Summary of Key Findings and Strength of Evidence

The key findings for each intervention appear in the Results section. Table 2 summarizes the findings, conclusions, and strength of evidence ratings that are reported in full in Appendix E.

Table 2

Summary strength of evidence.

In general, for the outcomes of interest, findings were insufficient to draw conclusions, or conclusions were supported by low levels of evidence. No conclusions were supported by a high level of evidence. This section highlights findings and conclusions for which we found moderate or low strength of evidence (SoE).

Summary of Findings in Relationship to What Is Already Known

Platelet-rich plasma. The current review identified beneficial short-term effects of PRP. Several 2015 SRs reviewed the effects of PRP, however all prior reviews included studies comparing PRP to hyaluronic acid or corticosteroid injections. We included only studies that compared PRP to saline injections to control for any placebo effect. Thus, we identified too few studies to pool.

Glucosamine with or without chondroitin. The 2007 SR found no significant benefit for glucosamine, glucosamine plus chondroitin, or chondroitin alone, compared with placebo, based on the large (n=1,583) GAIT trial.

New RCTs identified for this review provided conflicting evidence for effects of supplemental glucosamine, chondroitin, or the combination. A large noninferiority trial found comparable short-and medium-term effects for glucosamine plus chondroitin compared with NSAIDs, but no long-term effects of either. This trial did not include a placebo control. The 2008 post hoc analysis conducted by the authors of the GAIT trial found that when participants were stratified by baseline pain, those with moderate to severe pain demonstrated a trend toward improvement from glucosamine plus chondroitin (proportion experiencing 20 percent or greater improvement in pain).¹³⁴ The effect was moderated by the large placebo response. No new trials assessed short- or medium-term effects of glucosamine sulfate alone; three RCTs found no consistent long-term effects on outcomes of interest. Chondroitin showed evidence of medium-term effects but no long-term effects, in three new trials and a long-term followup of the GAIT trial. The analysis also found that the effect of chondroitin on swelling was seen predominantly in those with less-advanced disease.

Strength and resistance training. The 2012 SR found low-level evidence that “strengthening exercise” decreased pain and improved several other outcomes among individuals with OA of the knee, but no evidence for improvement in function was supported. That review did not describe their criteria for categorizing an intervention as a strengthening exercise intervention; therefore we have not attempted to pool studies identified for this report with theirs.

The current review strengthens the findings of the 2012 review on beneficial effects of strength and resistance training on pain. We identified evidence for a significant beneficial effect on total WOMAC scores and a nonstatistically significant beneficial effect on short-term pain and function based on pooled analysis of five RCTs. An ongoing RCT, the Strength Training for Arthritis Trial (START, ClinicalTrials.gov NCT01489462) is testing whether higher intensity, longer duration strength training can reduce long-term pain and OA progression by further increasing quadriceps strength and offloading stress on the knee joint.¹³⁵

Agility training. The current report identified low-strength evidence from six RCTs that strengthened the findings of the 2012 report on beneficial effects of agility training on long-term pain,¹⁹ as well as providing evidence on short-term benefits for pain (low strength of evidence).

Tai chi. The current report identified low-strength evidence supporting a beneficial effect of Tai chi on short-and medium-term pain and function, augmenting the findings of the 2012 report.

Yoga. The current report did not identify sufficient evidence to augment the findings of the 2012 report on aerobic exercise or yoga. A 2016 SR on the effects of yoga on OA of the knee found a significant short-term effect on pain; this review included six studies, some with very short follow-up times.¹³⁶

Manual therapy. For the current review, we found low-strength evidence for a lack of beneficial effect of manual therapy on short-term pain, based on three pooled RCTs, but no consistent effects on medium-term pain, function, or other outcomes, likely due to wide variation among the interventions, which included both physical therapist-applied manual therapy, therapeutic massage, and self-administered acupressure. The 2012 SR reported a low strength of evidence for an effect of massage on function based on two pooled studies (6–13 weeks) and reported improvements in disability and other outcomes based on three unpooled studies.

WBV. The current review identified a significant beneficial effect of WBV on medium-term function but not on medium-term pain, based on pooled analysis of three RCTs (low-strength evidence). Insufficient evidence was found for short- and long-term effects. The 2012 SR did not consider WBV as an intervention, and no other recent high-quality SRs assessed the effects of WBV on pain or function.

