Surgical repairs for the treatment of anterior instability of the
shoulder can be performed with use of either arthroscopic or open approaches.
The classic open Bankart procedure includes incision of the subscapularis
tendon and capsule to expose the anterior aspect of the labrum and capsule for
secure reattachment to the glenoid rim with suture passed through bone
tunnels1. Multiple
variations have been used to expose the joint, to repair the ligaments, and to
treat associated capsular
laxity2-11.
Open approaches have yielded consistently low rates of recurrent
instability12-15.
Arthroscopic techniques have also been utilized to treat anterior
instability of the shoulder. Results have been reported after the use of
staples16,
transglenoid
sutures17,18,
bioabsorbable
tacks19, and suture
anchors20-22,
with proponents describing benefits related to smaller incisions, less loss of
motion, lower risk of subscapularis failure, quicker return to sports, and
higher patient satisfaction.
Despite the advocates for each approach, the evidence regarding the
relative effectiveness of open and arthroscopic treatment of anterior
glenohumeral instability remains unclear. Two previous meta-analyses have been
published23,24,
but these studies did not:
1. include all available published and unpublished series.
2. utilize both fixed-effects and random-effects
models25,26.
Fixed-effects models assume homogeneity—i.e., that every study included
in the analysis evaluated the same treatment effect. Random-effects models
assume that the treatment effect may have been different in each study.
Heterogeneity can result from the inclusion of different populations, from the
application of different surgical techniques, from differences in followup, or
from differences in outcome measurement. If bias or heterogeneity is present,
the fixed and random-effects models may lead to different conclusions.
3. employ funnel plots to discover possible publication
bias27. Funnel
plots are a graphical representation of treatment effect versus sample size.
Optimally, the treatment effect should not change with sample size; funnel
plots reveal asymmetry in this relationship. Asymmetry (e.g., the relative
absence of small studies with negative findings) can bias conclusions away
from the true treatment effect.
4. include a number-needed-to-treat analysis to demonstrate the clinical
relevance of the different
outcomes28. The
number needed to treat describes the number of patients one would need to
treat with one technique (e.g., open repair) rather than the alternative
technique (e.g., arthroscopic repair) in order to prevent a single event
(e.g., recurrent dislocation).
5. compare what has been proposed as the "gold standard"
arthroscopic technique—i.e., repair with suture
anchors21,29-32—with
open approaches.
6. include an analysis of the effect of study quality on the observed
treatment effect. The reliability and utility of a metaanalysis depend in
large measure on the quality of the primary source studies that are evaluated.
Non-randomized trials are expected to have a variety of biases compared with
randomized controlled
trials33. A
meta-analysis that compares randomized with non-randomized trials offers one
window into this effect.
Because of the clinical importance of the question and the incompleteness
of the published systematic reviews, we performed a systematic review and
rigorous meta-analysis of all published and presented literature comparing
open and arthroscopic approaches to the treatment of anterior shoulder
instability. In this study, we tested the hypothesis that the literature
demonstrates significant differences between the effectiveness of arthroscopic
treatment and that of open treatment of anterior shoulder instability,
specifically with respect to (1) restoration of shoulder stability (indicated
by the absence of recurrent dislocation, subluxation, or recurrent
apprehension and no need for a reoperation), (2) the rate of recurrent
dislocation alone, (3) the rate of reoperations for instability alone, (4) the
ability of patients to return to work or sports, and (5) the Rowe scores. We
further hypothesized that the quality of the study influences the results.
Finally, we tested the hypothesis that the specific arthroscopic technique
influences the clinical outcome.
Inclusion and Exclusion Criteria
We identified articles and abstracts that met the following
inclusion criteria: (1) comparison of one or more arthroscopic techniques with
open techniques, (2) evaluation of patients who had predominantly anterior
instability (including those with a first-time dislocation, those who had
attempted rehabilitation, and those with multidirectional laxity but a
clinical diagnosis of anterior instability), and (3) use of an anterior
soft-tissue repair (i.e., a Bankart procedure and/or capsular shift). We
included retrospective comparative studies and observational case-control
trials as there is a paucity of randomized, controlled trials pertaining to
this topic; inclusion of these studies, which were analyzed separately from
the randomized trials, permitted evaluation of the hypothesis regarding the
effect of study quality on the size of the observed effect.
