Background
Osteoarthritis of the knee (OAK) is the most common type of osteoarthritis (OA)[
1], and its prevalence is rising in parallel with the increasing age of the population [
2]. The condition is associated with pain and inflammation of the joint capsule [
3], impaired muscular stabilisation [
4,
5], reduced range of motion [
6], and functional disability.
European League Against Rheumatism (EULAR) recommendations state that both pharmacological and non-pharmacologial interventions are needed for optimal treatment of OAK with at least 33 potentially effective interventions at the clinicians' disposal [
7]. Ten of these interventions are listed as non-pharmacological and 5 of these non-pharmacological interventions are physical agents: acupuncture; low level laser therapy (LLLT); pulsed electromagnetic fields (PEMF, including shortwave therapy SWT); transcutaneous electrical nerve stimulation (TENS), and ultrasound (US). While paracetamol, opioids and coxibs receive recommendations based on the second highest level of evidence (1B), no physical agents are recommended in spite of being listed as having the same evidence level (1B).
Inadequate dosageand inappropriate procedural technique can contaminate the findings of RCTs of physical agents but the EULAR analysis did not account for this. Recent findings suggest that most physical agents exhibit fairly distinct dose-response patterns, and failure to account for adequacy of TENS [
8] and LLLT [
9] interventions can markedly reduce ES estimates. Indeed, evidence-based guidelines for dosage and treatment procedures and the conduct of systematic reviews have been published for LLLT [
10], and for acupuncture [
11].
An appropriate approach would then be to investigate the short-term efficacy of physical agents for OAK, for all trials with each intervention and then to make sub-group analyses for trials according to their compliance with adequate dosageand procedural recommendations. Consistency in trial design and in the selection and timing of outcome measures must be assured to allow for comparison between interventions [
12]. The selected meta-analysis methodology was identical to that previously used by our group to assess common pharmacological interventions for OAK [
13].
Methods
Review protocol specification
A detailed review protocol was specified prior to analysis. This included a sequential three-step reviewing procedure of 1) harvesting randomised placebo-controlled trials where patients were treated with specified interventions for knee ostoarthritis, 2) evaluating their methodological quality according to predefined criteria, and 3) calculating their pooled effect as the weighted mean difference (WMD) in change between intervention and placebo in mm on a 100 mm visual analogue scale (VAS).
Literature search
A specified literature search was performed from 1966 through April 2006 on Medline, Embase, Cochrane Controlled Trials Register for RCTs, CINAHL, Database of Abstracts of Reviews of Effectiveness (DARE), International Network of Agencies for Health Technology Assessment (INAHTA) database, The Physiotherapy Evidence Database (PEDro), National Guideline Clearinghouse (NGC), PRODIGY Guidance, and NICE (National Institute for Clinical Excellence). In addition, hand searches were performed in the journal Laser Therapy from 1994, and in books of abstracts from congresses arranged after 1990 by the World Confederation of Physical Therapy and World Association for Laser Therapy.
The following search string was used: Osteoarthritis OR osteoarthrosis OR knee OR exercise OR electrotherapy OR laser therapy OR light therapy OR ultrasound OR electrostimulation OR transcutaneous electrical nerve stimulation OR electromagnetic AND randomized OR randomised.
In addition, handsearches of national Scandinavian physiotherapy journals, conference abstracts and reference lists of systematic reviews were performed, and experts in the field were consulted. No language restrictions were applied with papers in English, German and Scandinavian languages eligible for inclusion.
Inclusion criteria
The trials were subjected to 6 inclusion criteria:
1. Diagnosis
A statement in the report that knee osteoarthritis had been verified by clinical examination according to the American College of Rhematology criteria and/or by x-ray.
2. Symptom duration
More than 3 months.
3. Trial design
Randomised blinded placebo-controlled parallel and cross-over groups design.
4. Outcome measures
Primary outcome measure: Pain intensity within 4 weeks of treatment start scored on the Western Ontario and McMaster Universities osteoarthritis index (WOMAC) subscale of pain, or on a 100 mm VAS for global or walking pain.
Secondary outcome measure: Pain intensity, as measured for the primary outcome measure, at 5–12 weeks follow-up.
5. Threshold levels for clinical relevance
Mean threshold for OAK patients reporting "minimal clinical important improvement" has been determined to 19.9 mm on VAS [
14]. Likewise, the mean threshold for inducing a categorical change from "no change" to "slight improvement" has been determined to be 12.7 mm [
15], while the mean threshold for "minimal perceptible clinical improvement" is determined to be 9.7 mm [
16].