TENS and NMES. The current review found a beneficial short-term effect of TENS on pain, based on a MA of three RCTs (moderate-level evidence), but no consistent effects of TENS on function and no medium- or long-term effects. The review found insufficient evidence to draw conclusions regarding the effects of NMES on pain or function; further, strength, which is considered the primary outcome for NMES, was not included as an outcome of interest in the current study, and only two of the four studies that assessed NMES assessed strength.

The 2012 SR identified a beneficial effect of electrical stimulation, (including TENS and NMES) on short-term pain, based on meta-analysis of seven RCTs, but no other significant effects of electrical stimulation.¹⁹

Orthoses (knee braces, shoe inserts, custom shoes, and cane use. The 2012 SR identified low-level evidence for an effect of foot orthoses on function.¹⁹ That review did not identify studies on cane use, knee braces, or shoes.

The current review found no beneficial effects of shoe inserts on pain or function in pooled analyses. A 2015 Cochrane update review assessed the efficacy of orthoses (including one type of shoe, a custom variable-stiffness shoe) and knee braces.¹³⁷ That review included only one RCT that was published since the 2012 SR (included in the current review) and, in agreement with the current review, concluded that braces and orthoses had no consistent effects on pain or function.

Other physical interventions. The current report did not identify evidence of sufficient strength to augment or contradict the findings of the 2012 SR on therapeutic ultrasound, pulsed electromagnetic field therapy (PEMF), heat, aquatherapy, balneotherapy, or mud therapy.¹⁹

Weight loss. The 2012 SR did not consider the effects of weight loss, and no other systematic reviews were identified that assessed the effects of weight loss on the outcomes of interest for this review.

The current review identified moderate-level evidence from RCTs and single-arm trials supporting a beneficial of weight loss on medium-term pain and function and a low level of evidence supporting a beneficial effect of weight loss on long-term pain. Dose-response effects between weight loss and effect sizes were identified for medium-term pain but were inconsistent across most studies.

Home-based and self-management interventions. The 2012 SR included a number of studies that assessed the effects of home-based or self-management interventions but did not assess these interventions as a category. Two 2015 SRs reviewed the effects of home exercise programs¹³⁸ and self-management interventions¹³⁹ for the treatment of OA of the knee or knee conditions in general. These SRs reported positive effects of home exercise programs and self-management programs with exercise on pain and function but noted the heterogeneity of interventions and challenges in study design. Most RCTs of exercise interventions included in the current report expected participants to perform exercises at home, but the studies we analyzed in this category explicitly assessed home-based or self-management programs. These programs showed beneficial effects on short-term pain, and medium-term pain and function.

Adverse events. The 2007 SR reported that, in general, adverse events (AEs) for glucosamine with or without chondroitin did not differ between treatment and placebo groups, and no SAEs were reported. Likewise, the 2012 SR on physical interventions reported that AEs did not differ significantly between treatment and control groups and did not deter individuals from continued participation in trials. Approximately half of the studies included in the current review reported having assessed AEs. However, this number includes studies that simply reported that no AEs were found. Of the 13 RCTS that mentioned SAEs, most reported no SAEs or SAEs that could not be attributed to the intervention. Of note, AEs associated with glucosamine and chondroitin did not differ between groups in the placebo-controlled or noninferiority trials. WBV, which was not assessed in the 2012 SR, was not associated with any AEs. PRP was associated with pain and stiffness that increased with the number of injections.

Applicability

The applicability of the results of the trials included in the current review may be somewhat limited for several reasons.

First, the studies of glucosamine and chondroitin used forms and preparations of the dietary supplements that are not available commercially. In addition, the composition and purity of these supplements could be rigorously tested and ensured, unlike most commercial-grade supplements.

Likewise, the studies of PRP each prepared their material using proprietary processes, although at least one publication described the process.

As we discuss further below in the section on limitations of the literature, the results of studies of physical interventions may be influenced heavily by the ability of academic research centers to recruit highly motivated study participants. Even so, much of the success of such interventions in the community is likely to depend on the compliance of patients not only with respect to attending clinic appointments but also with their engaging in regular workouts away from the clinic. Only a small proportion of the studies were considered community based, and even those tend to attract the most motivated participants.