Studies were excluded if there was no comparison group, if a
bone-block-type of procedure was used, or if the predominant direction of the
instability was posterior. We did not exclude papers that included patients
with a first-time dislocation or those with glenoid defects or a Hill-Sachs
lesion.
Identification of Studies and Publication Bias
A search of the Medline database on
PubMed34, for the
years 1966 to November 2004, was conducted with use of five combinations of
search terms: (1) "Bankart," (2) "shoulder AND
instability," (3) "shoulder AND dislocation AND anterior,"
(4) "shoulder AND capsulorrhaphy," and (5) "shoulder AND
capsular shift." Following this, a search of the Cochrane Collaboration
Library35 was
performed with use of the same combination of search terms. We also performed
an online search of the Arthroscopy Association of North America (AANA) annual
meetings abstracts from 1998 to 2004 and the International Society of
Arthroscopy, Knee Surgery and Orthopaedic Sports Medicine (ISAKOS) meeting
abstracts available for 1997, 1999, and 2001. We performed a manual search of
the American Academy of Orthopaedic Surgeons (AAOS) annual meeting abstracts
from 2000 to 2005 and the American Shoulder and Elbow Surgeons (ASES) annual
open meeting abstracts from 1996 to 2005. Additional strategies included
searching the citations of several review articles and a prominent textbook,
The
Shoulder8. We
also contacted three subject-matter experts who were not involved in this
study in order to maximize the likelihood that all relevant literature would
be identified. In addition, attempts were made to contact authors of several
articles used in the study to obtain additional, unpublished data.
Two orthopaedic surgeons independently reviewed the titles to identify
articles that might meet our eligibility criteria. These abstracts or articles
were then collected and reviewed to determine if they were appropriate for
inclusion.
Data Extraction
Two coauthors (T.R.L. and A.K.F.) abstracted all relevant available
information regarding the type of study, the population, the intervention, and
the end points from each article. Any differences in the data that the two
collected were reconciled by consensus. Demographic data included age, sex,
hand dominance, number of preoperative dislocations, time to surgery, and
duration of follow-up. Surgical data included the technique that was used.
Outcome data included stability end points (recurrent dislocation,
subluxation, and/or apprehension and/or a reoperation for recurrent
instability), the Rowe score (a composite score in which up to 50 points is
assigned for stability, up to 30 points is assigned for function, and up to 20
points is assigned for
motion)1, the
ability to return to work and/or sports, and the range of motion.
Complications were recorded as presented in the report.
Assessment of Methodological Quality and Grouping of Studies
Each study was classified into one of three groups on the basis of the
overall study quality. Grouping was based on the Levels of Evidence statement
in the Instructions to Authors of The Journal of Bone and Joint
Surgery. Level-I therapeutic studies include high-quality randomized
controlled trials with =80% follow-up, blinding, and narrow confidence
intervals. Level-II therapeutic studies include lesser-quality randomized
clinical trials (no blinding, <80% follow-up, or improper randomization)
and prospective comparative studies. Level-III therapeutic studies include
case-control series and retrospective comparative series, and Level-IV and V
studies include case series and expert opinion, respectively. In our study,
the "best" group included Level-I and II randomized clinical
trials. The "good" group included Level-II prospective comparative
studies and Level III case-control studies (controlled clinical trials). The
"fair" group included Level-III series in which the
arthroscopically treated group differed from the group treated with an open
technique as treatment was indicated by the pathological findings at the time
of surgery or the inability to perform a secure repair with use of an
arthroscopic technique (finding-dictated trials). Placing the studies into one
of these three groups allowed us to perform a subgroup analysis of the
influence of study quality on the effect estimate.
