Main findings
This study investigated whether change in physical activity was associated with future clinical outcomes of pain and physical function in older adults with knee pain attributable to OA. This question is novel and important since change in physical activity may explain clinical improvements following exercise interventions and inform future interventions. The main finding from this RCT was that change in physical activity level was not associated with future pain, physical functioning or treatment response at either three or six month follow-up. Small β coefficients were expected given the difference in scale between the PASE (0–400+) and WOMAC pain and function scores (0–20 and 0–68 respectively) (since the PASE scale is larger by approximately a factor of 20 than the WOMAC pain scale). However, even taking this in to account and allowing for a 10 point change in PASE, the magnitude of associations were very small, non-statistically significant and do not appear to be of clinical importance (Tables
4,
5 and
6). For example, extrapolating from the β coefficients, changing physical activity by a full standard deviation (83 points on the PASE) would be associated with less than a 1 point change in WOMAC pain or function at either three or six months. Similarly, large changes in physical activity would only have a very small effect on the odds of being an OMERACT-OARSI responder.
The null association findings (Tables
4,
5 and
6) suggest that change in overall general physical activity level, as measured by the PASE, does not explain clinical outcome following exercise intervention within the BEEP trial and that other variables may be responsible for the observed improvements in pain and function (see Table
2). For example, lower limb strengthening [
32] or psychosocial factors (such as outcome expectations, attention and monitoring, the interest and empathy expressed by physiotherapists and the credibility of the intervention) may contribute to improvements in pain and function [
33,
34].
There is also a separate or further explanation for the null findings. Measuring change in physical activity using the self-report PASE involves a number of limitations that may increase the chance of a Type II error (i.e. rejecting an association between change in physical activity and clinical outcome if one exists). Although the PASE has been highlighted as a promising measure of physical activity in older adults with OA [
35], all self-report measures of physical activity can either over- or under- estimate actual physical activity level [
36] since they are at risk of recall bias (through errors in memory and activity recall), misclassification of physical activity intensity and duration [
15,
37], and social desirability bias by participants [
38]. Furthermore, modelling change in physical activity is methodologically challenging and using an absolute change score between two time-points may compound measurement errors and reduce regression coefficient precision, biasing our findings towards the null [
39].
The PASE minimal detectable change statistic (87), which measures the threshold for a “real” change that with 95% confidence is beyond measurement error [
40], is considerably larger than the mean change detected in the BEEP dataset (15.1). This suggests that the mean change in PASE scores detected during the exercise interventions was relatively small, potentially affected by measurement error and perhaps insufficient to influence clinical outcomes or alternatively that the PASE may have inadequate responsiveness in older adults with joint pain.
To the authors’ knowledge, this is the first study to explicitly investigate the association between change in physical activity level and clinical outcomes of pain and function in older adults with knee pain attributable to OA. However, similar studies have investigated the association between change in physical activity level and disability in those with low back pain [
41] and pain severity and physical function outcomes in those with fibromyalgia [
42,
43]. Similar to our findings, no association was found in the back pain study using either self-report or accelerometer measured physical activity [
41], however, associations were found between change in physical activity and future clinical outcomes in the two fibromyalgia studies [
42,
43]. Whilst the aetiological differences between knee pain attributed to OA and fibromyalgia are likely to be substantial and prevent direct comparison, these findings do suggest it is possible to detect associations if they exist between self-report change in physical activity and clinical outcomes despite the previously discussed challenges in measuring change in physical activity.
Strengths and limitations
The strengths of this study were the ability to investigate both univariable and adjusted associations between change in physical activity and future clinical outcomes at two separate time-points. The use of multiple imputation helped preserve sample size, reducing the risk of bias due to loss to follow-up [
29] whilst the sensitivity analysis added strength to the primary findings by exploring the dataset under different missing data assumptions.
The main study limitation, relating to the challenges in measuring change in physical activity, has been discussed above. With our available data we were unable to investigate levels of different types of physical activity, for example, time spent in strengthening or different cardiovascular intensities of exercise. Different types of physical activity may have had different effects on outcome. Another concern for our analysis is temporal bias. Temporal bias occurs when the inference about proper temporal sequence of cause and effect is erroneous [
44]. Attempts were made to measure the exposure of interest- change in physical activity level (baseline to three months) prior to the clinical outcomes (at three and six months). Nevertheless, change in pain or physical function may have occurred prior to any change in physical activity, meaning we cannot be sure about the direction of any potential cause and effect. In handling missing data using multiple imputation for our analysis, we made the assumption that our data was “missing at random” (MAR) [
29] since we deemed it likely that missing values could be estimated from observed values. If any missing data was “missing not at random” (MNAR) i.e. there are systematic differences between missing and observed values even after the observed data are taken into account then our multiple imputed analysis would be at risk of bias [
29].
Finally, before generalising our findings it is important to remember this sample population were older adults with knee pain attributed to clinically diagnosed OA from a RCT of exercise interventions. Other populations of older adults with knee pain, for example, those who did not consent to exercise interventions may have different clinical outcomes with changing their physical activity levels.
Implications
Although change in physical activity was not shown to be associated with clinical outcomes of pain and physical function, insufficiently active older adults with knee pain can still be advised to increase their physical activity levels as able, in order to achieve the associated general health benefits [
45‐
48] with the reassurance that modest increases in physical activity are not associated with increasing pain or deterioration in function at a group level.
In order to select the most appropriate physical activity measure for longitudinal studies measuring change in physical activity in older adults with knee pain, future research could investigate and compare the reliability and responsiveness of the PASE alongside other recommended measures of physical activity such as the International Physical Activity Questionnaire (IPAQ) [
37] and direct measures such as pedometry and accelerometry.
The inclusion criteria for this trial sample was based on older adults with knee pain regardless of baseline level of physical activity. Future studies could investigate specific subgroups of older adults with knee pain, such as those who are inactive, who may plausibly respond differently to increases in physical activity than all adults with knee pain or those currently meeting guideline recommended activity levels.