Journal of Rehabilitation Medicine 51-1CompleteIssue | Page 7

4
J. Oosterhaven et al.
Table I. Risk of bias due to various factors, and total score study quality (23)
Study Study
participation Study
attrition Prognostic factor Outcome
measurement
measurement Study
confounding Statistical analysis Study quality
and reporting
total score
Howard et al. 2009 (24)
Bendix et al. 1998 (27)
Biller et al. 2000 (28)
Carosella et al. 1994 (29)
Coughlan et al. 1995 (30)
Kvaal et al. 1999 (31)
Sloots et al. 2009 (32)
Richmond & Carmody 1999 (33) Low
Low
Low
Low
Low
Moderate
Low
Moderate Moderate
High
High
Moderate
Low
High
Low
High Low
Moderate
Moderate
Moderate
Moderate
Moderate
High
Moderate High
High
High
High
High
High
High
High Low
Low
Low
High
High
High
Moderate
Low
The reviewers familiarized themselves with the QUIPS through
a test session involving 2 excluded studies, before judging the
included studies. All ratings were entered into a spreadsheet.
Any difference between the 2 reviewers was resolved through
discussion and, if needed, a third reviewer was consulted to
reach consensus (WD). An overall score of the study quality
was based on the recommendations of Hayden and colleagues
(23). For each domain the risk of bias was classified as high,
moderate or low. Studies were considered of high quality if in
all 6 domains a low risk of bias was found and these studies
were labelled as an overall low risk of bias study (23).
Data-extraction, data-analyses and data-synthesis
Several steps were taken to extract and synthesize the data from
the included studies, all steps independently by 2 reviewers (JO,
HW), followed through a discussion, if needed a third reviewer
was consulted to reach consensus (WD). In step 1, an extraction
manual was designed to facilitate the data-collection process.
The following information was extracted from the included
studies: (i) general information: authors, journal, publication
date, country, language; (ii) research design: retrospective-or
prospective cohort study or RCT; (iii) research population;
(iv) analytical approach: univariate analyses using a variety of
methods (for example χ 2 tests, independent t tests and univariate
logistic regression analyses) and multiple logistic regression
analyses; (v) all possible factors associated with dropout in
univariate analyses and multiple logistic regression analyses
with statistical significance and strength of the associations,
number of studies that examined the associations.
In step 2 the factors were grouped into 5 domains of Meichenbaum
& Turk (14).
For each domain the presence of associations and the direction
of the associations of predictors and dropout was determined
in univariate analyses and multiple logistic regression analyses
(Tables II, III, and Table S2 1 and Table S3 1 ).
For data-synthesis in systematic reviews of studies on out-
come prediction models there is still no clear methodological
procedure for pooling the data. The heterogeneity of the study
populations, study interventions, predictors, statistical analyses
and statistical reporting and the fact that most predictors were
only investigated in one study (24), did not support applying
a best-evidence synthesis (25, 26). Therefore, in step 3, only
potential predictors from univariate analyses and multiple lo-
gistic regression analyses that were judged in at least 2 studies
were described in the results. To summarize the results for a
predictor that was investigated in more than 1 study the term:
(i) “significant” was assigned if ≥ 75% of the studies showed
significant results; (ii) “non-significant” was assigned if ≥ 75%
of the studies showed non-significant results; (iii) “conflicting
results” was assigned if the rule of ≥ 75% studies showing
significant or non-significant could not be applied, or if oppo-
site directions of the association were found in studies (e.g., if
www.medicaljournals.se/jrm
Moderate
Moderate
Low
Low
High
Low
Moderate
Moderate
Low
Low
Low
Low
Low
Low
Low
Low
dropout was associated with higher pain intensity in one study
and with lower pain intensity in another study) (26).
RESULTS
Study selection
The initial search identified a total of 1,954 studies.
One additional study was added through screening re-
ference lists (Fig. 1). Without the 555 duplicates, 1,400
studies remained for screening on title and abstract. A
total of 32 articles were considered for inclusion, but
after full-text screening, only 8 studies were selected
for the review. The main reason for exclusion was study
design, such as cohort studies with only analyses on
differences between completers and dropouts at ba-
seline without prospective or retrospective follow-up
and without univariate- or multiple logistic regression
analyses of factors that might be predictors for dropout.
Two studies were excluded due to the absence of an
interdisciplinary approach in the intervention or on
the grounds that the intervention under study was an
online programme.
Study characteristics
The 8 included studies were conducted between 1994
and 2009. Three studies took place in Europe and 5
studies in the USA. Table S1 1 provides an overview
of the studies included in this review. Most studies
focused in their main research objective on detecting
predictors of dropout in chronic pain management
programmes, 3 studies had a prospective cohort design
(24, 27, 28), 4 a retrospective cohort design (29–32)
and 1 randomized clinical trial (RCT) with a retrospec-
tive secondary analysis on dropout (33).
Interventions
Seven studies described outpatient chronic pain mana-
gement programmes with an interdisciplinary approach
(27, 29, 31–33). Three of these studies were outpatient
programmes with a focus on return to work, known as
functional restoration programmes (24, 27, 29). One
study investigated an inpatient programme with an
interdisciplinary approach in the UK (30).