Search Strategy
We conducted electronic searches for relevant randomized controlled trials
in the MEDLINE and EMBASE databases (from January 1966 to December 2001) and
the Cochrane Controlled Trials
Register9. In
combination with a search strategy that has optimal sensitivity in identifying
randomized controlled
trials10,
"osteoarthritis," "hyaluronic acid," and their
synonyms were entered as textwords for searches. In addition, supplemental
issues of Arthritis and Rheumatism, British Journal of Rheumatology,
Journal of Rheumatology, and Acta Orthopaedica Scandinavica,
published between January 1970 and December 2001, were searched by hand. The
references cited in review articles and retrieved articles were also reviewed.
Unpublished trials were not searched, and the search was restricted to
English-language reports.
Trial Inclusion Criteria
Only single-blind or double-blind randomized controlled trials that
compared the therapeutic effect of intra-articular injection of hyaluronic
acid with that of intra-articular injection of a placebo to treat
osteoarthritis of the knee were included in this meta-analysis. Outcome end
points for pain or function and quantitative data on therapeutic effects had
to be available. Comparison of the therapeutic effect of intra-articular
injection of hyaluronic acid with that of intra-articular injection of a
placebo is essential to demonstrate the true efficacy of hyaluronic acid.
Data Collection
All relevant randomized controlled trials were selected, and all relevant
data were extracted from the text, tables, and figures, by two investigators
working independently using a standardized form. Differences in data
interpretation were resolved by discussion between the investigators. We did
not calculate the interrater and intrarater agreement regarding trial
selection and data extraction. Authorship, time periods of enrollment, details
of treatment protocols, demographic data on the patients, and other
information in the randomized controlled trials were closely examined to avoid
inclusion of data on patients who had been described in multiple reports and
to ensure inclusion of the most complete data. The authors, institutions,
journals, and identities of the treatment groups were not masked during trial
selection and data extraction by the investigators. Available evidence
indicates that, for purposes of meta-analysis, such masking imposes
substantial burdens without significantly altering the results of a review (p
= 0.35)11. Masking
also hinders efforts to avoid duplication and ensure completeness of data.
The methodological quality of each trial was scored with use of a 28-point
checklist (see Appendix), which had been
validated12 and has
been used in the development of evidence-based guidelines for the management
of osteoarthritis of the knee by the European League Against
Rheumatism3. Pain
intensity and patient function at different times, which were measured with
validated and reliable outcome
instruments13 (see
Appendix), were recorded for statistical analysis. In addition, adverse events
were recorded and were classified as major or minor according to the
classification in each trial. Study-level data, but not individual patient
data, were used in the statistical analysis.
Outcome Measures of Efficacy and Safety of Hyaluronic Acid
We classified the outcome end points for efficacy evaluation into three
categories: pain with activities (walking, climbing, and so on), pain without
activities (at rest, at night, and so on), and function (see Appendix). To
incorporate efficacy outcomes that were measured with different types of
outcome instruments and were evaluated at different times after treatment, we
defined efficacy scores with a modification of the method of de Craen et
al.14 and Zhang et
al.15,16.
First, the extracted data were transformed in the following manner. The pain
intensity difference was considered to be the difference in pain intensity
between one time-point and the baseline. The sum of the pain intensity
differences (SPID) was the sum of the average of two consecutive pain
intensity differences multiplied by the time-interval between two
time-points—that is, the area under the pain intensity
difference-versus-time curve. The standard deviation of the sum of the pain
intensity differences was the sum of the average of standard deviations of two
consecutive pain intensity differences multiplied by the time-interval between
two time-points. For a trial that presented mean values without measures of
variability, the standard deviation of the sum of pain intensity differences
was calculated by multiplying the mean for the trial arm by the median
coefficient of variation from other included trial arms that used the same
category of
outcome17. The
function index difference, the sum of function index differences (SFID), and
the standard deviation of the sum of function index differences were defined
in the same manner. Then, we defined the efficacy scores, SPID%, adjusted
SPID% (ASPID%), and peak PID%, as follows:
SPID%=sum of pain intensity
differences(maximum scale of pain
intensity×trial duration)×100%
SPID%, an integrated score with time as a dimension, represents the overall
efficacy in the trial period.
ASPID%=sum of pain intensity
differences(baseline intensity×trial
duration)×100%
ASPID%, an integrated score with time as a dimension, represents the overall
efficacy in the trial period after adjustment for the baseline pain intensity.