6. Intervention groups, including criteria for modality-specific optimal dosage
Acupuncture
Interventions which produced somatic stimulation of 'acupuncture points' were included; i.e. manual or electrical dry needling.
Criteria for optimal dose (i.e. compliance with adequate dosageand procedural recommendations) were: manual or electrical dry needling of acupuncture 3 or more acupuncture points as defined in Traditional Chinese Medicine and performed by an acupuncturist with at least 2 years clinical experience. As it is plausible that manual acupuncture and electro-acupuncture trigger different biological mechanisms, we decided to group and assess manual acupuncture and electro-acupuncture separately. Categorisation as electro-acupuncture demanded electrical current intensity to be at a strong, near noxious level, which has been shown to be more effective than a low intensity level [
17].
Low level laser therapy (LLLT)
Criteria for optimal dose: GaAs 904 nm infrared pulse lasers = intensities between 12–60 mW/cm2 and doses between 1 – 4 Joule per session; GaAlAs 780–860 nm infrared pulse lasers = intensities between 30–200 mW/cm2 and doses between 6 – 24 Joule per session.
These doses are based on optimal location-specific dose ranges for osteoarthritis when the joint capsule is exposed [
9] and dosage recommendations from World Association of Laser Therapy for pain relief[
18].
Pulsed electromagnetic fields (PEMF), including shortwave therapy (SWT)
SWT (27 MHz)
Criteria for optimal dose: intensity between 14.2–76.7 Watts, pulse frequency between 100–800 Hz, treatment time 20–30 minutes and doses between 17–138 kJoule per session. These doses are based on a review of clinical trial literature to determine optimal treatment procedures and dose ranges for shortwave (27 MHz)[
19].
PEMF other than SWT
Criteria for optimal dose: There is a lack of consensus of optimal doses for intensity, so PEMF (other than SWT) delivered at any intensity was included. Frequencies between 10 and 200 Hz in line with those used in most animal studies.
Electrical stimulation using surface electrodes (TENS)
Interventions which delivered electrical currents in the milliampere range across the intact surface of the skin to stimulate nerves innervating the knee joint (L4-5, S1; [
20]) were included providing a standard TENS device or an interferential current stimulator was used [
21]. Interventions using any other TENS-like device were excluded because of the absence of a plausible physiological rationale (e.g. microcurrent electrical stimulation, high voltage pulse (galvanic) currents, high voltage TENS pens, transcranial electrical stimulation, transcutaneous spinal electroanalgesia (TSE), H-wave therapy and action potential simulation [
21]). No restrictions were placed on the electrode types.
Criteria for optimal dose: a strong, near-noxious intensity, pulse frequencies between 1–150 Hz, treatment time at least 20 minutes per session in at least 5 sessions. These doses are based on a meta-analysis with sub group analysis for optimal dose for TENS [
8].
Ultrasound therapy
Interventions which delivered mechanical vibration using an ultrasound device at frequencies between 1.0–3.0 MHz
Criteria for optimal dose: intensity 0.1–3 W/cm2, continuous or pulsed output, treatment time between 3–20 minutes and doses between 18–540 Joules per session. These doses are based on those commonly reported in clinical literature as optimal dose range has yet to be established [
22,
23].
Static magnets
Criteria for optimal dose for this modality remain uncertain, as does their anatomic location for placement on the human body.
Placebo control groups
Reports that stated that they had included a placebo or sham control were included. For LLLT, PEMF, TENS and US reports were checked to ensure that the placebo/sham intervention was inert in the form of an identical device delivering no output (i.e. a dummy device). For acupuncture, sham interventions were considered inert if they used non-acupuncture points and superficial needling (≤ 2 mm) or a specifically designed placebo needle. For sham magnet therapy, identical-looking devices without any or insignificant magnetic fields were considered.
Assessment of methodological quality
A criteria-list of methodological criteria was used for assessment of trial quality [
24]. Assessments of trial methodology were made by two independent reviewers (JMB and RABL-M). No specific cut-off limit for method scores was pre-planned as criterion for exclusion.