For subgroup analysis of the influence of the specific arthroscopic
technique on the effect estimate, studies were classified according to the
arthroscopic intervention that had been used, regardless of whether suture
anchors or bone tunnels had been employed for the open procedures. Three
groups were established according to whether the arthroscopic repair had been
done with suture anchors, bioabsorbable tacks, or transglenoid sutures. Three
studies36-38
were excluded from this portion of the analysis because they could not be
placed into one of these three groups. The authors of those studies had used
multiple arthroscopic techniques and did not report the results of the
techniques independently.
Meta-Analysis
For binary outcomes (dislocation, subluxation, apprehension, reoperation,
and return to work and/or sports), the relative risk and 95% confidence
interval were calculated with use of the Cochrane Collaboration's
meta-analysis program, Review Manager
4.239. Data were
then pooled within each subgroup and across all studies with use of both
fixed-effects and random-effects
models25,26.
When there was no difference between the findings derived with the two models,
we reported the results from the fixed-effects model. If there was a
difference, we reported the results of both models. Differences between the
models occurred when there was heterogeneity of results across studies and
could often be explained by variations among subgroups. A random-effects model
typically resulted in a more conservative estimate, meaning that it was less
likely to show a difference between treatment approaches than was a
fixed-effects model. If an effect was present, significant differences should
ideally be shown with use of both models. Results were presented with use of
forest plots summarizing effect size estimates and 95% confidence intervals.
If an end point was not reported in a study, it was excluded from the
analysis.
For continuous outcomes (Rowe scores), the standardized mean difference and
95% confidence interval were calculated with use of Review Manager
4.239. Again, data
were pooled within each subgroup and across all studies with use of both
fixed-effects and random-effects models. Rowe scores were reported in twelve
articles, but the reporting was inconsistent. In four of the
articles36,40-42,
the means and standard deviations were included, and those are presented in
our analysis. In five
articles43-47,
the median values were presented along with the variance; the technique of
Tomlinson and
Beyene48 was used
to convert those data to the equivalent mean and standard deviation values for
inclusion in our analysis. In three
articles30,37,49,
the mean values were given without standard deviations so the data could not
be included in the statistical analysis. We analyzed our data twice. First we
analyzed the initial group of four studies, and then we pooled both the
initial group of four studies and the adjusted group of five studies. We
reported only the results of the second, pooled analysis unless there was a
difference in the results of the two analyses.
The standard chi-square test for heterogeneity (Q) across study results for
each outcome was performed, with calculation of the accompanying I2
statistic50. The
major advantage in using this statistic instead of the more commonly used Q
statistic is that it provides an estimate of the proportion of total variation
in study results that is caused by heterogeneity, rather than sampling error.
I2 may be roughly interpreted as absent (0% to 25%), low (25.1% to
50%), moderate (50.1% to 75%), or high (75.1% to 100%)
heterogeneity50.
Funnel plots were constructed to examine the possibility of publication
bias27, with the
relative risk result from each study on the x axis and the standard error on
the y axis. The standard error reflects differences among studies of different
sample sizes. Publication bias would be suggested by asymmetry in this plot
(e.g., with a relative absence of publications showing small effects with
smaller sample
sizes)51.
In addition, a number needed to treat was calculated for each binary
outcome. The difference in relative risks and the 95% confidence interval of
the difference was calculated for each study and then pooled. The inverse of
this risk difference and the upper and lower confidence intervals were used to
calculate the number needed to treat with a 95% confidence interval. Results
can be reported as the number needed to harm or the number needed to benefit,
and they indicate how many procedures need to be performed with one approach
to prevent one adverse event (or conversely, to result in one good outcome)
from occurring with the alternative
approach28.
We attempted to estimate differences in the range of motion, but
inadequacies in the reporting of the data prevented analysis of this end
point. Authors reported the absolute loss of
motion30,31,36,40,44-46,52-55,
the percentage of normal
motion49, or the
range of
motion38,43.
No variances, standard deviations, or ranges were presented to allow an
estimate of effect differences.