Peak PID%=maximum pain intensity
differencemaximum scale of pain
intensity×100%
Peak PID% represents the maximum efficacy during the trial.
For function, SFID%, adjusted SFID% (ASFID%), and peak FID% were defined in
the same manner.
In this meta-analysis, the outcome measures of the efficacy of hyaluronic
acid were the differences in mean efficacy scores between the hyaluronic acid
and placebo groups. The outcome measure of the safety of hyaluronic acid was
the relative risk of adverse events.
Statistical Analysis
Pooled Efficacy and Safety
For efficacy outcomes, the METAN program with methods of unstandardized
mean difference18
was used to calculate pooled mean differences in the efficacy scores and
corresponding 95% confidence intervals. For safety outcomes, the METAN program
with the methods of Mantel and
Haenszel19 was used
to calculate pooled relative risk and corresponding 95% confidence intervals.
The chi-square test for Q statistics was used to test between-study
heterogeneity. If there was significant between-study heterogeneity (a p value
for chi-square test of =0.1), the random-effects models of the DerSimonian
and Laird methods20
were used to calculate pooled mean differences, pooled relative risk, and
corresponding 95% confidence intervals. If there was no significant
between-study heterogeneity (a p value for chi-square test of >0.1),
fixed-effects models of the inverse variance method were applied in the
calculation.
Subgroup Analysis and Meta-Regression Analysis
Subgroup analysis was planned and conducted to investigate the effects of
several categorical attributes on the estimates of hyaluronic acid efficacy
and to assess between-study heterogeneity within each subgroup of trials.
These categorical attributes included blinding status, single-center or
multicenter trial, intention-to-treat analysis (meaning that the data on all
study subjects were analyzed according to the groups to which they were
initially allocated), use of escape analgesics, mean age of the patients
(sixty-five years or less, or more than sixty-five years), inclusion of the
most advanced radiographic stage of osteoarthritis (defined as a complete loss
of joint space), effusion as an inclusion or exclusion criterion, trial
duration (twelve weeks or less, or more than twelve weeks), sample size (100
knees or fewer, or more than 100 knees), and whether the trial was
industry-funded.
Meta-regression analysis, performed with use of the METAREG program with
the method-of-moments estimator
method21, was
planned and conducted to investigate the effects of several continuous
attributes on the estimates of hyaluronic acid efficacy; these included the
year of publication, quality score of the methodology, molecular weight of the
hyaluronic acid, mean age of the patients, trial duration, and sample size. We
also calculated residual heterogeneity
variance2, which is
the residual between-study
variance2 after
accounting for the heterogeneity resulting from the difference in each
attribute. The meta-regression analysis was limited to univariate
analysis.
Assessment of Publication Bias
Funnel plots of the mean differences in the efficacy scores for all
included trials were constructed to assess the degree of publication bias.
Included Trials
Figure 1 shows the steps and
criteria for the search strategy and the number of trials evaluated at each
stage of the systematic review. A total of 665 potentially relevant studies
were identified and screened for retrieval: 647 were identified by electronic
searches and eighteen, by hand-searching of key journals. After the exclusion
of non-trial studies, sixty-one relevant trials were retrieved for more
detailed evaluation. Of these trials, thirty-six were excluded because of no
placebo control, no randomization, an open design, no pain or functional
outcomes, or duplication. Of the remaining twenty-five potentially appropriate
randomized controlled
trials8,22-45,
only twenty provided quantitative data on therapeutic effects. These twenty
trials were included in the meta-analysis (see
Appendix)8,22-34,39,41-45.
The twenty included trials provided data on therapeutic effects for a total
of 1647 randomly assigned knees (818 knees treated with hyaluronic acid
injection and 829 treated with placebo injection) and data on safety for a
total of 2252 knees (1141 knees treated with hyaluronic acid injection and
1111 treated with placebo injection). The twenty included trials differed in
many dimensions, including study design, characteristics of the patients,
inclusion and exclusion criteria, type of hyaluronic acid product, evaluated
outcome end points, outcome instruments, and status with respect to industry
funding. Of the hyaluronic acid products, only Synvisc (hylan G-F 20; Genzyme,
Cambridge, Massachusetts) is a cross-linked hyaluronic acid. Each type of
hyaluronic acid product has its own recommended dose schedule, including the
amount and number of doses. All of the included trials used validated and
reliable outcome
instruments13 (see
Appendix). Allocation concealment was unclear in all included trials.