The change in overall pain intensity between the active intervention group and placebo was used. If more than one attainable outcome measurement was obtained in the first 4 weeks after treatment started, the time point corresponding to the largest effect values was selected. If data on overall pain intensity were missing, data were obtained as a mean of the 5 items on the WOMAC pain subscale. If WOMAC data were registered on non-continuous (categorical, Likert) scales, they were converted to 100 mm VAS and checked against other subscales and overall WOMAC score, as this has been found to have good internal consistency [
25]. If overall pain or WOMAC pain subscale data were unavailable, pain on movement was used as registered on a 100 mm VAS.
Statistical analysis of pain-relieving effect
Mean differences of change for intervention groups and placebo groups and their respective standard deviations (SD) were included in a statistical pooling. If variance data were not reported as SDs, they were re-calculated algebraically from the trial data of sample size and other variance data such as p-values, t-values, standard error of mean, or 95% confidence intervals [CI]. As a control measure for the stability of the small (n < 40) trials results, we substituted reported SDs (or other variance data) with the arithmetic mean SD from the other trials with the same intervention if SD was lower than the the arithmetic mean [
26].
Results were presented as weighted mean difference (WMD) between intervention and placebo with 95% CI in mm on VAS, i.e., as a pooled estimate of the mean difference in change between the treatment and the placebo groups, weighted by the inverse of the variance for each study (Fleiss 1993). A fixed effects model was applied.
Subgroup analysis
In order to give as precise effect estimates as possible, care was taken to investigate discrepancies in trial samples and interventions. The validity of heterogeneity tests is equivocal, and their results were only used to support subgrouping in cases where clinical and methodological quality heterogeneity was evident. Heterogeneity was tested using Q-values, and statistical significance was defined at the 0.05 level for each intervention. To analyse heterogeneity and effect size for each intervention, trials were then subgrouped according to baseline pain, methodological quality, adequate dosageand procedural recommendations for each physical agent using the criteria listed previously (see criteria for optimal dose). Subgroup analyses were also performed for results during the 5–12 week follow-up period, and for funding sources.
Publication bias analysis
Effect size plots were used as a graphical test in order to detect possible publication bias [
27,
28].
Outcome measures
1) Best reduction in pain intensity during the first 4 weeks after initiation of treatment scored on the subscale of pain on the Western Ontario and McMaster Universities osteoarthritis index (WOMAC) [
29] or on a 100 mm visual analogue scale (VAS) for one, or the mean score of two or more pain dimensions. Variance was calculated from the trial data and given as 95% confidence intervals [95% CI] in mm on VAS. Effect size within 4 weeks was defined as a pooled estimate of the difference in change between the mean of the treatment and the placebo control groups, weighted by the inverse of the standard deviation for each study, i.e. weighted mean difference of change between groups.
2) Follow-up results at 1–12 weeks after end of treatment were used for pain intensity (as described under 1) or categorical data of global health status. Improved global health status was defined as any one of the following categories: "improved", "good", "better", "much improved", "pain-free","excellent". The numbers of "improved" patients were then pooled to calculate the relative risk for change in health status. A statistical software package (Revman 4.2) was used for calculations.
Discussion
It seems that all but two of the included physical agents (MA and ultrasound therapy), exhibit statistically significant effects over placebo within 1–4 weeks, regardless of what doses and treatment procedures were being used. However, effect sizes for PEMF and SM, failed to reach the mean threshold for "minimal perceptible clinical improvement" for OAK as defined by Ehrich et al. [
16]. It cannot be ruled completely out that more studies may contribute to optimise PEMF treatment procedure and dosage, but at present MA, PEMF, US and SM cannot be recommended for rapid pain relief in OAK management.
For TENS, the above findings are at odds with previous reviews of TENS in chronic pain [
64] and in chronic low back pain [
65], but not in knee osteoarthritis [
66].
The picture for acupuncture is mixed, and most studies with MA have been performed using fewer weekly treatment sessions than the other interventions. Consequently, the results at 4 weeks are similar to those of the placebo groups, while the effect at 8 weeks is statistically superior to placebo. In a systematic review of acupuncture reviews, it has been argued that the evidence in favour of acupuncture is weakened by lack of randomisation and lack of assessor or patient blinding [
67]. In this review, we have only included randomised and double-blinded (patient and assessor) trials, and the results are in line with a recent review of acupuncture OAK [
68]. However, the clinical relevance of the MA effect in OAK remains questionable, and the results infer that EA seems to be a better choice in OAK management.
