Page 7 - Cardiac testing for coronary artery disease in potential kidney transplant recipients

Reference standards These data were then collated in a spreadsheet. A third author was
Coronary angiography. available to adjudicate on disagreements.

Assessment of methodological quality
Search methods for identification of studies
Methodological quality of included primary studies was assessed
by two authors using a modified QUADAS tool (Smidt 2008;
Electronic searches Whiting 2003) that included 11 of the 14 mandatory items (rep-
resentative spectrum, acceptable reference standard, acceptable
We searched the following resources.
delay between tests, partial verification avoided, differential ver-
• MEDLINE (OvidSP) 1950 - 1 November 2010
ification avoided, incorporation avoided, reference standard re-
• EMBASE (OvidSP) 1980 - November 2010, Week 44
sults blinded, index test results blinded, relevant clinical infor-
A Trials Search Co-ordinator of the Cochrane Renal Group (RM) mation, uninterpretable results explained, withdrawals explained)
formulated specific search strategies for the MEDLINE and EM- (Smidt 2008; Whiting 2003). The operational definitions of the
BASE searches (Appendix 1). QUADAS items are presented in Appendix 2.
Citation tracking was performed using Web of Science. No re-
strictions were imposed in terms of language of publication or
publication status. To maximise the sensitivity of our search, we Statistical analysis and data synthesis
avoided the use of methodology filters when searching for diag- Extracted data were used to create forest plots of sensitivity and
nostic accuracy studies because even the most sensitive filters have specificity, to depict study-specific estimates of sensitivity and
been found to miss relevant studies (de Vet 2008; Doust 2005). specificityinreceiveroperatingcharacteristic(ROC)space foreach
index test, and to investigate:
1. the diagnostic performance of each index test
Searching other resources
2. heterogeneity in the diagnostic performance of each index
We handsearched the reference lists of all primary studies and
test according to patient characteristics, study design, and study
reviews identified by the initial search.
quality factors (identified in Table 2 where sufficient data were
available)
3. the relative diagnostic performance of alternate tests based
Data collection and analysis on all available studies that provided data for at least one test,
and when the analysis was restricted to studies that provided data
for both tests.
Selection of studies Hierarchical summary receiver operating curve (HSROC) models
Two authors independently reviewed the search results, first by were fitted using the PROC NLMIXED procedure in SAS9.2®.
title and abstract, and where necessary by review of full text of the We applied the HSROC model to derive inferences about diag-
study report, to determine inclusion or exclusion. Resulting sets nostic test accuracy and heterogeneity in test performance where
of citations for inclusion were also compared. A third author was sufficient studies (n ≥ 5) for tests were available. The HSROC
available to arbitrate final decisions to include or exclude. model used study specific estimates of sensitivity and specificity
to estimate the position and shape of the summary curve (Rutter
2001). The curve was defined by three parameters: threshold (the
Data extraction and management underlyingtestpositivityrate: aproxyforthe cut-pointthatdefines
A standardised data extraction form was used to abstract study a positive test); accuracy (the diagnostic log odds ratio); and shape
design features and results data from each publication. For each (the dependence of accuracy on threshold). Each study provided
study data were extracted independently by two authors. We ex- an estimate for threshold and accuracy which were assumed to be
tracted: year of publication, country of study, study design, clin- random effects in the model. When there was no evidence of an
ical setting, definition of CAD (stenosis percentage on coronary association between accuracy and threshold, the summary curve
angiogram), the Quality Assessment of Diagnostic Accuracy Stud- was considered symmetric and its position defined by a constant
ies (QUADAS) methodological items (Reitsma 2009), prevalence diagnostic odds ratio (DOR). The model estimates were used to
of cardiovascular risk factors in the study population (percentages obtain summary estimates for sensitivity, specificity, positive and
of participants on haemodialysis; with ESKD, diabetes mellitus negative likelihood ratios, DORs and 95% confidence intervals
(DM), hypertension; who were male; with history of smoking; (CI), and the corresponding 95% confidence region for each index
and symptomatic of heart disease). We also recorded the numbers test. The corresponding area under the curve (AUC) was com-
of true positives, true negatives, false positives and false negatives. puted from the constant DOR as part of the analysis.

Cardiac testing for coronary artery disease in potential kidney transplant recipients (Review) 5
Copyright © 2011 The Cochrane Collaboration. Published by John Wiley & Sons, Ltd.