Pooled Efficacy and Safety
With regard to pain with activities in all included trials, the pooled mean
difference for SPID%, which represents the overall hyaluronic acid efficacy in
the trial period, was 7.9% (95% confidence interval, 4.1% to 11.7%). The
pooled mean difference for ASPID%, which represents the overall hyaluronic
acid efficacy in the trial period after adjustment for the baseline pain
intensity, was 13.4% (95% confidence interval, 5.5% to 21.3%). The pooled mean
difference for peak PID%, which represents the maximum hyaluronic acid
efficacy during the trial, was 9.9% (95% confidence interval, 4.8% to 15.0%)
(see Appendix).
Both the trials involving cross-linked hyaluronic acid and those involving
non-cross-linked hyaluronic acid had significantly positive pooled mean
differences in the efficacy scores at each outcome end point (see Appendix).
The trials involving cross-linked hyaluronic acid had much greater pooled mean
differences than did those involving non-cross-linked hyaluronic acid (pain
with activities: 23.6% compared with 5.4% for SPID%, 34.8% compared with 8.7%
for ASPID%, and 27.1% compared with 7.4% for peak PID%; function: 21.9%
compared with 5.3% for SFID%, 38.3% compared with 11.7% for ASFID%, and 26.8%
compared with 8.2% for peak FID%). However, there was significant
between-study heterogeneity in the estimates of hyaluronic acid efficacy among
the trials involving non-cross-linked hyaluronic acid (p = 0.1, chi-square
test) (see Appendix).
No mortality was associated with hyaluronic acid injection. Major adverse
events were noted in three of 1002 knees treated with injection of
non-cross-linked hyaluronic acid (severe swelling in one, vasculitis in one,
and hypersensitivity reaction in one) and in one of 139 knees treated with
injection of cross-linked hyaluronic acid (Synvisc) (a painful acute local
reaction)22,34,42.
Minor adverse events consisted of a transient mild increase in local pain or
swelling. The pooled relative risk of minor adverse events for all included
trials was 1.19 (95% confidence interval, 1.01 to 1.41). There was no
significant between-study heterogeneity in the relative risk of minor adverse
events (p = 0.585, chi-square test).
Subgroup Analysis and Meta-Regression Analysis
Considering a potential so-called drug-class difference between
non-cross-linked and cross-linked hyaluronic acid products, we conducted a
subgroup analysis and a meta-regression analysis of only the trials involving
non-cross-linked hyaluronic acid. Subgroup analysis of pain with activities
showed no significant between-study heterogeneity (p > 0.1, chi-square
test) in the mean difference for at least one efficacy score within most
subgroups of trials. It indicated that the data from the different trials
within each of these subgroups can be meaningfully combined and thus the
pooled estimates of these efficacy outcomes may be more realistic (see
Appendix). Similar results were found in the subgroup analysis of the mean
differences in the efficacy scores for pain without activities and function
(see Appendix).
Subgroup analysis of the efficacy scores for the three outcome end points
showed that all of the single-blind trials, double-blind trials, single-center
trials, and multicenter trials had significantly positive pooled mean
differences in pain with activities. Moreover, the single-blind trials had
greater pooled mean differences than did the double-blind trials, and the
single-center trials had greater pooled mean differences than did the
multicenter trials, indicating that the single-blind trials and the
single-center trials provided higher estimates of hyaluronic acid efficacy.
The trials with acetaminophen as an escape analgesic had smaller pooled mean
differences than did those without escape analgesics, indicating that
introduction of acetaminophen as an escape analgesic in trials results in
lower estimates of hyaluronic acid efficacy.
The trials with a mean patient age of sixty-five years or less had
significantly positive pooled mean differences, whereas those with a mean
patient age of more than sixty-five years did not. Moreover, the trials with a
mean patient age of more than sixty-five years had smaller pooled mean
differences than did those with a mean patient age of sixty-five years or
less, indicating that hyaluronic acid may be less effective for patients over
sixty-five years of age. The trials that included patients with the most
advanced radiographic stage of osteoarthritis had smaller pooled mean
differences than did those in which no patient had the most advanced
radiographic stage of osteoarthritis, indicating that hyaluronic acid may be
less effective for the most advanced radiographic stage of osteoarthritis. In
addition, while both the trials with a sample size of 100 or fewer and those
with a sample size of more than 100 had significantly positive pooled mean
differences, the trials with a sample size of 100 or fewer had greater pooled
mean differences than did those with the larger sample size. This indicated
that there may be a small-trial bias favoring the estimation of hyaluronic
acid efficacy. The industry-funded trials were found to have smaller pooled
mean differences in the scores for pain with activities and function than did
the non-industry-funded trials. However, the industry-funded trials had
greater pooled mean differences in the scores for pain without activities than
did the non-industry-funded trials (see Appendix).