Literature Search
(Fig. 1)
The PubMed search identified 2108 studies, and fourteen of these met our
inclusion criteria. One of them was excluded as a result of duplicate
publication56,
leaving thirteen articles. The search of the Cochrane Collaboration identified
ninety-four studies, ten of which met our inclusion criteria. Four had not
been identified by the PubMed
search46,57-59.
One of these was excluded as a result of duplicate
publication46, and
another59 was
excluded because of inadequate data. Our search of the AANA and ISAKOS
abstracts identified three abstracts, two of which met our inclusion criteria.
Both articles were later published and were identified by the PubMed
search44,53.
The third was excluded because it included patients in whom a bone graft had
been placed on the anteroinferior aspect of the
glenoid60. Our
search of the ASES meeting abstracts identified four abstracts, two of which
met our inclusion
criteria40,55;
an author of one of these
abstracts40
responded to our invitation to provide more data for the analysis. This search
identified two other
studies44,58,
which later were published and were identified with the other search
strategies. Our search of the other sources did not identify any additional
studies. One other study was published during our project, and it was
included38. This
resulted in eighteen
studies31,36-38,40-45,47,49,52-55,57,58
available for analysis (see Appendix).
Analysis of the funnel plots was limited by the relatively sparse
distribution of data points on most of the plots, which made it difficult for
us to draw any firm conclusions. The plot with the most data points (recurrent
instability) does not suggest a publication bias, as the points are relatively
evenly distributed (see Appendix).
Grouping of Studies (see Appendix)
Of the eighteen articles that were identified, four were randomized
clinical trials (two Level-I
studies40,47
and two Level-II
studies)41,54.
This was the "best" group of studies. Ten studies, all
non-randomized comparative trials, were included in the "good"
group (the controlled clinical trials). In these studies, the selected
approach was based on patient
preference36,43,49,55,
surgeon
preference58, or a
retrospective analysis of a surgeon's
experience38,44,52,53,57.
The "fair" group included four studies in which the intervention
was based on the pathological findings seen at the time of
surgery30,31,42
or in which an open approach was used because an arthroscopic approach had
failed37,45.
We identified six studies in which suture anchors had been used in the
arthroscopic
procedures40,41,44,54,57,58,
four studies in which bioabsorbable tacks had been used in the arthroscopic
procedures30,43,45,53,
and five studies in which transglenoid sutures had been used in the
arthroscopic
procedures42,47,49,52,53.
Bone tunnels had been utilized in the open procedures only in the studies by
Green and
Christensen52 and
Geiger et al.49.
Field et al.58 and
Sisto and Cook45
did not specify the fixation used in the open procedures. The remaining
authors utilized suture anchors in the open repairs.
Analysis of Stability, Rowe Score, and Return to Work and/or
Sports
When all studies were included in the analysis, it appeared that open
approaches were more reliable in restoring stability to the shoulder (see
Appendix). The pooled estimate from all studies demonstrated that arthroscopic
repairs were associated with a significantly higher risk of recurrent
instability (p < 0.00001, relative risk = 2.37, 95% confidence interval =
1.66 to 3.38), recurrent dislocation alone (p < 0.0001, relative risk =
2.74, 95% confidence interval = 1.75 to 4.28), and a reoperation (p = 0.002,
relative risk = 2.32, 95% confidence interval = 1.35 to 3.99). The rates of
recurrent instability were 18% and 8% after arthroscopic and open approaches,
respectively, whereas the rates of recurrent dislocation alone were 12% and
5%, respectively. Number-needed-to-treat analysis demonstrated that nine
arthroscopic procedures would lead to one additional case of recurrent
instability (95% confidence interval = 7 to 14).
Open approaches were more successful in enabling patients to return to
their previous work and/or sport (p = 0.03, relative risk = 0.87, 95%
confidence interval = 0.77 to 0.99).
In contrast to restoration of stability, the pooled data demonstrated no
difference in Rowe scores between open and arthroscopic approaches
(Fig. 2). In the Rowe scoring
system, a maximum of 50 points is assigned to stability; 20 points, to motion;
and 30 points, to function. Because half of the Rowe score is determined by
stability, we were interested in the relationship between the Rowe score and
recurrent instability (Fig. 3).
