INTRODUCTION

Osteoarthritis is a major contributor to chronic pain and physical disabilities globally.1, 2 Current pharmaceutical treatment for OA is largely restricted to analgesics including non-steroidal anti-inflammatory drugs (NSAIDs), which are both palliative in nature and accompanied by adverse effects.3 Fortunately, an array of etiologically targeted agents, known as disease-modifying OA drugs, have been developed in the hope of slowing its progression and provide symptomatic benefits4 (Appendix 1.1). These drugs act via a variety of mechanisms and targets including subchondral bone, cartilage, and synovium. While well-known potential DMOADs, such as glucosamine and chondroitin, are popular and used worldwide, they remain controversial in official recommendations and meta-analytic conclusions.5–8 Although there have been increasingly emerging potential disease-modifying drugs for OA, many unanswered questions persist regarding their treatment efficacy as well as underlying physiological mechanism and mode of action.

The aim of our study was to compare the long-term efficacy and safety of potential DMOADs on hip and knee OA.9

METHODS

Search Strategy and Selection Criteria

We included randomized placebo-controlled and comparative efficacy trials that evaluated orally administered DMOADs for knee or hip OA. Additional inclusion criteria included the duration of treatment was at least 12 months (6 months for MRI10) and included at least one structural outcome. We excluded hand OA patients because they are non-weight-bearing joints. There is evidence that OA at different joint types bears distinguishable pathophysiological mechanisms due to different anatomy and physiology and may respond differently to pharmacologic interventions.11–13 We included the following candidate drugs14, 15: chondroitin, glucosamine, diacerein, matrix metalloproteinase (MMP) inhibitors, collagen hydrolysate, vitamin E, vitamin D, inducible nitric oxide synthase (iNOS) inhibitors, doxycycline, avocado-soybean unsaponifiables (ASUs), hyaluronic acid, bisphosphonates, strontium ranelate, calcitonin, and licofelone.

The Cochrane Central Register of Controlled Trials, MEDLINE, EMBASE, and Web of Knowledge were searched without language, publication, or date restrictions on our own.16 A representative search strategy (Appendix 1.2) was further modified for grey literature. Reference lists from selected publications were screened by three reviewers for additional studies. The same three reviewers independently examined the article abstracts, and upon selection, full-text reports were evaluated for final inclusion.

Data Analysis

Data were extracted independently by two reviewers using a predesigned standardized form. Attempts were made to contact the authors to supplement incomplete reports; for those failing to respond, means and variance were estimated from available text and figures.14–19 Changes from baseline values were used to assess continuous outcomes,16 but if unreported, these values were derived from the baseline and final measurements14, 20–27 (Appendix 1.5, with correlation coefficients for standard deviation (SD) set as 0.528). Wherever possible, the results from the intention to treat (ITT) analysis were used.

The quality of evidence was assessed following the GRADE system.29 Study limitations were described in the risk of bias tables, plots, and contribution matrices.16 Common heterogeneity (Tau2) was assumed across all treatment comparisons in the NMA,30 while 95% confidence and prediction intervals of the NMA estimate were applied to assess the magnitude of heterogeneity.31 Consistency was tested both globally (design by treatment inconsistency model) and locally (side-splitting and loop-specific approach), with significance set at p <0.1.30 NMA transitivity was evaluated to assess the indirectness of studies. Finally, sensitivity analyses were undertaken for subgroups (including joint types, doses, dosing schedules, and product subtypes) and publication bias (using funnel plots, and analyses after excluding outliers).

Joint space narrowing on radiograph remains the only validated primary endpoint to assess the structural progression of OA and is recommended by both professional societies and regulatory agencies in the USA and Europe.17–19 The predefined primary outcome was therefore the structural efficacy on the minimum joint space width (weight-bearing X-ray). The relatively new cartilage volume measurement by MRI was also accepted in the analysis.20–22 Radiographic progression and total joint replacement were also included as adjunct structural outcomes. Changes in pain and function were analyzed as symptom outcomes, and if a study utilized more than one pain or function scale, then the result for the highest scale on the hierarchy list was extracted.23, 24 To assess safety profiles, drug-specific adverse events and study withdrawal rates were analyzed.

