Reductions in co-contraction following neuromuscular re-education in people with knee osteoarthritis

A total of 22 participants with knee OA were recruited and offered AT lessons. As no previous data for the clinical effectiveness of AT lessons for knee OA were available, the sample size calculation was based on data from a study of a self-management and exercise rehabilitation programme [37]. This study reported a change in WOMAC pain scores of 1.5 SD immediately after the 20-sessson intervention. Based on these data, an a priori, within-sample calculation was performed using the g*power software. This showed that a target sample of n = 20 (assuming 10 % attrition), would be sufficient to detect a change of 0.75 SD in WOMAC pain score (half that reported in [37]) with a statistical power of 0.8 and an α = 0.05.

Participants with knee OA were recruited through four general practitioners in the Greater Manchester area. In each of the five practices, electronic patient records were reviewed to identify all participants who satisfied the following criteria:

A total of 150 letters of invitation were sent out. Those who replied (n = 38) underwent further telephone screening to ensure that they did not have any problems with balance, satisfied all the above criteria and regularly experienced knee pain during walking. A total of 22 individuals satisfied these criteria and all were invited to participate in the study. A healthy control group (n = 20) was also recruited in order to address the second research question, relating to biomechanical joint loading. This group was recruited via advert around the university and through local community groups, such as U3A and Rotary. Each healthy participant had to satisfy criteria 2–5 above and have no history of musculoskeletal disorders of the lower limb or spine. Healthy participants were selected so that the mean age and BMI of this group was matched to that of the knee OA group. Before testing all subjects provided written informed consent to participate in the study and ethical approval was obtained from the NRES Greater Manchester North ethics committee, reference: 11/NW/0057.

The Alexander Technique (AT) intervention was delivered by an experienced local practitioner who was a certified member of the Society of Teachers of the Alexander Technique. The AT is usually delivered on a one-to-one basis and, for this study, each participant was offered 20 one-to-one AT lessons each lasting 40 min. The lessons were delivered over a 12 week period, twice a week for the first 8 weeks and weekly for the final 4 weeks. This rate of delivery was chosen to ensure similarity with other clinical trials investigating the AT, as well as with routine practice [31, 32].

One of the primary objectives of AT instruction is to improve one’s overall pattern of postural muscle tension. This is achieved by guiding an individual to prioritise attention to maintaining a dynamic coordinated and lengthening central body axis and to improved awareness of appropriate and inappropriate postural and tensional patterns. Typically, these skills are taught using such common movements as sit-to-stand or walking, as well as with the person lying in a semi-supine position. The teacher uses gentle manual guidance and verbal instruction, together with constructive feedback, to enable the individual to lessen habitual interference with and maintain the lengthened axial tensional pattern, thereby reducing inappropriate tensional patterns in general, and typically leading to movement that perceivably requires less effort. In addition, they are taught to be aware of subtle increases in anticipatory muscle activity which are triggered during movement and also in anticipation of pain. Individuals are taught to interpose mental ‘directions’ between their intention to move and their execution of the movement so as to maintain postural support activity in the spine and torso, and to avoid overreacting in anticipation of the usual effort (or pain) involved. Following the guided sessions, participants were encouraged to continue to apply the skills learnt in the lessons as they went about their daily activities, maintaining an improved postural awareness and state and continuing to inhibit inappropriate tensional patterns.

Clinical effectiveness of the AT intervention was assessed using the WOMAC self-report questionnaire which captures information on pain, stiffness and function [38]. Given our hypothesis that AT instruction may lead to reduced co-contraction and subsequent pain, we defined the WOMAC pain score (5 items from the full WOMAC questionnaire) as the primary outcome measure. The full WOMAC score (20 items) was also analysed as a secondary outcome measure. Clinical outcomes were collected at baseline, immediately post intervention (within 1 week of the final AT lesson) and also at 15 months post baseline. In addition, a record of analgesia use during the week before the baseline assessment and the week before the post intervention assessment was taken as well as a record of any other therapy accessed during the intervention period.