Implications for Clinical and Policy Decisionmaking

OA of the knee is an increasingly prevalent, progressively debilitating condition. Decisions regarding therapies for OA of the knee depend on a number of factors. Patient preferences have the strongest influence and are based on a combination of pain and perceived functional limitations and their influence on quality of life. Treatments for the condition range from the most minimal and least invasive (dietary supplements and over-the-counter analgesics) to total knee replacement. The current report considered only a subset of available interventions, and all fell along the less invasive end of the continuum.

A number of the interventions assessed in the report showed short- or medium-term benefit but either were not assessed sufficiently over the long term (meaning after a long intervention or after a shorter intervention with a long follow-up time, e.g., Tai chi, TENS, or PRP) or showed minimal benefits in the long term (e.g., glucosamine chondroitin). Several interventions showed beneficial long-term effects, including weight loss and several forms of physical activity (e.g., general exercise programs of the type taught by physical therapists). Because of study design and the numbers and duration of studies, it is not clear which physical activities are most effective, whether they are most effective in combination, or if benefit depends entirely on the individual patient. Adherence, which is obviously an important factor, was seldom assessed in the studies that met inclusion criteria.

One intervention that showed some medium-term benefit, intraarticular injection of PRP, has undergone limited testing for OA of the knee, especially regarding the effects of repeated injections. In addition, this intervention may not currently be covered by most insurers and its use as an intraarticular injection is considered off label by the FDA.

Pending longer RCTs of therapies that show promise for benefits in the short term, the implementation of progressive treatment plans, guided entirely by patient preference is supported by the findings of this review.

Limitations of the Evidence Base

Limitations due to study quality. The results of the RoB assessments for each study appear in Table F1 in Appendix F of the report. In the Results section of the full report, we have provided summary RoB scores for each study. The most prevalent limit to study quality was participant blinding: Only 33 of 85 RCTs reported an attempt to blind participants appropriately, using sham injections, placebo pills, sham applications of a treatment such as TENS, or in the case of exercise interventions, a control condition that could be considered an intervention itself. Many RCTs of physical interventions reported that participants were not or could not be blinded. Although outcome assessors were often reported to have been blinded in these studies, many of the outcomes of interest to this report were self-assessed (such as pain and WOMAC function). This lack of blinding significantly limits conclusions we can draw from the literature and is further discussed below in regard to comparators.

Another quality issue is the large number of RCTs for which adequate concealment of allocation could not be ascertained: 46 of 85. The inability to ascertain allocation concealment might sometimes be attributed to word limitations in publications, but is still a concern.

A third quality concern is the finding that 41 studies did not indicate use of intent-to treat analysis; since participants who are not experiencing benefit from treatment are more likely to drop out before study completion, per protocol analysis could artificially inflate apparent effects.

Fourth, 31 RCTs indicated evidence of incomplete adherence. This figure is actually deceptively low, as most interventions involving exercise require that participants work out on their own on days when they are not being supervised. Most studies did not attempt to monitor offsite compliance, and no studies assessed the effect of such compliance or adherence on outcomes.

Finally, although most studies demonstrated that participants were similar at baseline, some similarities were not routinely considered, such as weight status, or disease stage or severity, and almost no studies stratified outcomes by any baseline characteristics.

Additional limitations. The applicability of the findings of many of the studies to community settings may be limited by their having been conducted in an academic setting and enrolling highly motivated participants. For this reason, we attempted to assess the effects of home-based interventions; however, these interventions are limited in number, and also tend to be highly supervised. Related to this concern, compliance or adherence was almost never reported.

The applicability of studies of the dietary supplements, glucosamine and chondroitin, may be limited as they either did not report sources, did not ensure purity and concentration of active ingredients, or used forms and preparations that are not available commercially. Likewise, the studies of PRP each prepared their material using proprietary processes, although at least one publication described the process.

Another intervention-related limitation concerns the fact that many studies employed (or failed to prevent) multicomponent interventions. We purposely excluded studies whose multicomponent intervention design precluded assessment of the effect of a single component of interest. However, studies of physical modality interventions often implemented or focused on one type of activity added to a regimen of other activities (with the control group receiving the “other activities” only). In addition, many of the studies permitted continued use of analgesics or other treatments, preventing attribution of improvement to a specific intervention (or blunting the potential effects of an intervention). This problem is discussed further below.

