This systematic review and meta-analysis was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis - Diagnostic Test Accuracy (PRISMA-DTA) guidelines [7]. The approaches to evidence searching and data processing described in this article were based on the Cochrane Collaboration’s diagnostic test accuracy method [8]. No ethical approval or informed consent was required for this article because all data were obtained from published literatures. Searching for studies, identifying eligibility, extracting data, and assessing quality were performed by two investigators independently. Any disagreement was resolved through discussion until they reach a consensus.
Search strategy
Four electronic databases, PubMed, EMBASE, Scopus and Cochrane Library, were searched for entries recorded from the time of database inception to July 31, 2021. Vocabulary and syntax were specifically adapted according to the database. The initial search included variations of title/abstract/keywords and medical subject heading terms including “dual energy” AND (“computed tomography” OR “CT”) AND “bone” AND “edema” and were modified as needed by individual database. No language restrictions were applied [9]. To find any additional possible records, a gray literature search was performed by one author with evaluation of recent annual meeting for the American Roentgen Ray Society (ARRS), as well as recent years of conference proceedings from the Radiological Society of North America (RSNA) and the European Congress of Radiology (ECR). Conference abstracts that met the inclusion criteria and were not yet available in full text were included in analysis.
Literatures from each database and other sources were combined into a list from which duplicates were removed and were then screened initially with title and abstract for relevance. Subsequently, full-text review of the potentially relevant studies was conducted.
Inclusion criteria
The following criteria needed to be met for the studies included in this systematic review: (1) BME was evaluated around knee joints, (2) BME was the target finding and DECT was the index test, (3) MRI was used as the reference standard, (4) at least 10 patients were included, (5) sufficient data could be extracted to construct a 2 × 2 contingency table. Non-original researches (including case reports, commentaries, consensus statements, guidelines, and narrative reviews) were excluded. If more than one study provided overlapping data, only the most comprehensive or latest study was included.
Data extraction
Requisite data extracted and recorded in the standardized Excel files (version 16.54, Microsoft). Study characteristics including surname of the first author, year and country of publication, prospective versus retrospective study design, time between DECT and MRI, number of readers, presence of consensus reading, and reader experience were recorded. Patient characteristics including total number and inclusion interval of patients, mean age (Range), number of male and female, and number of knees (regions) with BME after acute knee injury were recorded. Additionally, specific site of pathology, true positive (TP), false positive (FP), false negative (FN), true negative (TN), and threshold value was recorded for each assessment.
Risk of bias assessment
The methodological quality of the included studies was appraised according to the Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 tool, which based on four separate domains: (1) patient selection, (2) index test(s), (3) reference standard, and (4) flow and timing [10]. Risk of bias (ROB) was assessed in each domain, and concerns about applicability were assessed in the first three domains with signaling questions. These questions were answered with “yes” for a low risk of bias/concern, “no” for a high risk of bias/concern, or “unclear” when relevant information was not clearly provided. Studies considered high risk for any single signaling question within a domain were considered to have a high ROB for that respective domain.
Statistical analyses
Generally, a 2 × 2 contingency table was developed for each included study. If both qualitative and quantitative analysis were reported in a study, two separate contingency tables were created. Pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were calculated using the bivariate meta-analysis framework (bivariate mixed effects regression model). In addition, summarized receiver operating characteristic (sROC) curves were constructed, with the area under the curve (AUC) depicting the accuracy of the tests.
To examine potential sources of variability, extensive subgroup analysis was performed on the studies. And univariate analysis was performed when bivariate regression model did not converge due to limited number of studies (< 4) and zero values in the 2 × 2 contingency table. A two-sided p value < 0.05 was considered statistically significant in all statistical tests. Simultaneously, statistical evaluation for publication bias was performed. Stata version 16 (StataCorp, College Station, TX) was used to analyze data from the included studies, and Review Manager Software version 5.4 (Cochrane Collaboration, Oxford, UK) was used to assess the methodological quality of the included studies.