After arthroscopic repairs, higher rates of recurrent instability were
associated with lower Rowe scores, as might have been predicted.
Interestingly, this relationship was not observed after open repairs,
suggesting that the variance in Rowe scores for patients treated with open
repair was related to factors other than recurrent instability, such as motion
and function.
Complications (see Appendix)
No subscapularis failures were identified in association with either open
or arthroscopic techniques. Stiffness, unexplained pain, and loose hardware
were seen in both groups. There were dysesthesias in both groups, with no
mention of anesthesia technique (interscalene block and/or general). We did
not find sufficient data for a meaningful comparison of complication rates
between the two approaches.
Influence of Study Design on Effect Size
Study quality had important influences on the results
(Fig. 4). No differences in any
of the stability end points were found between the treatment groups in the
randomized controlled trials or in the finding-dictated studies (see
Appendix). In the controlled clinical trials, the groups were seen to differ
significantly with regard to recurrent instability (p < 0.00001, relative
risk = 3.02, 95% confidence interval = 1.88 to 4.86), recurrent dislocation
alone (p < 0.0001, relative risk = 3.28, 95% confidence interval = 1.86 to
5.77), and the need for a reoperation (p = 0.002, relative risk = 2.86, 95%
confidence interval = 1.49 to 5.47).
Study quality was also found to have an effect on the detection of
differences in Rowe score (Fig.
2). While Rowe scores were found to be no different between the
open and arthroscopic treatment groups when we used a fixed-effects model
after pooling all of the studies, analysis of only the randomized clinical
trials showed better Rowe scores in the arthroscopic group (p = 0.002,
standardized mean difference = 0.43, 95% confidence interval = 0.16 to 0.70).
Heterogeneity was found to be moderate in this group of studies (I2
= 66.8%). Thus, differences in this group of studies could be due to
differences in study design, patient demographics, surgical technique, or
other systematic features. No differences in Rowe scores were seen in any of
the subgroups or with the pooled data when a random-effects model was used,
indicating that the fixed-effects results are suggestive and that additional
studies are needed. Despite higher Rowe scores in the arthroscopic group in
the randomized clinical trials, no difference was seen in terms of return to
work and/or sports with the numbers available.
Influence of Arthroscopic Technique on Effect Size
Subgroup analysis of specific arthroscopic techniques again showed that
open techniques more reliably provided stability
(Fig. 5). When the analysis was
confined to arthroscopic suture anchor techniques, significantly more
recurrent instability (p = 0.01, relative risk = 2.25, 95% confidence interval
= 1.21 to 4.17) and recurrent dislocation alone (p = 0.004, relative risk =
2.57, 95% confidence interval = 1.35 to 4.92) were found in the arthroscopic
group (see Appendix). When the analysis was confined to arthroscopic
transglenoid suture techniques, stability was again seen to be more reliably
provided by open techniques, which were associated with lower rates of
recurrent instability (p = 0.0006, relative risk = 3.98, 95% confidence
interval = 1.81 to 8.73), recurrent dislocation alone (p = 0.01, relative risk
= 4.2, 95% confidence interval = 1.34 to 13.16), recurrent subluxation alone
(p = 0.03, relative risk = 2.95, 95% confidence interval = 1.12 to 7.73), and
a reoperation (p = 0.007, relative risk = 9.81, 95% confidence interval = 1.86
to 51.58). Heterogeneity was low for the end point of recurrent instability
(I2 = 36.4%) and moderate for subluxation (I2 = 60.2%)
in the transglenoid suture group, and this perhaps contributed to differences
seen in this subgroup for these end points. Bioabsorbable tacks seemed to be
the most reliable arthroscopic technique for restoring stability, with no
differences from open techniques with regard to any of the stability end
points. The number-needed-to-treat analysis suggested that seventeen
arthroscopic procedures with a suture anchor technique would lead to one
additional case of recurrent instability (95% confidence interval = 9 to
100).
