Back and neck pain are the primary musculoskeletal complaints responsible for absenteeism from work as well as high healthcare costs. A large proportion of back and neck pain sufferers still experience pain at 12 months and most patients experience relapses. Although musculoskeletal complaints are the second most common reason people take sick leave in the UK, guidelines produced by various bodies both in the UK and abroad for managing back and neck pain have failed to significantly reduce work disability. Self-report subjective outcome measures are being used increasingly in healthcare to assess the quality of care received by patients. However, despite years of research, there is no consensus in the literature as to how one categorises improved patients from those who have not improved using these outcome measures. Moreover, although researchers have investigated predictors of outcome for back and neck pain patients undergoing manual therapy, these is a lack of consensus about whether or not it is possible to predict outcomes in patients, and if it is, which factors consistently do so.
Consecutive new patients with back and neck pain presenting at a private chiropractic practice in Bristol completed a questionnaire consisting of a pain diagram and a self-report outcome measure, namely the Bournemouth Questionnaire (BQ). This questionnaire consists of seven questions which assess the commonly-measured domains of the biopsychosocial model of musculoskeletal conditions. This questionnaire is preceded by demographic information on the patient and clinical questions about the patient’s complaint. For the purposes of reassessment on the 4th/5th treatment and the 10th treatment, a post-treatment BQ, which includes a Patient Global Improvement Scale (PGIS), was completed by the patients. Three methods were used to determine ‘improvement’ in patients, (i) the direct method, using the PGIS, (ii) the anchorbased method, comparing total BQ change scores with the response on the PGIS as the anchor, and (iii) the distribution-based method, using 0.5 of the standard deviation of the group change score as the cut-off for ‘improvement’. All potential predictor variables that were significantly associated with self-reported ‘improvement’ were entered into multivariable regression analysis to determine final predictive models consisting of predictors independently associated with the outcome. The sensitivity and specificity of the model was calculated to determine the ability of the model to discriminate between improved and non-improved patients. To calculate the diagnostic accuracy of the model, the area under the Receiver Operator Characteristic Curve (AUC) was calculated.
The clinical and demographic characteristics were similar to other back and neck research populations. There was moderate to good agreement between the three methods of defining improvement in these cohorts with best agreement between the anchor and distribution-based methods. As a result the anchor-based method was used to identify patients who had improved. In the back pain cohort, the final predictive models had good sensitivity (86.1% at treatment 4/5 and 92.4% at treatment 10) for correctly identifying improvement but less specificity (48.5% at treatment 4/5 and 48.9% at treatment 10) for correctly identifying nonimprovers. The discriminative ability of the models (AUC) was 0.75 at treatment 4/5 and 0.83 at treatment 10. In both models a short history of pain, a smaller area shaded on the pain diagram and a total BQ score of over 30/70 were identified as predictors of outcome. At treatment 4/5 patients who had good treatment expectation and rated their general health as good were also associated with improvement. At treatment 10 inappropriate markings on the pain diagram and regular alcohol consumption were also associated with improvement.
In common with back pain, the predictive model in the neck pain cohort included a total baseline BQ score over 30/70 as associated with the outcome. At treatment 4/5, in common with the back pain, patients improved if they had a short pain history, good general health perception and shaded a smaller area on the pain diagram. At treatment 10 neck pain patients who smoked were associated with improvement. The final predictive model for neck pain patients had good sensitivity (77.8%) and moderate specificity (66.5%) at treatment 4/5, and the same sensitivity and specificity (67%) at treatment 10. The discriminative ability of the models (AUC) at treatment 4/5 and 10 was 0.76 and 0.71 respectively.
Being able to distinguish between and identify patients at an early stage with good and poor prognosis gives the clinician the opportunity to direct treatment approaches and manage patients more effectively. This study has shown that it is possible to identify characteristics of patients and their condition at baseline that can predict outcome at a later stage in their treatment. As might be expected, patients with more acute conditions, less severe pain and disability and less area in pain are more likely to improve. It is believed that this is the first time the area shaded on the pain diagram has been identified as a predictor of outcome.
|Date of Award||Oct 2011|
|Supervisor||Graham Mills (Supervisor), Jennifer E. Bolton (Supervisor) & Ann Dewey (Supervisor)|