Development, internal and external validation of a prognostic model for symptom dissatisfaction among older adults with a new episode of back pain

Objectives The objective was to develop, internally and externally validate a prognostic model for symptom dissatisfaction, assessed by the patient acceptable symptom state (PASS), for older adults (≥55 years) seeking primary care for a new episode of back pain. Design Development, internal and external validation of a prognostic model using data from two prospective cohort studies with a 1-year follow-up was conducted. Participants and setting The Norwegian cohort (n=452) was used for model development and internal validation. External validation was conducted using the Dutch cohort (n=675). Outcome measures and method The outcome was defined as symptom dissatisfaction based on the PASS at 12 months follow-up. A logistic regression model was developed using backward selection and internally validated using 200 bootstrap samples. External validation included recalibration of the model intercept and slope. Model performance was measured using Nagelkerke-R ², area under the curve (AUC) and calibration slope, calibration in-the-large (CITL) and calibration plots. Results At 12 months, ~55% reported dissatisfaction in both cohorts. The final model included disability, catastrophising, recent back pain episode, spinal rotation pain, baseline symptom satisfaction, symptom duration and recovery expectation as predictors. The internally validated model showed acceptable discrimination (AUC 0.75, 95% CI 0.71 to 0.78), R ² was 0.23, the calibration slope and CITL being 0.89 (95% CI 0.73 to 1.08) and 0.01 (95% CI −0.16 to 0.15), respectively. External validation performance after recalibration yielded AUC 0.68 (95% CI 0.65 to 0.70), slope 0.86 (95% CI 0.67 to 1.05) and CITL 0.08 (95% CI −0.01 to 0.16). Conclusions This prognostic model could be a useful tool for predicting PASS outcomes among older adults with back pain. The external validation results imply that more research is needed to optimise predictions.