The subgroup analysis was confounded by several differences between the
fixed-effects and random-effects models. The only significant differences seen
with random-effects modeling were higher rates of recurrent dislocation and
reoperations in the transglenoid suture group compared with those in the
open-treatment group. While these findings were in agreement with the results
with the fixed-effects model, none of the other significant differences that
were seen in the fixed-effects models were shown by the random-effects models.
This indicates that the differences shown by the fixed-effects model are
suggestive but additional studies are needed.
In contrast to the results for stability, Rowe scores after arthroscopic
procedures involving suture anchors or bioabsorbable tacks were better than
those following open techniques (p = 0.04, standardized mean difference =
0.29, 95% confidence interval = 0.01 to 0.56 for the suture anchor group and p
= 0.007, standardized mean difference = 0.41, 95% confidence interval = 0.11
to 0.71 for the bioabsorbable tack group)
(Fig. 6). There was high
heterogeneity in the bioabsorbable tack group (I2 = 79.5%). In
addition, when a random-effects model was used, no difference was found
between the bioabsorbable-tack group and the open-treatment group. Because no
reports on arthroscopic suture anchors included data regarding return to work
or sports, the analysis of the effect of the technique was not performed for
this end point.
Although recurrent instability is one of the most common shoulder
problems being treated with arthroscopic or open surgical approaches, there
have been few rigorous trials comparing these methods. The goal of a
meticulous systematic review of evidence is to combine the results of all
available comparative studies; however, it is critical to recognize that these
analyses are intrinsically limited by several factors: (1) each surgeon
applies each technique somewhat differently, (2) the needs of individual
patients may prompt modifications of the methods within a given surgeon's
practice, and (3) surgeons are unlikely to be equally competent with two
different techniques. The goal of the present study was to apply the best
current methodology to determine whether there is clear evidence in the
literature supporting the superiority of either open or arthroscopic
approaches.
Our analysis indicates that open approaches are more reliable for restoring
stability. Pooled data demonstrated significantly lower risks of recurrent
instability, dislocation alone, and a reoperation after open procedures. In
addition, arthroscopic techniques involving use of suture
anchors40,41,44,55,57,58
were shown to be significantly inferior to open techniques with respect to the
resulting rate of recurrent instability.
In contrast, we found evidence that arthroscopic approaches resulted in
better Rowe scores. This was the case for both arthroscopic procedures done
with suture anchors and those done with bioabsorbable tacks. As half of the
Rowe score is determined by stability, differences were likely due to higher
scores for function and motion (which account for the other half of the score)
after arthroscopic repair. Although arthroscopic approaches resulted in better
Rowe scores, they were not as good as open approaches in enabling patients to
return to work or sports.
As an additional part of our analysis, we tried to determine if study
quality influenced the results. Given the relative lack of randomized
controlled trials in the orthopaedic
literature61,62,
this is a critically important question. We found that the results differed
among the randomized clinical trials, the controlled clinical trials, and the
finding-dictated studies. The controlled clinical trials (and the pooled
effect estimate) showed significant differences between the open and
arthroscopic approaches with regard to the end points of recurrent
instability, dislocation alone, and the need for a reoperation, whereas these
differences were not found in the randomized clinical trials or the
finding-dictated trials. This observation indicates that the pooled-effect
estimates were influenced heavily by the controlled clinical trials, where
heterogeneity may have been a factor. With respect to Rowe scores, significant
differences between the open and arthroscopic approaches were found in the
randomized clinical trials but not in the controlled clinical trials, the
finding-dictated studies, or the pooled estimate. Although no differences with
regard to return to work or sports after the two approaches were seen in any
subgroup, the pooled estimate demonstrated a difference. All of these results
indicate that the quality of the study strongly influenced the results,
suggesting the need for care in interpreting results from systematic reviews
of studies of different or uncertain quality.
We also sought to determine if the arthroscopic technique influenced the
results. Proponents of the arthroscopic approach claim that newer techniques
involving use of suture anchors yield outcomes approaching those with open
techniques21,29-32.
