Background: Detection of coronary artery disease (CAD) is important due to its high prevalence and its medical and economic implications. Purpose: A systematic review of the diagnostic performance of stress echocardiography (Echo), SPECT, cardiac magnetic resonance (CMR), CT Perfusion (CTP) and PET versus invasive coronary angiography (ICA) or fractional flow reserve (FFR) using hierarchical summary ROC (HSROC) methods. Data Sources: MEDLINE, EMBASE and SCOPUS for literature published in English or Spanish from January 1970 to December 2015. Study Selection: For inclusion, studies had to meet the Cochrane guidelines, had to evaluate the sensitivity and specificity methods, and use ICA and/or FFR. Only those studies with STARD methodology ≥60% were included. Data Extraction: Ten investigators extracted patient and study characteristics and 4 resolved any disagreements.

Review question: What non-invasive cardiovascular imaging tests (SPECT, Stress Echo [SE], Cardiovascular Magnetic Resonance [CMR], CT Perfusion [CTP] and PET myocardial perfusion) have the best diagnostic accuracy to detect obstructive coronary artery disease using four different cutoffs, two anatomic (invasive coronary angiography) and two functional (invasive fractional flow reserve)? Which non-invasive cardiovascular imaging tests (SPECT, Stress Echo, CMR, CTP and PET myocardial perfusion) have the least bias and heterogeneity in their published data?

Searches: Search in MEDLINE, EMBASE and SCOPUS databases for the literature published in English or Spanish from January 1970 to December 2015 of all prospective and retrospective studies performed with Stress Echo, SPECT, PET, CMR and CTP in patients with suspicion or known CAD compared against the reference standard of ICA (two cutoffs: lesions >50% and >70%) and/or invasive FFR (two cutoffs: <0.80 and <0.75).

Condition being studied: Obstructive coronary artery disease. Non-invasive imaging modalities (SPECT, Stress Echo, CMR, CTP and PET) to evaluate myocardial perfusion as indicative of obstructive coronary artery disease.

Intervention: There was no intervention or exposure, since it was an evaluation of diagnostic accuracy of worldwide current non-invasive cardiovascular imaging modalities compared to the accepted standard of reference.

Comparator: No control group used.

Context: Studies in English and Spanish were included to avoid missing any relevant, high quality contributions in different languages. Other languages different from English and Spanish were not included due to language barriers and limitations regarding understanding the full text document and to obtaining the necessary data, since those papers generally only have the abstract in English. There were no restrictions in country of origin, type of technique or equipment used, since we made performed a sub-analysis for those minor variables not reported in all papers.

Data extraction: After study planning, we decided to decided to included all published studies and decided not to include data that hadn't been formally published. Our search strategy with three terms related to the research question considering: patient population, intervention types of the different imaging modalities and the design type; we decided to include two types of search terms, free words and standardized words such as those use in Medical Subject Headings (MeSH) for PubMed and for EMBASE we used EMTREE terms; we also used a combination of thematic terms selected by a controlled vocabulary, the Thesaurus, open when necessary and with broad range of free text terms. As we described in detail as follows, the data extraction was always with more than two researchers. The team was divided into a searcher, a search advisor, and five junior and four senior reviewers. Four senior reviewers gathered together to define the MeSH terms necessary for the literature search. The searcher and the search advisor (one senior reviewer) worked together to build up the corresponding search algorithms as follows: for prospective or retrospective clinical trials, case series, abstracts and gray literature in English or Spanish for PubMed, for EMBASE, and for Scopus; also for meta-analysis and systemic reviews for PubMed, for EMBASE and for Scopus for the period comprised between January 1970 to December 2015. One senior author (first author) was designed as "administrator", then all retrievals from the searcher went directly and only to the administrator to avoid duplications or gaps in the review process. First the administrator received a search with titles and abstracts, which s/he distributed equally to the four senior reviewers (including the administrator) to make an initial selection and to eliminate all titles and abstracts that did not meet the inclusion or exclusion criteria, which was then gathered together by the administrator and sent back to the searcher, whose used the list of selected documents (titles and abstracts) to send a second package of searches to the administrator with all documents in full text version, who again equally distributed these documents to all four senior reviewers, who then validated the inclusion and exclusion criteria, which may have been missed in the abstract alone; the resulting package was again returned to the administrator who equally distributed the documents to all junior and senior reviewers to score each study according to the STARD checklist for this type of project. We previously established a minimum of 60% of points in the STARD checklist for a study to be approved and be included in the next stage of the review. All scoring sheets were sent back to the administrator within the timeline period. All disagreements were solved during our research meetings by a consensus of the senior reviewers. The administrator then selected the studies to be included in the meta-analysis based on the STARD score and distributed them equally to all junior and senior reviewers (including the administrator) to extract data in a pre- designed data capture sheet. At the deadline, the administrator gathered together all data capture sheets and then re-distributed equally with the original papers to the four senior reviewers who did a second review of each study to assure quality of the extracted data. All disagreements were solved during extra research meetings of the senior reviewers by consensus. After the data were accepted by the four senior reviewers, the administrator provided another senior reviewer, acting as a "statistician" with all captured data to start the statistical analysis. At the same time, all studies included in the meta-analysis were again distributed by the administrator to two senior reviewers (the administrator and the statistician) who did the data extraction for the sub-analysis, each senior reviewer took two imaging modalities for review, and after completion those sub-analysis data were gather together by the statistician to be included in the meta-analysis.

