Journal of Rehabilitation Medicine 51-7 | Page 12

OSTs (for each) Table II. Possible combinations of index test/reference standard/target condition for meta-analyses Examination of the LHBT in the clinical setting Index test Reference standard LHBT pathology identified HRUS Surgery (open or arthroscopy) One of: Tendinopathy Dislocation Rupture – partial Rupture – total Effusion (bicipital recess) One pathology which each OST is designed to detect (see Appendix I): SLAP lesion Tendinopathy Proximal LHBT pathology other than SLAP (dislocation, rupture, tendinopathy) Another diagnostic imaging modality Any reference standard Surgery (open or arthroscopy) HRUS MR imaging/arthrography Any reference standard 481 HRUS: high-resolution ultrasound; OSTs: orthopaedic special tests; MR: magnetic resonance; SLAP: superior labrum anterior and posterior; LHBT: long head of the biceps tendon. participants they recruit and the test that they evaluate (19). In that respect, data were combined where studies measured the accuracy of the same index test for the diagnosis of the same LHBT pathology: (i) according to the same reference standard; and (ii) according to all reference standards. Meta-analysis tools were used when a minimum of 4 primary studies were identified (Table II) (20). Where a limited number of studies prevented the use of meta-analysis tools, only sensitivity (Sn) and specificity (Sp) estimates are presented from each study, together with forest plots. Meta-analyses were conducted using the approach developed by Rutter & Gatsonis with the V3.3.3 of R statistical software (http://www.r-project.org/) (21). The HSROC package was used to calculate overall pooled estimates of the included diagnostic studies taking into account the between-study and within-study variability. This routine, based on Bayesian statistics, estimates the overall sensitivity (Sn) and specificity (Sp) for group of studies and produces a receiver operating characteristic (ROC) curve with credible interval and a 95% prediction region. The classical confidence interval (CI) presumes that differences in Sn and Sp between studies are caused only by a statistical instability related to sampling or measurement errors. All es- timates would turn around a unique value of Sn and a unique value of Sp. In reality, for the same technique, Sn and Sp may vary in time, with different populations, with different opera- tors or any other relevant conditions that change the nature of the test. Across different conditions, Sn and Sp could fluctuate among a range of values that reflect a change in reality rather than a statistical instability. The credible intervals delimit how Sn and Sp could fluctuate for reasons other than sampling or measurement errors. In this context, the CI adds to the credible interval the uncertainty caused by sampling and measurement errors. The credible intervals are narrower than the CI. The prediction region is defined by pairing the CI with the credible interval. Heterogeneity was explored graphically using forest plots. Positive (LR+) and negative (LR–) likelihood ratios were calculated from the overall Sn and Sp. However, confidence and credible intervals could not be calculated for likelihood ratios. Studies with cells containing zero in the 2 × 2 table lead to statistical model instabilities. A continuity correction, consisting of a small positive number (0.5 as suggested in the literature) was then added to the observed frequency (20). For SLAP lesions, because the degenerative fraying of the SLAP I lesion is often considered a normal variant and asymp- tomatic, type II–IV and type I–IV lesions studies were isolated (22). The type II–IV group comprised studies either designed to assess the diagnosis of SLAP II–IV lesions or where only SLAP II–IV lesions were ascertained by the reference standard. RESULTS Search results Searches resulted in 777 citations (duplicates remo- ved). Twenty-eight articles were accepted for the review after full-text screen. Fourteen articles were obtained by scrutiny of the reference lists of reviews and primary studies. Of the 42 eligible studies, 30 were included in the analysis of the review (8 for HRUS, 22 for OSTs; Fig. 1, Table III). Methodological quality of included studies For the risk of bias assessment, inter-rater agreement was excellent (Gwet’s AC1 of 0.85). The overall Records screened on basis of title/abstracts (duplicates removed n= 777 Records identified from reference lists of articles n=14 Full-text of potentially relevant studies retrieved n=101 Full-text elligible article n=28 Records excluded n=676 Excluded studies and reasons – Studies did not address LHBT (25) – Reviews (34) – Highly selected population (1) – Studies with lacking or incomplete data (4) – Not English or French (2) – No index test (1) – Index test is a cluster (1) – Reference standard inade- quate (1) – Target condition inadequate (4) Total 73 Potentially appropriate studies to be included in analysis of review n=42 Studies included in analysis of review – By source: MEDLINE (15), CIHAHL (4), EMBASE (4), References (7) – By index test: HRUS (8), OSTs (22) Excluded studies and reasons – Studies had highly selected population (2) – Studies had discrepancy in 2X2 tables or between text and tables (2) – Studies had 100% prevalan- ce (1) – Study lacking data to draw 2X2 tables (6) – Study was retrospective (1) Total 12 n=30 Fig. 1. Flow diagram of the bibliographic search. HRUS: high-resolution ultrasound; OSTs: orthopaedic special tests. J Rehabil Med 51, 2019