Meta-regression analysis of the efficacy scores for the three outcome end
points showed that nearly all three estimated regression coefficients of three
covariates (quality score of the methodology, mean age of the patients, and
sample size) were negative, indicating that these three attributes were
negatively correlated with the estimates of hyaluronic acid efficacy. The
analysis suggested that the trials with lower methodological quality provided
higher estimates of hyaluronic acid efficacy, that hyaluronic acid may be less
effective for older patients with osteoarthritis, and that there may be a
small-trial bias favoring the estimation of hyaluronic acid efficacy (see
Appendix).
Publication Bias
The funnel plots of the mean differences in the efficacy scores for pain
with activities in all included trials showed that the data are clustered
symmetrically around the vertical dashed line
(Fig. 2). This indicated that
publication bias was unlikely to have had an important influence on the
overall analysis of hyaluronic acid efficacy.
This meta-analysis of randomized controlled trials confirmed that
intra-articular injection of cross-linked hyaluronic acid (Synvisc) and of
non-cross-linked hyaluronic acid (Hyalgan [FIDIA, Padua, Italy], Orthovisc
[Anika Therapeutics, Woburn, Massachusetts], Artz [Seikagaku, Tokyo, Japan],
and BioHy [Savient, East Brunswick, New Jersey]) can decrease symptoms of
osteoarthritis of the knee. We found significant improvements in pain and
functional outcomes with few adverse events. The trials involving cross-linked
hyaluronic acid had much greater pooled estimates of hyaluronic acid efficacy
than did the trials involving non-cross-linked hyaluronic acid. However, there
was significant between-study heterogeneity in the estimates of hyaluronic
acid efficacy. In addition, whether cross-linked hyaluronic acid has greater
therapeutic efficacy than non-cross-linked hyaluronic acid is
controversial38,46.
Therefore, more randomized controlled trials comparing different types of
hyaluronic acid products are needed to clarify the effects of cross-linked
status on hyaluronic acid efficacy.
In the subgroup analysis and meta-regression analysis of the trials
involving non-cross-linked hyaluronic acid, we found that all of the
single-blind trials, double-blind trials, single-center trials, and
multicenter trials showed a significant decrease in pain with activities.
Moreover, lower methodological quality such as a single-blind or single-center
design resulted in higher estimates of hyaluronic acid efficacy. In addition,
we found that introduction of acetaminophen as an escape analgesic in trials
resulted in lower estimates of hyaluronic acid efficacy. This may be due to
the additional therapeutic effect of acetaminophen in the placebo groups.
We found that the patients who were older than sixty-five years of age and
those with the most advanced radiographic stage of osteoarthritis (complete
loss of joint space) were less likely to benefit from intra-articular
injection of hyaluronic acid. Understanding the differences in hyaluronic acid
efficacy among different patient populations is important when selecting
patients for this therapy.
Both the small trials and the large trials showed significant improvements
in pain and functional outcomes. However, there may be a small-trial bias
favoring the estimation of hyaluronic acid efficacy, suggesting that some
small trials that did not show a positive therapeutic effect of hyaluronic
acid may not have been published. This may introduce the likelihood that the
effect is skewed in the positive direction. Furthermore, we found that the
study's status with regard to industry funding did not have a consistent
effect on the estimates of hyaluronic acid efficacy based on the three outcome
end points, although there is now reasonable evidence that the funding source
is strongly associated with published outcomes and publication bias in
biomedical
research47,48.
This meta-analysis showed that intra-articular injection of hyaluronic acid
was associated with few adverse events. Despite the low rate of complications,
injection of cross-linked hyaluronic acid may be associated with a painful
acute local reaction, which should be considered a major adverse
event49-52.
However, the rate of painful acute local reactions to Synvisc injection in
this meta-analysis (one of 139 knees) was lower than those in some other
reports49-52,
in which the rates have ranged between 2% and 8%. One possible reason for this
is that all of the three Synvisc trials in this meta-analysis were
industry-funded. There is now reasonable evidence that adverse events may be
minimally reported in industry-funded
trials47,48.