Biomechanical data, used to characterise joint loading, and EEG data, used to characterise pain processing, were collected at baseline and immediately post intervention. Biomechanical data from a healthy cohort were also collected separately to characterise joint loading in this group. During each of the biomechanical testing sessions, participants walked barefoot at a speed of 1.25 ms⁻¹ along a walkway. Speed was measured with optical timing gates and only speeds within ±10 % considered acceptable. All subjects apart from one participant with knee OA were able to walk at this speed. In this one participant a walking speed of 1 ms⁻¹ was found to be appropriate and used for both the baseline and repeat testing. Although some previous studies of muscle activation, in people with knee OA, have instructed participants to walk at a self-selected speed [8], EMG amplitudes are known to vary with walking speed [39]. Therefore, in order to ensure appropriate comparison across testing sessions and between healthy and OA groups, we opted to control walking speed.

EMG data from the hamstrings and quadriceps was collected during walking to quantify co-contraction of the knee muscles. Data were collected using a Noraxon Telemyo system (3000Hz) with electrodes placed on vastus lateralis, vastus medialis, biceps femoris and semimembranosus according to SENIAM guidelines [40]. Following the walking trials, reference EMG data were collected during maximal voluntary isometric contractions (MVIC) for amplitude normalisation of the final EMG signals. To collect these data, the participant was seated in an isometric dynamometer with the knee flexed at 45° and, following a warm-up period, MVIC data collected first from the quadriceps and then the hamstrings. Three MVICs were performed for each muscle group with a 1 min rest between each contraction. During these contractions the net joint torque (measured with the dynamometer) was recorded in order to quantify knee extensor/flexor moments.

Following data collection, EMG data was exported to Matlab for processing. After applying a 20Hz high pass FFT filter to remove noise and movement artefact, the signal was rectified and then low pass filtered (6Hz Butterworth) to create a linear envelop [8]. The EMG signals were then time normalised to the stance period of the most affected limb using gait event data captured from the force platforms (see below). The MVIC data was filtered in the same way as the movement data and then a moving window algorithm used to determine the 0.1 s window in which the maximum EMG amplitude occurred [8]. This MVIC value was then used to normalise the EMG data from the walking trials. In order to characterise knee extensor/flexor strength, the peak torque was identified across the three maximal quadriceps/hamstring contractions.

We characterised muscular co-contraction during two specific phases of the gait cycle: an early stance phase and a pre-contact phase. Modelling studies have shown that, during early stance, there is a peak in the knee contact force which occurs between 15 and 25 % of stance [17]. It has also been shown that this peak increases with knee muscle co-contraction [17]. We therefore characterised muscular co-contraction during this period. We also characterised co-contraction over a pre-contact phase period (−5 % to 0 % of stance), just before initial contact. It has been shown that patients with knee OA increase background muscular co-contraction in anticipation of a destabilising perturbation [41] and it is possible that such anticipatory muscle activity may occur during walking, in preparation for contact with the ground. Such increased anticipatory contraction of the hamstring and quadriceps may lead to a subsequent increase in knee joint stiffness during the loading period. This may, in turn, affect the rate of increase of the knee contact force which could subsequently affect nociceptor input. As AT training aims to develop awareness and influence anticipatory muscle activity, it is possible that this intervention may influence co-contraction during the pre-contact phase.

A range of different algorithms have been proposed to calculate co-contraction of the knee muscles during walking. However, results from a recent modelling study suggested that simply summing the activity of the agonist and antagonist may give the best indication of articular loading [16]. Therefore, separate medial and lateral co-contraction EMG curves were obtained by summing the medial quadriceps and hamstrings and the lateral quadriceps and hamstrings respectively [10]. The final two co-contraction outcomes were then calculated as the respective means of the medial and lateral co-contraction curves over both the pre-contact phase (−5 % to 0 %) and the early stance phase (15 % to 25 %). Each time window was adjusted backwards to account for a 30 ms electromechanical delay.