Wherever possible, data from primary studies were summarised in the meta-analysis, and if not, they were discussed in a narrative manner. Continuous variables were described with standardized mean differences (SMDs) using 95% confidence intervals (CI), while relative risk (RR) with a 95% CI was used to report dichotomous data. SMDs greater than 0.8 and RR greater than 2 or less than 0.5 were defined as a large effect, SMDs between 0.5 and 0.8 and RR between 1.25 and 2 or 0.5 and 0.8 were defined as a moderate effect, and SMDs of less than 0.5 and RR between 1 and 1.25 or 0.8 and 1 were defined as a small effect. Clinical effectiveness was defined as at least with a moderate or large SMD.25, 26, 58 The combination at different time points from a single study was performed following formulations in Appendix 1.5 (correlation coefficients for SD set as 1.0).16, 27 For those trials reporting multiple intervention groups (e.g., different dosages) with a common control group, groups were combined based on the formulations in Appendix 1.5.16, 27

In the pairwise meta-analysis, random-effects modeling was employed. The number needed to harm (NNH) was calculated for adverse events. Forest plots were employed to compare the treatment effects of each candidate. A NMA under a frequentist framework was introduced to optimize direct and indirect evidence and to rank all competing DMOAD candidates. In the NMA, random-effects modeling was adopted. The geometry and patterns of the network were prepared for each outcome, and the efficacy/safety hierarchy for a specific outcome was obtained via the surface under the cumulative ranking curve (SUCRA).28 Clustered ranking plots were based on the cluster analysis of SUCRA values for two outcomes and displayed using the two dimensions of the x- and y-axes.28

Analyses were performed using RevMan (version 5.3, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2015), Confidence in Network Meta-Analysis (University of Bern 2017, available from cinema.ispm.ch), and STATA (version 14, STATA cooperation, USA, 2015). P values were two-sided with α = 0.05. Patients or the public were not involved in the design, or conduct, or reporting, or dissemination of our research.

RESULTS

After application of inclusion and exclusion criteria (Fig. 1), 30 records (28 studies) were included.21, 22, 29–56 Overall, 7356 patients were randomized to treatment with a DMOAD, and 4534 were randomized to comparator groups, with well-balanced baseline characteristics (Appendix 1.3). Although a small-scale trial of collagen hydrolysate was identified,25 it was excluded due to the cartilage volume measurement approach. Hyaluronic acid was excluded based on its administration route. Details of the included trials are listed in Appendix 1.3. The network geometry (Fig. 2 and Appendices 2.2.4, 2.3.4, 3.1.4.3.2.4) revealed a star-shaped network with multi-arm trials (Appendices 2.1.3, 2.2.5, 2.3.5, 3.1.5, 3.2.5, 4.3.1). For our primary structural outcomes of minimum joint space or cartilage volume, the pairwise meta-analysis (Fig. 3) revealed significant effects at combined time points for chondroitin, doxycycline, and strontium ranelate. The network ranking (Appendix 2.1.4, 2.1.5 and Fig. 4) verified statistical significance but demonstrated low clinical meaningfulness of chondroitin (network SMD 0.21; CI [0.10, 0.32]; SUCRA 78.1; rank 1), strontium (0.20; [0.02, 0.38]; 74.4; 2), and glucosamine (0.16; [0.04, 0.28]; 65.2; 6). There was a trend towards greater improvement from 1 to 3 years for most medications (Appendix 2.1.1 and 2.1.2). In terms of radiographic progression, a pairwise meta-analysis identified strontium and diacerein as statistically effective agents (Appendix 2.2.1, 2.2.2, and 2.2.3), although the NMA indicated that chondroitin (network RR 0.59; CI [0.45, 0.77]; SUCRA 93.2; rank 1) and glucosamine (0.62; [0.45, 0.86]; 88.0; 2) were better (Appendix 2.2.6, 2.2.7, and 2.2.8). Seven studies31, 36, 37, 42, 51, 53, 56 reported the outcome of total joint replacement and the pooled estimate effects were insignificant in either the pairwise (Appendix 2.3.1, 2.3.2, and 2.3.3) or the NMA for any of the medications (Appendix 2.3.6, 2.3.7, and 2.3.8).