The outcome measures of hyaluronic acid efficacy used in this meta-analysis
have several advantages. First, they can incorporate efficacy outcomes that
are measured by different types of outcome instruments and are evaluated at
different times after treatment. Second, the mean differences in the efficacy
scores used in this meta-analysis have more comprehensive clinical meanings
than does the so-called effect size used in other meta-analyses. With use of
the mean differences in the efficacy scores, we provided easily understandable
evidence regarding the magnitude of therapeutic effects, which can facilitate
evidence-based decisions in clinical practice. Third, in this meta-analysis,
different types of outcome measures had different clinical meanings: the mean
difference for SPID% (or SFID%) represents the overall hyaluronic acid
efficacy in the trial period, the mean difference for ASPID% (or ASFID%)
represents the overall hyaluronic acid efficacy in the trial period after
adjustment for the baseline pain intensity (or function index), and the mean
difference for peak PID% (or peak FID%) represents the maximum hyaluronic acid
efficacy during the trial. All three types of outcome measures are important
for efficacy evaluation of slow-acting drugs used to treat the symptoms of
osteoarthritis because the therapeutic effects of these drugs usually last
several months. Among these efficacy scores, SPID% and peak PID% have been
previously
described14-16.
Fourth, we placed the outcome end points for efficacy evaluation into three
categories: pain with activities, pain without activities, and function.
Because these end points are different, it is not suitable to pool them
together. Clarifying the therapeutic effect of hyaluronic acid on different
symptoms of osteoarthritis of the knee may help clinicians to more
appropriately prescribe intra-articular injection of hyaluronic acid to
relieve specific symptoms.
This meta-analysis has several limitations. First, unpublished trials were
not searched. Since trials with "negative" results tend to be
unpublished53,54,
exclusion of unpublished trials may introduce the likelihood that the effect
is skewed in the positive direction because of publication bias. However, the
funnel plots in this meta-analysis showed the data to be clustered
symmetrically around a vertical line. This indicated that publication bias was
unlikely to have had an important influence on the overall analysis of
hyaluronic acid efficacy. Second, interrater and intrarater agreement was not
calculated, and this limits the confidence that the reader can have in some of
the assessments. Third, not all validated outcome instruments fit neatly into
categories like "pain with activities," "pain without
activities," and "function." This classification might be
somewhat arbitrary and might affect results, depending on how it was
performed. However, the classification was done independently by two
investigators and the differences in classification were resolved by
discussion between the investigators, which could minimize the likelihood of
misclassification. Fourth, the twenty included trials differed in many
dimensions, including study design, characteristics of the patients, inclusion
and exclusion criteria, type of hyaluronic acid product, evaluated outcome end
points, outcome instruments, and status with regard to industry funding.
Meanwhile, there was significant between-study heterogeneity in the estimates
of hyaluronic acid efficacy. The heterogeneity of included trials is a common
limitation of this and other meta-analyses summarizing treatment effects.
Therefore, we conducted a subgroup analysis, in which we found the studies
within the same subgroup to be similar enough to allow combination. Fifth, the
results of subgroup analysis, which were calculated from study-level data but
not from individual patient data and thus were hypothesis-generating only,
should be applied cautiously when substantiating any inference.
In conclusion, this meta-analysis of randomized controlled trials confirmed
that intra-articular injection of cross-linked hyaluronic acid and
non-cross-linked hyaluronic acid can decrease symptoms of osteoarthritis of
the knee. We found significant improvements in pain and functional outcomes
with few adverse events. However, the patients over sixty-five years of age
and those with the most advanced radiographic stage of osteoarthritis
(complete loss of joint space) were less likely to benefit from
intra-articular injection of hyaluronic acid. More well-designed randomized
controlled trials with high methodological quality are needed to resolve the
continued uncertainty about the therapeutic effects of different types of
hyaluronic acid products on osteoarthritis of the knee in various clinical
situations and patient populations.
A figure demonstrating the mean differences in the efficacy scores for pain
with activities between the hyaluronic acid and placebo groups in all of the
individual trials as well as tables showing (1) the checklist for the
assessment of the methodological quality, (2) characteristics of the
randomized controlled trials included in the meta-analysis, (3) subgroup
analysis of the mean differences in the efficacy scores for pain with
activities, pain without activities, and function, and (4) a meta-regression
analysis of the mean differences in the efficacy scores for different outcome
end points are available with the electronic versions of this article, on our
web site at
(go to the article citation and click on "Supplementary
Material"), and on our quarterly CD-ROM (call our subscription
department, at 781-449-9780, to order the CD-ROM).