Force and 3D motion plate data were also collected in order to quantify other aspects of joint loading. These data were collected using a 10-camera Qualisys Pro-reflex motion capture system (100 Hz) with two AMIT force plates (1500Hz) embedded in the walkway. Rigid clusters of 4 markers were used to track the motions of the thigh and shank, and a system of 4 markers, placed over anatomical landmarks, used to track motion of the foot [42]. Ankle and knee joint centres were calculated as midpoints between the malleoli and femoral epicondyles respectively and hip joint centres obtained using the regression model of Bell et al. [43]. Following data collection, the Visual 3D software (C-Motion, Rockville, Maryland) was used to derive the sagittal plane knee angle angles and also the sagittal and frontal plane knee moments from the kinematic data using a 6DOF model. These data were then time normalised to the stance phase. Peak sagittal angle, peak sagittal moment and peak frontal moment were then derived to characterise knee loading. Both peak moments were subsequently normalised to the participant’s body mass to define the final outcomes.

We used a 64-channel EEG system (BrainAmp MR, BrainVision UK) to measure brain activity (sampling rate of 500Hz, FCz reference), and a thulium laser stimulator to induce 30 very brief (<150 ms) heat sensations on the right forearm. Prior to EEG recording, a suitable stimulus intensity was determined for each participant that induced moderately painful sensations. This was tailored to each individual patient using a psychophysics procedure, such that they judged a moderately pain sensation (described to the participant as clearly painful but easily tolerable) as a rating of 7 on a numerical rating scale (NRS) from 0 to 10. The resulting laser energy output (mean (standard deviation) across patients was 1.2 (0.3) Joules. During EEG recording, participants provided ratings of pain intensity for each laser pulse on the same 0 – 10 NRS. At 3 s, 2 s and 1 s prior to each stimulus onset, they were cued with an auditory stimulus to anticipate the timing of the pain, allowing for the recording of anticipatory brain responses. After each laser pulse, laser-evoked potentials were also recorded, allowing for measurement of the brain’s reaction to pain sensations. These anticipatory and laser-evoked responses were derived from the average, across all trials, of EEG responses time-locked to the stimulus onset, after pre-processing of the data using standard analysis techniques. These included filtering of the data to retain 0 – 30Hz frequencies, cleaning of eye-movement and other artefacts using Independent Components Analysis, manual rejection of remaining artefactual epochs, baseline correction to -3500 ms to –3000 ms pre-stimulus and re-referencing to the common average of all scalp channels.

The analysis of the resulting event-related potentials (ERPs) focussed on two time windows. For anticipation, the mean amplitude of the ERP at electrode Cz and its eight surrounding electrodes during the late anticipatory period (−500 ms to 0 ms pre-stimulus – see Fig. 1) was analysed, as this previously showed enhanced amplitude in a study of OA patients relative to healthy controls [6]. To measure the neural response to pain, the relative difference between the N2 and P2 peaks of the laser-evoked potential was analysed, as these peaks have been previously shown to be modulated by cognitive factors such as attention and expectations of pain intensity [44, 45]. The N2 peak was specified as the largest negative deflection in the ERP in the time period between 200 ms and 350 ms after laser stimulation at electrode Cz, while P2 was the largest positive deflection between 350 ms and 500 ms at electrode Cz. For each peak, activity was averaged across electrode Cz and its eight surrounding electrodes over a 20 ms time window centred on the patients’ peak latencies.

We used a Wilcoxon signed rank test to investigate changes in WOMAC pain and full WOMAC score following the AT intervention. Paired sample t-tests were used to establish whether any of the biomechanical or pain processing outcomes changed significantly following AT instruction. Following this, independent t-tests were used to compare the biomechanical loading variables between the healthy and knee OA groups. Finally, to investigate the link between changes in mechanistic outcomes and changes in clinical outcomes, we used a Pearson’s correlation analysis. For all analyses, α < 0.05 was chosen as the significance level. Although, with the relatively large number of statistical tests, this increases the likelihood of a type 1 error, it was deemed appropriate given the exploratory nature of this study.

A total of 22 individuals with knee OA were recruited and began the AT intervention. One participant dropped out (after 10 AT lessons) and was removed from the study; however the remaining 21 participants completed all 20 AT lessons. Biomechanical and clinical data was collected from all 21 participants at baseline and immediately post-intervention, however, only 19 participants agreed to undergo the pain processing assessment. Six participants were lost to long-term follow up, which left 15 patients with OA at the 15 month follow up.