Figure 1
figure 1

Flow diagram of the study selection process.

Full size image
Figure 2
figure 2

Network maps of DMOADs regarding minimum JSW or cartilage volume (a) and withdrawals due to adverse events (b). The width of lines is proportional to the number of studies compared in every pair of treatments, and the size of nodes is proportional to the total sample size of each treatment. * licofelone study adopted naproxen instead of a placebo as a comparator. ASU, avocado soy unsaponifiables; DMOADs, disease-modifying osteoarthritis drugs; iNOS, inducible nitric oxide synthase; JSW, joint space width; MMP, matrix metalloproteinase.

Full size image
Figure 3
figure 3

Pairwise forest plots of DMOADs regarding minimum JSW or cartilage volume (a) and withdrawals due to adverse events (b). Random effects model with combined time points was used. ASU, avocado soy unsaponifiables; CI, confidence interval; DMOADs, disease-modifying osteoarthritis drugs; iNOS, inducible nitric oxide synthase; JSW, joint space width; MMP, matrix metalloproteinase; RR, relative risk; SMD, standardised mean of difference.

Full size image
Figure 4
figure 4

Head-to-head comparisons for the DMOADs on minimum JSW or cartilage volume and withdrawals due to adverse events. Consistent network meta-analysis model with combined time points was used. Data are SMDs/RRs (95% CI) of the column treatment relative to the row treatment. For minimum JSW or cartilage volume, SMDs higher than 0 favour the column treatment. For withdrawals due to adverse events, RRs lower than 1 favour the column treatment. Significant results are in bold and underscored. ASU, avocado soy unsaponifiables; CI, confidence interval; DMOADs, disease-modifying osteoarthritis drugs; iNOS, inducible nitric oxide synthase; JSW, joint space width; MMP, matrix metalloproteinase; RR, relative risk.

Full size image

There were significant but clinically low pain-relieving benefits from doxycycline, glucosamine, and vitamin D in the pairwise analysis (Appendix 3.1.1, 3.1.2, and 3.1.3). This was corroborated by network rankings: doxycycline (network SMD − 0.20; CI [− 0.39, − 0.01]; SUCRA 88.7; rank 1), glucosamine (− 0.15; [− 0.25, − 0.05]; 84.7; 2), and vitamin D (− 0.15; [− 0.27, − 0.03]; 83.4; 3) (Appendix 3.1.6, 3.1.7, and 3.1.8). A similar pattern was noted for the outcome measure of joint function, and the pairwise analysis detected favorable results for vitamin D and glucosamine (Appendix 3.2.1, 3.2.2, and 3.2.3). NMA further confirmed the positive functional effects for vitamin D (network SMD −0.18; CI [0.31, −0.06]; SUCRA 87.2; rank 1), glucosamine (−0.17; [−0.28, −0.07]; 84.8; 2), and chondroitin (−0.15; [−0.26, −0.03]; 78.6; 3) (Appendix 3.2.6, 3.2.7, and 3.2.8). Clustered ranking methods demonstrated that glucosamine and chondroitin were more effective than other DMOADs in slowing down structural changes, promoting pain management, and improving joint functions (Appendices 3.1.9, 3.2.9, 3.3).

Based on the NMA results (Appendix 4.3.2 and 4.3.3 and Fig. 4), withdrawals due to adverse events were significantly higher amongst those patients taking MMP inhibitors (network RR 3.80; CI [1.42, 10.14]; SUCRA 2.7; Rank 12), diacerein (2.14; [1.45, 3.18]; 10.9; 11), and doxycycline (1.69; [1.03, 2.75]; 22.1; 10). The clustered ranking with the structural outcome (Fig. 5) revealed high safety for glucosamine, chondroitin, bisphosphonate, strontium, and licofelone. For those agents with significantly negative safety profiles, the most frequent adverse events were musculoskeletal issues for MMP inhibitors and gastrointestinal discomforts for diacerein and doxycycline (Appendices 4.1, 4.2).

