Model development of a multivariable prediction model for long-term work-related neck disability among high risk...

Clinical prediction models have been identified as a research priority to identify individuals at risk of persistent musculoskeletal disability. As neck pain is a leading cause of disability among workers exposed to cumulative, repetitive and sustained force, model development in such populations is needed. This study describes the development of a clinical prediction model to predict moderate-to-severe neck disability, measured by the neck disability index (NDI≥22%) at 12-months. Sonographers working in Australia and New Zealand (n = 262) with and without neck disability completed questionnaires evaluating demographic, ergonomic and psychosocial factors. Individual scores, rather than summary scores, were entered into LASSO (Least Absolute Shrinkage and Selection Operator) regression with 5-fold cross validation to identify predictors of persistent disability. The penalised analysis identified two models: an optimal model with 18-items and a simpler model with five items. Following expert consensus, one item (medical history) was removed from the optimal model. Multivariate logistic regression models were then developed using the full sample, demonstrating excellent discrimination for the optimal model (area under the curve (AUC) 0.922, 95% CI 0.875-0.968, sensitivity 79% and specificity 91%) and good discrimination for the 5-item model (AUC 0.894, 95% CI 0.832-0.956). The models suggest that work- or pain-related beliefs and the impact of neck pain on work, headaches or sleep, but not ergonomic factors are the most important predictors of disability in workers at high risk of non-traumatic injuries. However, these models are not definitive and require external validation in larger samples to confirm model calibration and predictive accuracy.