There were no significant differences in baseline demographics between the healthy and OA groups (Table 1). The main WOMAC pain score of the knee OA participants was relatively high, reflecting our inclusion criteria of knee pain on walking. Kellgren and Lawrence (KL) [46] grades of the OA participants ranged from 2 to 4 with 11 participants being classified as KL = 2, nine participants as KL = 3 and four participants with KL = 4. Only six participants experienced unilateral symptoms with the rest reporting bilateral knee OA pain. Of the 21 participants, 15 reported other musculoskeletal pain, including shoulder, hip, neck and low back pain and 15 reported taking analgesia for their knee OA pain at baseline.

At the end of the AT instruction period there was a reduction from baseline of 56 % (p < 0.01) in the WOMAC pain score and 54 % (p < 0.01) in the full WOMAC score (Table 2). These reductions were maintained at 15 months (p < 0.01, Table 2). Of the 15 participants who had taken analgesia for their knee OA pain, 10 participants had reduced or stopped taking the medication and 5 had maintained the same level. Interestingly of the 15 participants reporting other musculoskeletal pain at baseline, 11 reported improvements in this non-knee related musculoskeletal pain immediately after the intervention. None of the participants accessed additional therapy during the intervention period.

Following AT instruction, medial co-contraction was observed to decrease by 13 % (p < 0.05) during the pre-contact phase, however, there was no significant change in lateral co-contraction during this period (n = 19). Note that, due to problems with the instrumentation, it was only possible to obtain EMG for 19 of the 21 OA participants. Comparison with the healthy controls showed medial co-contraction to be 39 % higher (p < 0.02) in the knee OA group at baseline but only 22 % higher following the AT intervention, a difference which was no longer significant (p = 0.12). There was no significant difference between the healthy controls and the OA group in lateral co-contraction either before or after the AT intervention.

Similar changes in muscle activation were observed during the early-stance period following the AT intervention. Specifically, there was a reduction in medial hamstring (Fig. 2d) and medial quadriceps (Fig. 2b) activity which led to a 15 % reduction in medial co-contraction (p < 0.01, Fig. 2f). However the reduction in lateral co-contraction was not significant (p = 0.07, Fig. 2e). Comparison with the healthy controls showed medial co-contraction to be 57 % higher (p = 0.001) in the knee OA group at baseline. Although this reduced to 34 % following the intervention, the difference was still significant (p = 0.02). Analysis of the lateral co-contraction data revealed a similar pattern being 81 % higher (P < 0.001) in the OA group at baseline and 66 % (p < 0.001) higher following the AT intervention.

The strength data, measured using the isokinetic dynamometer, showed no differences in either peak knee extensor torque (p = 0.72) or peak flexor torque (p = 0.27) following AT instruction. However, although healthy participants were able to generate peak knee extensor torques which were 47 % greater (p < 0.002) than those produced by the patients with knee OA, there were no group differences in peak knee flexor torque. A similar pattern was observed during walking, with no statistical changes in peak knee joint angle or peak moments (Fig. 3) following AT instruction. However, there was a significantly higher peak sagittal moment in the healthy subjects compared to the baseline OA data (p = 0.03) and a significantly higher (p < 0.01) peak knee adduction moment in the OA participants when compared to the control subjects.

When the pain processing outcomes were analysed, no statistical changes were observed in either the late-anticipatory potential (p = 0.77) or the N2-P2 (Fig. 1) difference in the laser-evoked potential (p = 0.32).

Correlational analyses were carried out to test for a relationship between the change in WOMAC pain score and the change in medial/lateral co-contraction following AT instruction. These analyses were performed separately for the two different phases: the pre-contact phase and the early stance phase. For the pre-contact phase, a moderate correlation (r = 0.45, p < 0.05) was observed between the change in the WOMAC pain score and the change in medial co-contraction (Fig. 4a). This relationship showed a clear outlier (Fig. 4a), which when excluded, increased the correlation to r = 0.63, p < 0.01. There was some evidence of a correlation between the change in lateral co-contraction and change in the WOMAC pain during the pre-contact period, however this failed to reach significance (r = 0.37, p = 0.13, Fig. 4b). Furthermore, there was no evidence of a correlation between either of the co-contraction measures during the early stance period nor were any meaningful correlations observed between the changes in pain processing outcomes and the change in WOMAC pain.