The systematic review was performed according to the recommendations of Eden et al. (Eden et al. 2011) on review methods, data sources, and search strategies. We addressed two review questions:
-
Using the Tubiana scale as a reference test, how sensitive and specific is the URAM scale for defining quality of life in patients with in DC?
-
How sensitive to change is the URAM scale after treatment with FSC and CCH?
We performed a systematic search of PubMed, EMBASE, Cochrane, Google Scholar, LILACs, and Web of Science for articles published between January 1, 1990 and June 1, 2019.
The search criteria used in all the databases were combinations of the terms “Unité Rhumatologique des Affections de la Main”, “URAM”, "Dupuytren Contracture", and “Dupuytren”.
Two reviewers (PVF and DGH) independently searched the databases and reviewed the articles retrieved. They also hand searched the reference lists of relevant articles and reviewed the gray literature to identify clinical trial reports and conference proceedings.
Clinical trials, cohort studies, and case-control studies that had used the URAM scale to evaluate DC were included. Authors were contacted when specific information on the use of this scale was missing. To minimize publication bias, no language constraints were placed.
Study Selection
Two researchers (PVF and DGH) independently screened the titles and abstracts to identify suitable texts, which they then reviewed in depth. When the researchers disagreed on whether a particular article should be included or excluded, the article was reviewed by a third researcher (FJCH) to break the tie.
Data Extraction and Risk of Bias Assessment
Working separately, PVF and DGH transferred all relevant data from the selected articles into standardized forms. The reliability of the entries was checked by another researcher (JEPJ). In addition to effect variables (mean [SD] pre- and post-intervention URAM and Tubiana scores), the data recorded included demographic variables (age, gender, and hand and radius affected) and variables for the stratification analyses in the meta-analysis (e.g., quality, language, study type).
As the studies included in the meta-analysis differed in type, their quality was assessed using the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) checklist (von Elm et al. 2014) applied separately by two researchers for each article. To minimize bias, a score of 15 or higher was used to identify high-quality studies. Discrepancies (i.e., differences in scores that placed a given study above or below the cutoff of 15) were resolved by a third researcher (RSC).
Statistics
Three meta-analysis models were used to answer the research questions: a hierarchical summary receiver operating characteristic (HSROC) model, a difference in means model for pre- and post-treatment URAM scores, and a meta-regression model adjusted for time since treatment.
For the HSROC model, tables summarizing Tubiana and URAM scores reported in each of the studies were created. In both cases, it was assumed that the scores were normally distributed. The data were then presented in 2 × 2 contingency tables with the URAM scale as the index test and the Tubiana scale as the reference test. The respective thresholds used were 2.5 and 1. Prevalence of DC was established at 100%. In other words, it was assumed that there were no true negatives, that it is that all the negative results for the reference test were false negatives. Enabling continuity correction, we then built a hierarchical multinomial regression HSROC model (Rutter et al. 2001), which converts the distribution of the two variables, allowing calculation of the overall ROC curve under the assumption that there is an underlying curve for each of the studies included. Each curve is determined by two parameters, α and β, which denote accuracy and asymmetry, respectively. Using these parameters and a θ parameter to denote the positivity threshold, distribution tables were generated for each study assuming that while the distribution of parameters would vary between studies, it would be normal and random (random-effects model). We then estimated the overall ROC curve together with the optimal threshold and corresponding confidence interval. The bivariate model was applied to directly model specificity and sensitivity based on the assumption that the Napierian logarithm of the odds ratio had a normal bivariate distribution in the different studies analyzed (Reitsma et al. 2005).
For the second model, standardized mean differences in pre- and post-treatment URAM scores were computed using Cohen’s D and appropriate weighting. The most conservative model was selected in each case (Higgins et al. 2011). Differences of over 10% were considered to be clinically significant and the results were stratified by type of intervention (FSC or CCH). Each group was finally assigned an overall value.
For the meta-regression model, the dependent variable was change in URAM scores after treatment (differences in means before and after FSC or CCH) and the independent variables were Tubiana scores, time since treatment, type of treatment, age, and sex. The model with the greatest explanatory power was selected.
Heterogeneity between studies was investigated using the I2 statistic, with high heterogeneity defined as a value of over 50% (Higgins et al. 2002). Potential sources of heterogeneity were investigated by subgroup analyses (study setting, language, ethnic origin), and the effect of outliers was analyzed in a sensitivity analysis in which studies were excluded one by one.
Analyses were conducted using the metan, metacum, metafunnel, and metandi features in Stata version 15. Differences in means were considered to be significant when the confidence intervals did not cross 0 and clinically significant when there was a difference of at least 10%. Publication bias was assessed using funnel plots and the Begg-Mazumdar test (Begg et al. 1994).
Funding and potential conflicts of interest
The whole investigation has been financed by own funds. Each author certifies that he has no commercial associations that might pose a conflict of interest in connection with the submitted article. All authors have completed and submitted the Conflict of Interest Disclosure Form and none were reported.