Optimising outcome prediction in primary care: Use of longitudinal data in prognosis research

Talk Code: 

The problem

Musculoskeletal conditions are a common problem in primary care, and account for around 14% of UK general practice consultations. The prognosis of these conditions is variable with around 30% of patients still reporting symptoms a year later. Research investigating prognosis in musculoskeletal conditions has only been moderately successful in predicting which patients have a low probability of recovery from their pain and functional limitations. When estimating future outcome the majority of prognostic studies only include information on factors assessed at baseline, despite the fact that re-consultation for musculoskeletal pain is common in clinical practice. The aim of this study was to explore if repeated short-term measurements of brief pain scores or prognostic factors (monitoring) can contribute to better prediction of long-term functional outcomes in primary care patients with musculoskeletal pain.

The approach

We obtained data on four cohort studies which included patients with back or shoulder pain, assessed functional limitations as the long-term outcome (6 or 12 months) and included baseline and short-term (within two months) assessments of pain and prognostic factors. We firstly compared the predictive performance of baseline pain, disability and psychological factors versus repeated short-term assessments of these factors in predicting long-term disability improvement (a reduction of 30% or more in disability score). Secondly, we investigated the predictive performance of previously validated multivariable prediction models compared to baseline pain score only. We finally compared predictive performance of potential clinical scenarios to help identify those patients who are likely to have a poor prognosis and decide on referral or further treatment. Logistic regression analyses were used to investigate the associations between predictors and outcome, with the C-statistic and sensitivity and specificity calculations being used to assess predictive performance.


Overall, a repeat short-term assessment of either pain or disability was more predictive of long-term disability improvement compared to baseline scores only or short-term changes in pain or disability. A repeated assessment of psychological factors showed little improvement in predictive performance over a baseline assessment only. Including more detailed prognostic information at baseline in the form of a multivariable prognostic model was more predictive of recovery (C-statistic 0.71 (95% CI 0.66 to 0.75)) compared to baseline (C-statistic 0.61 (95% CI 0.56 to 0.66)). The optimal clinical scenario, although still of limited predictive performance, included assessment of a multivariable prognostic score at baseline (presenting consultation) to identify people at either high or low risk of poor outcome followed by a repeat assessment (follow-up consultation) in those at intermediate risk (sensitivity 0.64, specificity 0.65). Consequence: Obtaining more detailed information when a patient first consults, and inviting patients back for a follow-up consultation, can result in more accurate prediction of the risk of future disability, and support referral decisions.


  • Gemma Mansell, University of Oxford, Oxford, UK
  • Danielle van der Windt, University of Oxford, Oxford, UK
  • George Peat, University of Oxford, Oxford, UK
  • Daniel Lasserson
  • Kate Dunn, University of Oxford, Oxford, UK
  • Kelvin Jordan, University of Oxford, Oxford, UK