Duration of interventions and followup was a concern. We limited inclusion to studies with a minimum followup of four weeks, because OA of the knee is a chronic, progressive condition. This decision had several implications; for example, no studies of taping met inclusion criteria, as the follow-up time was usually brief. Also, we did not consider the duration of an intervention as an inclusion criterion (as interventions such as PRP injection have no duration). Thus, the interval between the end of an intervention and outcome assessment, especially medium- or long-term followup, differed across studies. In categorizing studies by the length of followup times for potential pooling, we did not always consider the duration of the intervention, itself. This limitation could explain a lack of significant medium- and long-term effects, as few, if any, of the interventions included in this report are thought to have disease-modifying effects that last beyond the intervention.

A related limitation concerns the lack of sufficient numbers of studies with similar interventions to enable assessment of the effects of dose (or intensity, frequency, and duration of physical activity sessions). A 2015 Cochrane review found no evidence for significant differences in the effects of low vs. high intensity interventions on knee or hip OA patients but regarded that the evidence was insufficient to draw firm conclusions.¹⁴⁰

Another major challenge concerns the choice of study comparators. Contributing to this challenge is the self-reported, subjective nature of pain as an outcome. The placebo effect observed in large placebo-controlled RCTs of glucosamine with or without chondroitin diminished the effect of the active intervention. At the same time, a recent trial comparing glucosamine plus chondroitin to an NSAID found comparable beneficial effects of both. For the current report, we excluded studies that used only comparators of unclear efficacy (e.g., HA as a comparator for PRP) to make it possible to discern the magnitude of the placebo effect. We also excluded studies that used a participant’s less-painful knee as the comparator. However, the selection of appropriate comparators is a concern, particularly for studies of physical interventions such as strength training. Many of the studies we included employed usual care as a control; however, as described above, usual care often included a physical therapy program (usually some combination of strength and agility exercises and manipulation). Therefore, the failure to see a difference in outcomes between an intervention and a usual care control group might be attributable to there simply being a limit to the improvement that might be possible over that from standard physical therapy (especially over the often short duration of a study, and without major effort being expended by participants to work out on their own on days they do not attend the study classes). This conclusion is particularly likely, given that most studies that reported no differences in outcomes between interventions and active controls did report significant improvements from baseline. It is unclear what the most appropriate control is for studies of physical interventions or even studies of weight loss that include exercise: the findings of studies that compared diet alone to exercise and to diet plus exercise were difficult to interpret because exercise might have the same beneficial effects as weight loss, and whether they are synergistic or one actually masks the other could not be determined. That some studies used only active comparators while others used only inactive comparators also limited the numbers of studies that could be pooled or even compared.

A number of outcomes of interest were not reported in the included studies or were reported only sporadically. Risk for undergoing TKR was a prespecified outcome of interest in only one RCT. Many factors that cannot be accounted for influence the decision to undergo TKR. Thus, TKR has not proven to be a useful outcome for assessing the effectiveness of interventions.

We ideally hoped to assess the clinical as well as the statistical significance of any beneficial findings. To do so, we would assess whether statistically significant outcomes met a prespecified minimum clinically important difference (MCID). However, we encountered several major challenges in trying to do so. First, some publications failed to include the numerical scales used with their assessment tools. As a result, it was impossible to assess the potential clinical significance of their findings. Second, published MCIDs depend on the disease severity of the participants; the included studies varied widely in the disease severity of included participants, and some did not report it. Nevertheless, a wide variety of MCIDs have been derived and applied in reviews of similar patient populations (see Appendix I for a summary of published values). We selected and applied one set of values that has been applied in a number of similar reviews¹⁴¹ to the small number of statistically significant outcomes for which we had pooled standardized mean differences or for which we were able to identify the numerical measurement scales. But, thirdly, it is important to note that MCIDs are derived by translating patients’ responses on a scale of multiple items (e.g., the full WOMAC scale contains 24 items), each item graded using numerical rating scales of 4–100 points, to their response to a small number of subjective anchoring question; thus, their validity continues to be debated. Further, in studies with continuous outcomes, even if the mean difference is less than the MCID, a proportion of participants experience outcomes that exceed the MCID. Thus rigorously applying the MCID could prevent patients from obtaining potentially effective treatments.

Small sample size was an additional limitation of concern for many of the studies we identified. For such studies, the importance of significant findings cannot be assured.

