohort study with data from patients with neck pain aged 18–84 years consulting Norwegian chiropractors. We used three different outcomes: pain intensity (Numeric Pain Rating Scale), the Neck Disability Index (NDI), and health-related quality of Life (EQ-5D). We assessed associations between change in outcome scores at 12-weeks follow-up with Pearson’s correlation coefficient. We used multivariable linear regression models to explore differences in explained variance and relationship between predictors and outcomes.

Results The study sample included 1313 patients and 941 (72%) completed follow-up at 12 weeks. The strongest correlation was between NDI and EQ-5D (r = 0.57) while the weakest correlation was between EQ-5D and pain intensity (r = 0.39). The correlation between NDI and pain intensity was moderate (r = 0.53) In the final regression models, the explained variance ranged from adjusted R2 of 0.26 to 0.60, highest with NDI and lowest with pain intensity as outcome. The predictive contributions of the included predictors were similar across outcomes. Among the investigated predictors, pain patterns and the baseline measure of the corresponding outcome measure contributed the most to explained variance across all outcomes.

Conclusions The highest correlation was found between NDI and EQ-5D and the lowest with pain intensity. The same prognostic model showed highest predictive performance with NDI as outcome and poorest with pain intensity as outcome. These results suggest that we need more knowledge on the reasons for the differences in predictive performance variation across outcomes.