However, our subgroup analysis did not support this contention. Our analysis
of the six trials in which this technique had been used revealed that
significantly more recurrent instability and recurrent dislocation alone were
seen in the arthroscopic groups. In contrast, the Rowe scores seen following
use of this arthroscopic technique were better than those observed after open
techniques. Although the data did not allow a direct comparison between
arthroscopic techniques, bioabsorbable tacks seemed to perform better than the
other two arthroscopic techniques, as there were no differences in the
stability end points between the bioabsorbable-tack and open techniques. The
transglenoid sutures performed poorly compared with open techniques. Poor
results have been described in many previous studies of the transglenoid
suture
technique18,63-66.
Two previous meta-analyses also demonstrated that instability is more
likely to recur following arthroscopic
repairs23,24.
In a systematic review of six studies, Freedman et al. found that the odds of
recurrent dislocation were 2.3 times greater after an arthroscopic
technique23, and
Mohtadi et al. found that the odds were 2.0 times greater in a review of
eleven studies24.
Unlike us, Freedman et al. found better Rowe scores after the open approach,
although none of the series that they included were treated with an
arthroscopic suture anchor technique. Mohtadi et al. found odds of 2.9 in
favor of open techniques with regard to the end point of the patient returning
to work, an observation that is similar to ours. Several high-quality studies,
some of which have included an arthroscopic suture anchor technique, have
become available since these prior meta-analyses were performed. We included
eighteen studies in our systematic review, including four randomized clinical
trials (involving a total of 218 patients) and six reports presenting the
results of an arthroscopic suture anchor technique (483 patients). This
allowed us to examine the influence of both study design and arthroscopic
technique. We found that heterogeneity possibly affected some of our
conclusions, as discussed. It did not seem that publication bias was present.
Utilization of fixed-effects and random-effects models was enlightening, as
there was disagreement between the models with regard to several end points
and subgroups in the analysis of arthroscopic technique, indicating
heterogeneity in the studies.
The results of this study must be considered in light of certain
limitations. First, many surgeons will not attempt an arthroscopic repair in
patients who have a large osseous defect (=25% of the glenoid or =21% of
the glenoid
length30,67,68),
in athletes who play contact
sports3,69,
or in patients with multiple
recurrences69.
Inclusion of such patients in previous series may have contributed to the
inferior results seen with the arthroscopic approach. We did not perform a
subgroup analysis of these factors, as the data were not presented in a way
that allowed this to be done, so unfortunately we were unable to clarify this
issue. Second, analysis of shoulder function was limited by the data
available; ideally such data would include not only a scoring system (such as
the Rowe score) and return-to-activity data but also analyzable data on range
of motion and subscapularis function. We attempted to evaluate all of these
factors, but we were limited by sparse data. Statistical analysis would be
facilitated by consistent reporting of mean values with standard deviations
for range of motion in degrees of forward elevation, external rotation at the
side, and external rotation in abduction. Little information was available on
the function and integrity of the subscapularis after surgery, one of the
primary concerns with open surgical approaches. Use of the Rowe score to
evaluate shoulder function is another limitation as that score is compromised
by the weight placed on stability. We found that higher recurrence rates after
arthroscopic techniques seemed to be related to lower Rowe scores. Low Rowe
scores following open repairs were not associated with recurrence, suggesting
that points were lost as a result of diminished range of motion or function
rather than because of instability. We included the Rowe score in this
analysis because it was commonly included in the literature that we analyzed.
The results discussed here demonstrate the shortcomings of any scale that
attempts to combine points for attributes as disparate as stability and range
of motion in the same score. Such an attempt requires arbitrary weighting of
one attribute in relation to another. It is obviously preferable to use an
evaluation system that enables stability, range of motion, function, and
comfort to be assessed separately without making an a priori
assignment of relative value.
In conclusion, the available evidence indicates that recurrence rates are
higher after use of arthroscopic techniques, even those involving suture
anchors. While return to work and/or sports was better after open repairs,
Rowe scores were better following arthroscopic repairs.
Tables showing the demographics and the characteristics (including reported
complications) of the included trials, results of selected subgroup analyses,
and a funnel plot for recurrent instability 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). ?