Non-surgical interventions for primary frozen shoulder: systematic review
June 2017 (CML and an experienced healthcare librarian). The
original review (21) searched 19 sources, but concluded that 2
databases (Cochrane CENTRAL and either the Science Citation
Index or MEDLINE), plus reference checking, proved effective in
identifying included papers and this approach was utilized in this
review (23). The original MEDLINE search (Table SI 1 ) was re-run
in OVID (Epub Ahead of Print, In Process & Other Non-Indexed
Citations, Ovid MEDLINE (R) Daily and Ovid MEDLINE (R)
1946 to Present) with search date limits 1 January 2010 to 8
December 2016. In addition, Cochrane CENTRAL, SCI MED-
LINE, CENTRAL databases were searched, following the search
strategies detailed by Maund et al. (21), to retrieve randomized
clinical trials published between 1 January 2010 and 6 June 2017.
Reference lists of included trials were manually searched for
other relevant trials (no new hits). Clinical trials registers were
searched to identify any additional publications and to identify
on-going trials to further inform the review. ClinicalTrials.gov
(searched 8 June 2017, n = 42: no new hits), ISRCTN (searched
7 June 2017, n = 20: no new hits) and the European Union Clini-
cal Trials Register (searched 8 June 2017, n = 3: no new hits).
Although resources for translation were not available, searches
were not restricted to English language, so that the quantity of
non-English-language research could be established. Data synthesis
Study selection Study selection
Two reviewers independently screened the retrieved title and
abstracts against the inclusion criteria (EB and NDB in 2014,
CML and JL in 2016/17). When potentially relevant studies
were identified, or information was insufficient, the full-length
article was screened and discussed by 2 reviewers. Two revie-
wers independently screened the full articles for inclusion (EB
and NDB in 2014, CML and JL in 2016/17) and then discus-
sed them together. In the case of disagreement between the 2
reviewers, a third reviewer was consulted (JL in 2014 and NDB
in 2016). This occurred in the case of 2 trials in 2014, but was
unnecessary in 2016/17. In the OVID MEDLINE search for trials 2,724 records
were identified, 76 records were screened, and 55 full
papers were subsequently assessed against the review’s
inclusion criteria. Of these, a total of 30 trials were
included in the review. No additional studies were
identified from searches of the additional databases. A
summary is provided in the PRISMA flow diagram in
Fig. 1. Table SII 1 contains a list of excluded studies. The
characteristics of the trials included in the review and
their risk of bias assessments are summarized in Tables
I and II. Four studies were included in data syntheses.
Risk of bias assessment
Risk of bias was assessed by 2 independent reviewers using
a domain-based, risk of bias assessment approach (24). Do-
mains included sequence generation, allocation concealment,
intervention integrity, effective blinding and whether outcomes
were pre-specified, analysed, and reported appropriately. Any
additional aspects of study methodology relevant to validity or
generalizability were also evaluated (e.g. appropriateness of
study measures and sample size). Items relevant to each trial’s
internal validity were reviewed and graded as adequate (low
risk of bias), partial (moderate risk of bias), inadequate (high
risk of bias), or unclear (uncertain risk of bias). Blinding of
participants and intervention providers was not thought feasible,
and thus lack of blinding here was rated as low risk. Any items
judged differently by reviewers were discussed and resolved
by 2 reviewers (CML, JL) until consensus was reached and an
evaluation of overall risk of bias was achieved.
Data extraction
541
Study characteristics for all trials included in the review were
extracted by one reviewer using a standardized form (CML) and
checked against their source article by a second reviewer (JL).
http://www.medicaljournals.se/jrm/content/?doi=10.2340/16501977-2578
1
Data syntheses were undertaken for trials evaluated as low risk
of bias. There are different methods and approaches towards un-
dertaking systematic reviews, and a lack of consensus regarding
the optimal approach researchers should utilize in reviews. Whilst
some researchers decide to include all studies, including those at
high risk of bias, in syntheses, others consider it to be inappropriate
and potentially misleading to include studies at high risk of bias
in all parts of data analyses/syntheses within a review (25, 26).
It was unclear from the written publication whether participants
with diabetes mellitus were included or excluded in trials for 10
trials evaluated at low risk of bias. The corresponding authors for
these trials were emailed requesting additional information, with a
follow-up email at least 2 weeks later, where necessary. Following
this, it was confirmed that 4 trials were eligible for inclusion and
6 were excluded because they included participants with diabetes
mellitus or because they did not reply to confirm eligibility. A
further author (Ma et al.) (27) was emailed to clarify how parti-
cipants were randomized; no reply was received, therefore this
trial could not be evaluated as low risk of bias.
RESULTS
Comparison of joint mobilization and stretching vs
stretching alone.
Celik & Mutlu (28) compared the effectiveness of
joint mobilization and stretching vs stretching alone.
Joint mobilization techniques included glenohumeral
joint distractions, caudal glides, posterior and anterior
glides at a rate of 2–3 oscillations/s for 1–2 min and
progressing from Grades I–II to III–IV if pain allowed.
The stretching programme involved 20 s of stretching
and 10 s of rest and was performed 10 times in the
directions of flexion, abduction in the scapular plane,
external and internal rotation. Both groups performed
a home exercise programme; self-stretching and exer-
cises (scapular retraction, external rotation, extension
against resistance, wall and table push-ups, and sca-
pular adduction in prone). All participants received
18 sessions, including the home exercise programme
(Table I). This study is a pilot trial, but is reported as a
trial, including a sample size calculation and statistical
testing, rather than as a pilot trial (29). Both groups
J Rehabil Med 51, 2019