Figure 5
figure 5

Clustered ranking for the DMOADs on minimum JSW or cartilage volume and withdrawals due to adverse events. The plot is based on the clustered analysis of SUCRA values (horizontal and vertical axes values). Treatments lying in the upper right corner are considered to perform well for both outcomes. Each color represents a group of treatments that belong to the same cluster. ASU, avocado soy unsaponifiables; DMOADs, disease-modifying osteoarthritis drugs; iNOS, inducible nitric oxide synthase; JSW, joint space width; MMP, matrix metalloproteinase; SUCRA, surface under the cumulative ranking curves.

Full size image

Additional subgroup NMAs were undertaken to assess each outcome for knees and hips separately (Appendix 6.1). The structural modifying effect of glucosamine appeared statistically more powerful at the knee than the hip, while diacerein seemed the most effective agent to structural changes at the hip. Bisphosphonate and glucosamine have the best statistical results for relieving pain and improving hip joint function, respectively. In addition, the subgroup analysis was also performed on the glucosamine and chondroitin regarding dosages, dosing schedules, or product subtypes indicated no significant difference in outcomes (Appendix 6.3.1 and 6.3.2). The analysis of publication bias suggested that excluding outliers led to little change in the rankings (Appendix 6.2).

We graded the strength of our conclusions to be of moderate quality for most of the comparisons the outcomes assessed (details in Appendix 1.4 and Appendix 5).

DISCUSSION

In our review, we found moderate-quality evidence that glucosamine and chondroitin had statistically significant but marginally small structural and symptomatic effects, in addition to favorable safety profiles. Statistically, strontium displayed therapeutic benefit in terms of structure modification, while vitamin D displayed improvement in symptoms. Doxycycline also showed promising rankings in terms of structure and function, but its clinical use may be hindered by safety issues. The therapeutic effects of diacerein, bisphosphonate, iNOS inhibitors, ASU, and licofelone on OA were not as significant as the medications indicated above, while MMP inhibitors and vitamin E promoted little structural or symptomatic improvement.

To our knowledge, this study is one of the most comprehensive systematic reviews performed on a wide range of DMOAD candidates for the treatment of hip and knee OA. The predefined criterion for including trials allowed the assessment of the long-term effects. Most of the trials provided direct measurements, ensuring the reliability of the results. Although joint space narrowing from X-rays is a validated primary endpoint in assessing structural progression, we also exploited MRI findings,30, 33, 46, 47 for its high sensitivity. Pieces of information were maximally preserved when combinations of dosage groups and time points were employed to avoid multiple testing. Additional subgroup analyses assessed the relative merit of different time points, dosages, regimens, or product sources. Although knee and hip osteoarthritis may share similar etiology, presentations, and treatment guidelines,57 sensitivity analyses of joint subgroups were performed to explore possible distinctive effects.

It is worth noting that the magnitude of the effect sizes of all the included drugs in our study was probably too small to be clinically meaningful. We conclude that none of our included DMOAD candidates has convincing long-term disease-modifying abilities. However, for most of the included medications, the analyses suggest improved structural modifying effects over time, though these effects were not statistically significant. Future studies should include long-term interventions to assess potential structural modifying effects for DMOADs.

In addition, distinctions between different drugs on the hip and knee joint suggest that OA could display distinct pathophysiological mechanisms and respond differentially to pharmacological interventions at different joint types.59 In our study, it was interesting to see that symptomatic improvement was associated with structural changes for some agents (glucosamine, chondroitin), but not for others (vitamin D, strontium ranelate). While the dissociation between the radiographic stage of OA and the severity of symptoms has long been recognized,60, 61 this remains a subject of debate.62 Some evidence suggests the involvement of different etiological pathways as the causes of structural damage and symptomology.63, 64