Finally, because of the heterogeneity among studies with regard to interventions, comparators, outcome measures, durations of treatment and followup, and even reporting of the scales used for some outcome measures, few studies could be pooled. Although, we describe each study narratively in the report, the inability to pool results limits our confidence in the strength of evidence.

Future Research Recommendations

In general, future studies need to enroll sufficient numbers of participants to enable prespecified subgroup analysis according to important participant characteristics and to enable assessment of both statistical and clinical improvement. Studies also need to employ designs that permit assessing the effects of specific interventions and to consider including both active (sham) and passive comparison groups to enable participant blinding. Isolation of the interventions being assessed needs to be accomplished both by careful design of the interventions themselves and by prohibiting participants from using alternative modes of therapy. In addition, many interventions need to be conducted for longer durations and mechanisms need to be developed to better measure compliance. Reported outcomes need to include the percent of participants who experience improvement as well as an estimate of whether the effect size achieves a MCID. In addition, the use of imaging and other nonclinical measures will help clarify structure-function relationships and outcomes of interventions.

Recent OARSI guidelines on design of clinical trials for knee OA therapies include 25 recommendations. Among them are clear definition (of and rational for) inclusion/exclusion criteria; assessment and reporting of disease severity; ensuring randomization, blinding (to the extent possible), and similarity of important characteristics at baseline; use of validated outcome measures and steps to minimize bias in patient-reported outcomes.¹⁴² Recommendations specific to particular interventions are described below.

Cell-based therapies. Based on our finding of a significant effect of PRP in a small number of small, high RoB studies, and the number of studies that did not meet inclusion criteria because they compared PRP only to HA, we believe a large, saline-controlled trial is needed. Although corticosteroids could provide an additional comparator for noninferiority, the immediate adverse effects of intraarticular injection of corticosteroids would be impossible to mask. Residual benefits that remain after the intervention is discontinued (and the effect of follow up treatment) also need to be assessed.

In addition, no studies of stem-cell therapy or other cell-based therapies met inclusion criteria. A large multisite commercial clinic that was contacted for trial results did not respond to the request. Clinicaltrials.gov lists several registered trials of stem-cell treatments for OA of the knee, which should be monitored for published findings. We also identified four published studies of gene therapies (using autologous chondrocytes genetically modified to deliver a growth factor and designed to be injected intraarticularly), which to date, have been tested only in Phase II trials.^143–146

Glucosamine with or without chondroitin. The 2016 MOVES Trial found significant beneficial medium-term effects on pain, function, stiffness, and quality of life for a prescription form of glucosamine hydrochloride plus chondroitin that were comparable with those of a Cox-2 inhibitor in a large patient population with severe pain. The rate of AEs was relatively small and similar across groups (individuals with cardiovascular conditions were excluded). Thus far, longer-term outcomes have not been reported but would need to be considered in formulating guidelines regarding the use of a prescription grade form of the supplement, especially in light of the findings of the LEGS Trial that glucosamine, chondroitin, and the combination had no beneficial effects at 1 and 2 years compared with placebo. In addition, a head-to-head trial similar to MOVES should be conducted using a combination of glucosamine sulfate and chondroitin, as some evidence has suggested glucosamine sulfate is more effective than glucosamine hydrochloride.

Physical interventions. The studies on strength, agility, and aerobic training that met inclusion criteria usually combined the training modality that was being tested with additional exercises, for example, a strength training intervention would include aerobic exercise as a warm-up and would sometimes include a brief session of exercises aimed at improving agility or gait as well. This design matches the physical therapy regimens in current use and probably makes sense as a therapeutic regimen, but it requires that studies that aim to test a specific modality are carefully designed to ensure that the results can be attributed to the intervention being tested. Other SRs have also noted the difficulty in drawing conclusions regarding the clinical utility of various physical interventions.

Studies are needed to assess the effects of varying the “dose” of physical interventions, by comparing different numbers, durations, and/or intensities of treatments.

The efficacy of individually tailored multicomponent interventions also needs to be assessed but traditional clinical trial methods may not be well-suited to assess such interventions, because testing custom interventions essentially requires that patients serve as their own controls. A number of the trials included in our review modified interventions based on an assessment of individual participant deficits but only one assessed the effects of doing so and found no differences from participants who received a nontailored therapy.