Risk of bias: In the selection process bias risk was managed by using a double check of the selection criteria by senior reviewers, by scoring each study by STARD methodology using its checklist and a rigid criteria of >60% to be considered for further analysis, any disagreements between the junior and/ senior review authors over the risk of bias in particular studies were solved by discussion during an extra research meeting of two senior authors, with involvement of a third reviewer when necessary. The publication bias was evaluated in graphic manner using Funnel plots and in mathematical form with Begg and Egger tests. A value of p<0.05 was defined as significant.

Strategy for data synthesis: We performed Forrest plots for each test and HSROC curves with the Moses-Littenberg method. We calculated AUC and Q* for each test. Total values of sensitivity, specificity, DOR, LR+ and LR-, as well as 95% CI were calculated using the bivariate model of Teitsma and the models of HSROC (hierarchical summary ROC) of Rutter and Gatsonis. The differences between the tests were evaluated using meta-regression comparing different models with the likelihood ratio test. The analysis of heterogeneity between studies was evaluated in graphic form in the HSROC curves, and in mathematical manner with meta-regression and the calculation of Higgins' I2; this last test, Higgins' I2, was not reported within the results, since in diagnostic test accuracy studies, this I2 statistic alone may not be informative as they do not consider threshold effect; in comparison to this same statistical test when it is used for the analysis of heterogeneity between therapeutic and or interventional studies, since a threshold effect can be considered. In our study the publication bias was evaluated in graphic manner using Funnel plots and in mathematical form with Begg and Egger tests. A value of p<0.05 was defined as significant.

Analysis of subgroups: To analyze statistical differences in diagnostic accuracy between modalities, we based the analyses on mathematical combinations and used the term "certainty" to refer to diagnostic accuracy that one imaging modality has over the other one; then we performed the meta-regression analysis for comparison of "certainty" between two imaging modalities at the same time. We used a graphic representation for the results. We did a post-hoc analysis by sub-groups to all other variables, common to every technique and/or specific for each imaging modality studied.

Dissemination plans: A paper will be submitted to a leading journal in this field. The findings of the review produced solid evidence that can change current worldwide practice for the benefit of patients, institutions and public health services by obtaining higher accuracy in the diagnosis of significant coronary artery disease in terms of myocardial ischemia, thus avoiding extra unnecessary tests, less harm to the patients and less unnecessary expenses.

Current review status: Completed but not published.

Inclusion Criteria:

• All studies (prospective, retrospective and even case series) that included patients of any age and gender with known or suspected coronary artery disease and evaluated the sensitivity and specificity of SE, SPECT, CMR, CTP and PET compared with ICA and/or FFR, and to that met the Cochrane guidelines including a score of >60% with STARD methodology.

Exclusion Criteria:

• Studies that included patients with known previous myocardial infarction, previous PCI with or without stent implantation, previous cardiac bypass surgery, heart transplantation, and the absence of invasive coronary angiography as a gold standard.

We searched all literature in English and Spanish from January 1970 to December 2015 that meet the inclusion criteria.

We are intended to publish the corresponding paper in a high-impact journal, with all information available to other researchers in the corresponding appendix.