Abstract
Background: Knee osteoarthritis (OA) is a prevalent disease which contributes to significant pain and disability in older individuals. There is increasing evidence that physiotherapy, in the form of exercise or manual therapy, is an effective intervention. However, not all knee OA patients will benefit from physiotherapy treatment, and there are numerous other treatment options. Matching effective interventions to individuals who are most likely to benefit is desirable.
Aim: To identify predictors of successful response to physiotherapy treatment, evaluated at one year, for individuals with knee OA.
Methods: This observational cohort study was nested within a randomised controlled trial investigating effectiveness of physiotherapy for individuals with lower limb OA. Selection of variables for a standardised baseline assessment was informed by a systematic review identifying predictors of knee OA progression, and an inter-rater reliability study. A scale was developed to provide a valid and reliable method for assessing irritability, a potential predictor of outcome. Participants received physiotherapy including exercise and manual therapy. Outcome was evaluated at nine weeks and one year using the Western Ontario and McMaster osteoarthritis index (WOMAC) and global rating of change (GRC) transition scale; with response determined using OMERACT-OARSI (Clinical Trials Response Criteria Initiative and Outcome Measures in Rheumatology/ Osteoarthritis Research Society International) responder criteria. Nine week response was investigated as an additional predictor of successful outcome.
Data Analysis: Pre-test probability of success was the number of participants with a successful response following physiotherapy treatment, divided by the total number of participants in the treatment cohort.
Baseline variables with a univariate association with one year outcome (p<0.2), were entered into multivariate logistic regression with backward elimination, to identify a set of variables (p < 0.1) that predicted successful outcome at one year. Accuracy statistics were used to calculate post-test probability of success for different numbers of retained predictors. A novel approach used post-estimation analysis to identify named combinations of variables, and associated probability of success.
Results: Pre-test probability of success was 35%. Six predictors of success were identified: posterior knee pain, disturbed sleep, absence of knee injury, instability, symptom duration (> 5 years), and female sex. The optimal model, with at least four out of six predictors, increased post-test probability of success to 66%. Presence of less than three out of six predictors decreased post-test probability of success to 11%. Using the post-estimation test, a five variable model (posterior knee pain, disturbed sleep, absence of knee injury, instability, and female sex), gave an 87% probability of success.
Data from the usual care control group did not fit the model, providing some evidence that variables were predictors of treatment response rather than natural progression of knee OA.
Nine week response gave an inferior prediction of probability of success at one year (52%), but combined with baseline predictors produced an optimal model of at least five out of seven predictors, with a post-test probability of 86%.
Conclusion: It is possible to use baseline variables, with or without nine week response, to predict physiotherapy treatment response at one year, for patients with knee OA. These findings represent the derivation stage of a clinical prediction rule which requires validation in future research.