Previous meta-analyses have suggested conflicting or ambiguous structural and symptomatic efficacies for glucosamine and chondroitin in OA treatment,5–8 which may largely be attributable to the different inclusion criteria. In our study, with the duration of 1 year and above, statistically significant top rankings were achieved for both medications, although the clinical meaningfulness needs further investigation. Doxycycline, believed to have cartilage protective effects through collagen degradation inhibition,65 was ranked highly by our study in terms of efficacies; however, it showed an unsatisfactory safety profile. Interestingly, a previous meta-analysis,66 which included the short-term application of doxycycline (< 1 year), showed reduced benefits and increased safety when compared to our study. Strontium ranelate presented promising structural modifying effects in our study and is believed to suppress subchondral bone resorption and stimulate cartilage matrix formation.67 However, strontium had no symptomatic relief; this may be explained by the different etiological pathways involved in OA structural damage and symptomology.64 While our safety profile of strontium was acceptable for daily treatment, caution must be exercised as the drug appears to cause serious cardiovascular side effects when treating osteoporosis.68, 69 In contrast, vitamin D, as shown in previous studies5, 70, 71 and this study, showed few structural effects, but demonstrated a potential to improve symptoms. Hence, combining vitamin D with structure-modifying agents such as glucosamine could be theoretically beneficial, though studies need to be done to verify this. It is possible that particular drug regimens could be combined to target different pathways for the optimized treatment of OA. Diacerein, a purified compound with an anthraquinonic structure, was shown to be beneficial for cartilage by inhibiting cytokine production and activation,41, 42 and a recent meta-analysis,72 including some short-term studies (< 1 year), suggested that diacerein had minimal effects on pain reduction and questionable relevance to joint space narrowing. Our study confirmed that for long-term treatment diacerein was not the best agent to choose. Although it provided structural effects at the hip joint, safety issues could limit its use.73 Agents suppressing bone turnover, including bisphosphonates, have been associated with fewer subchondral bone lesions,40 a source of osteoarthritic pain. Despite limited evidence74 that bisphosphonates are effective in the treatment of OA, our study suggests hip pain reduction for alendronate. It would be interesting to see if studies could differentiate the effects of different bisphosphonates on hip and knee joints, and for those with both OA and osteoporosis, bisphosphonates such as alendronate may be a first-line therapy. With regard to the other agents investigated, iNOS inhibitors, ASU, licofelone, MMP inhibitors, and vitamin E, all showed no significant outcomes in either structural progression or symptomatic relief in our study.

Our study has several limitations. First, several DOMADs9 were only recently. For five candidate drugs, we identified only one eligible trial for each,21, 22, 29, 34, 35 thereby making our analyses susceptible to false-negative errors. Secondly, we were unable to perform a stratified analysis of OA stage (Kellgren-Lawrence). Although some of the trials22, 29–32, 35–37, 41, 43, 44, 48, 50, 52, 53 provided information about patients’ OA stages, none of them stratified the outcome data according to the staging. Third, despite the assumption that all trials were jointly randomizable for NMA, the potential barriers to transitivity such as the various durations of follow-up and different usage of medications could make our results questionable. In this study, we combined the dosage and time groups to minimize this potential bias. The combination of dosages may represent a slight underestimation of the desired SD (84), and the combination of time points conservatively (set correlations at 1 comparing to the widely used 0.5) may slightly underestimate the significance. Additional analyses were made with correlations set at 0 for the time points combination, with no obvious change to the effects. Fourth, there were few publications with direct comparisons (14%) between medications, and as a result, the medium heterogeneity and inconsistency presented in our study should be interpreted with caution, given the fact that a star-shaped network limits the possibility to detect both. However, we have provided the confidence intervals for the heterogeneity estimates for the readers to assess. Caution also needs to be employed when interpreting the clustered ranking plots, since they are not based on joint analyses of the two completely independent outcomes. Considering the limited clinical meaningfulness of candidate drugs, clustered ranking plots can be misleading if only taking into account the statistical ranking.

In conclusion, none of the 12 classes of included drugs could be confirmed as clinically effective DMOADs, although some of them showed promising potential in this regard. It is hoped that this review will help physicians, patients, and researchers make informed decisions regarding disease-modifying OA candidate drugs for treatment, clinical studies, and basic research into OA.