Only one study of aquatherapy, and few studies of yoga or tai chi, met inclusion criteria. Larger trials of these interventions alone compared with both active comparators (to mask the intervention of interest) and waiting list (or other passive) comparators are needed, as they can easily be undertaken by sedentary individuals with no prior training.

OARSI recently published guidelines for the design and conduct of clinical trials of rehabilitation interventions, which include the physical interventions.^{142, 147} Recommendations are similar to those of the OARSI guidelines for assessing interventions for OA of the knee.¹⁴² Emphasis is on participant blinding when possible; assessor blinding; use of both sham (active) and passive comparators; description of baseline severity (with clinical measures, if desired); prespecification of adverse events for assessment; use of valid outcome measures with a benchmark, if possible; and assessment of the percent of participants who achieve improvement. Comparative effectiveness trials are advocated for testing novel treatments against those with established effectiveness or when blinding is not otherwise possible. Caution is suggested in applying published MCIDs, as they have been shown to differ by population and other factors.¹⁴⁸

Weight loss. This review showed beneficial effects of weight loss interventions on pain and function. Future studies need to clarify the roles of exercise and self-efficacy education in the observed effect to assess whether exercise and/or self-efficacy have their own effects, independent of caloric restriction and weight loss or if these co-interventions assist with weight loss and weight maintenance.

The OARSI recently released guidelines on design and conduct of diet and exercise interventions for OA.¹⁴⁷ Most of the recommendations were similar to those provided for rehabilitation and for OA of the knee interventions in general, in copublications. However, they also provided several additional noteworthy recommendations. These include the need to determine in Phase 1 trials whether high-intensity strength training, aimed at increasing quadriceps muscle strength, is safe in older adults with knee OA. Also recommended is allowing monitored use of rescue medication (analgesics), as weight loss trials tend to be longer in duration than other studies.

Home-based therapies. Our results, based on only a small number of studies, suggest home-based therapies with periodic supervision show beneficial effects on pain and function. This model has the advantage of requiring few clinic visits but the disadvantages of lack of monitoring of compliance and correct form when performing activities. The 2016 SR of home-based therapies by Anwer and colleagues also cites the issue of difficulty assessing compliance with home-based interventions.¹³⁸ Future research studies of home-based exercise could easily employ any one of a number of fitness monitoring devises to assess adherence and could use applications like Skype to periodically monitor performance.

Adverse effects. Future studies need to prespecify AEs of concern. Researchers need to actively and systematically collect information on adverse effects of interventions at defined intervals, particularly for cell-based therapies and intensive exercise programs.

Finally, because of the heterogeneity among studies with regard to interventions, comparators, outcome measures, and even reporting of the scales used for some outcome measures, few studies could be pooled. Therefore, no attempt could be made to assess whether any pooled effect sizes met or exceeded established MCIDs or MCIIs. For that reason, when individual studies reported their findings in those terms, we attempted to capture those data.

Conclusions

Among the interventions assessed in this report, many had insufficient evidence to determine their benefit for managing OA of the knee. Interventions that show beneficial effects on short-term outcomes of interest include TENS (moderate strength of evidence [SoE]), agility training, home-based programs, and PEMF on pain (low SoE); Tai chi on pain and function; and strength and resistance training on WOMAC total scores (low SoE).

Interventions that show beneficial effects on medium-term outcomes include weight loss for pain (moderate SoE) and function, intraarticular platelet-rich plasma on pain and quality of life, glucosamine plus chondroitin on pain and function, chondroitin sulfate alone on pain, general exercise programs on pain and function, Tai chi on pain and function, whole-body vibration on function, and home-based programs on pain and function (low SoE).

Interventions that show beneficial long-term effects include agility training and general exercise programs for pain and function, and manual therapy and weight loss for pain (low SoE). A moderate SoE supports a lack of long-term benefit of glucosamine-chondroitin on pain or function, and glucosamine or chondroitin sulfate alone on pain.

Insufficient evidence was found for long-term effects of most interventions, and for additional outcomes, such as stiffness, swelling, quality of life, and avoidance of knee replacement, for most interventions.

Larger randomized controlled trials are needed, with more attention to appropriate comparison groups and longer duration, to assess newer therapies and to determine which types of interventions are most effective for which patients.