For LLLT, a Cochrane review has found limited evidence in favour of LLLT in rheumatoid arthritis and inconclusive evidence in osteoarthritis [
69]. But we have previously pointed out that the findings in osteoarthritis could be caused by inherent methodological weaknesses [
70] such as lack of adequate dose-response analyses[
71]. In line with the dosage recommendations from World Association for Laser Therapy, the findings above suggest that 904 nm is only effective with doses of 2–12 Joules and 830 nm with doses of 20–48 Joules when applied to 2–8 points over the joint capsule.
The small sample size of some trials on EA, TENS and LLLT may undermine the validity of our conclusions. It has been argued that evidence for most interventions lack sufficient statistical power to make valid conclusions [
72]. The Oxford pain research group suggests that reasonably robust conclusions can be be made from systematic reviews including 200 patients and/or more than 4 trials [
67]. Cochrane reviews offer positive conclusions for pharmacological interventions for pain based on the inclusion of 40 patients for neck pain [
73] and 185 patients for OAK [
74]. The sample size for our total and subgroup analyses for optimal treatment for EA, TENS and LLLT met the criteria stated by the Oxford group (EA n = 242, LLLT n = 222, TENS n = 272). Nevertheless, we remain cautious in our conclusion until larger scale clinical trials are available to verify the results. Methodological trial quality also undermines review conclusions [
36,
63], although the majority of trials on which our conclusions rest, were of acceptable quality.
The biological rationales for the observed effects seem somewhat clearer for EA, TENS and LLLT than for the interventions that demonstrated lesser effects. EA and TENS has been shown to inhibit ongoing nocicpetive transmission at a segmental level and that this is dose-dependent [
75]. The EA-trials included in the review delivered electrical stimulation with needles placed in the painful area, similar to that used for TENS. This is consistent with established physiological principles whereby stimulation in dermatomes and myotomes related to the pain are likely to elicit segmental analgesia mechanisms. It has been shown in experimental studies that electrical stimulation by both needle and skin electrode can produce similar analgesic effects [
76]. The observed similarities between TENS and EA in effect size and time-effect profiles after cessation of treatment, may be indices that similar physiological mechanisms are being induced by these two interventions. Adding the data from trials using acupuncture to trials using electrical stimulation in the form of EA, did not increase effect size over TENS to any appreciable extent.
During the last three years, controlled LLLT-trials have found dose-dependent anti-inflammatory effects under
in vitro,
in vivo, and
in situ conditions [
77,
78]. Another possible explanation for the observed positive LLLT effects may arise from local dose-dependent biostimulatory effects on cell activity which have been observed in controlled
in vitro [
79]
and vivo [
80] trials with lower, but overlapping, dose intervals.
The value of standardising treatment procedures and dosage in the treatment with physical agents is highlighted by the finding that doses which work well in laboratory settings also can be extrapolated to induce better pain reduction in the clinical subgroups of EA, TENS and LLLT-trials with optimal treatment. But the heterogeneity of treatment procedures, application techniques and doses still call for careful interpretation of the results.
Until now, physical therapies have often been neglected in editorials and reviews of treatments for OAK [
81,
82] and this may have resulted in the under-utilization of physical agents in OAK management [
83]. The safety of the physical therapies seems good as no serious adverse events were reported in the 36 RCTs reviewed. The advantage of physical agents is that they can be used in combination with drug therapy, thus reducing drug dosage and adverse effects. There is also some evidence that effects from adequately administered TENS, EA and LLLT remain clinically relevant even 1–2 months after the end of treatment. It may be difficult to directly compare the results of trials of physical agents with those of pharmacological interventions because of differences in the nature of the placebo's used in the trials.
In the pharmacological literature publication bias in favour of small trials with positive results has previously been detected. There seems to be no support for this tendency from asymmetry in the graphical plot [
27]. On the contrary, a small asymmetry towards publication bias in favour of small trials with negative results seems to be present for these physical interventions.
Exercise therapy, education and weight loss still remain the cornerstones of long-term OAK management [
81], but our results suggest that EA, TENS and LLLT have potential to become useful adjuncts in OAK pain management.
Competing interests
The author(s) declare that they have no competing interests.
Authors' contributions
JMB and AEL conceived of the study, and participated in its design and coordination BB, and RLM performed the literature search in the databases, while MIJ and RC handsearched for additional studies. JMB, BB and AEL and RLM made the methodological assessment of studies, and JMB, RLM and RC performed the statistical analysis and the meta-analyses. JMB, MIJ and RC drafted the manuscript, BB made the figures and AEL revised the manus for typographical errors. All authors read and approved the final manuscript.