This is a protocol for a systematic review and meta-analysis on the diagnostic test accuracy (DTA) of BNP and NT-proBNP for the detection of CPE in patients with acute respiratory failure. We will adhere to the standards of the methodology of the Cochrane Handbook for Systematic Reviews of DTA[17] and the Preferred Reporting Items for Systematic Reviews and Meta-analyses of Diagnostic Test Accuracy Studies [16] in reporting the findings of this review.
Objectives
Primary objective
To determine the accuracy of BNP and NT-proBNP for the diagnosis of CPE in patients with acute respiratory failure.
Secondary objective
To explore the possible sources of heterogeneity, including risk of bias and presence of past medical history, such as renal deficiency and left ventricular dysfunction, among studies assessing BNP and NT-proBNP.
Criteria for studies included in this review
Types of studies
We will include all reports on the accuracy of plasma BNP or NT-proBNP for the diagnosis of CPE among adult patients with acute respiratory failure. Moreover, the study will comprise prospective or retrospective observational (cohort or cross-sectional) studies or secondary analysis of randomized controlled trials. However, diagnostic case-control studies and case studies without sufficient diagnostic test accuracy data, namely true-positive (TP), false-positive (FP), true-negative (TN), and false-negative (FN) values, based on the reference standard will be excluded.
Participants
The target participants are as follows:
- Adult patients aged 15 years or older.
- Patients with acute respiratory failure, dyspnea, and hypoxia who were admitted in the ED or ICU.
- Patients with bilateral pulmonary edema on imaging studies, such as radiography and CT scan.
Index test
The index tests are plasma BNP and NT-proBNP assays using any types of method. We will report these index tests as positive or negative based on the study threshold cutoffs. Studies evaluating both BNP and NT-proBNP in a similar study population will also be included.
BNP and NT-proBNP are different indexes widely used to diagnose heart failure[18, 19]. Currently, an alternative test for cardiac biomarkers is not available. To distinguish ARDS from CPE in patients with acute respiratory failure and bilateral pulmonary edema, the BNP and NT-proBNP tests can be used in addition to echocardiogram, chest radiography, and physical examination.
BNP is synthesized as a prohormone (proBNP), which is then cleaved into the active fragment BNP (32-amino-acid, C-terminal fragment) and the inert fragment NT-proBNP (inactive 76 amino-acid, N-terminal fragment). They are synthesized and released into the circulation by cardiac ventricular myocytes in response to volume expansion and possible increased wall stress. Both are cleared mainly by the kidneys. However, NT-proBNP has a longer half-life (mean: 120 vs. 20 min), and it is more stable than BNP in vitro[20, 21]. The serum BNP and NT-proBNP levels may vary due to kits used in the examination or under some conditions, such as renal dysfunction, obesity, drug-related disorder, inflammation, and cancer[22-25].
Reference standard
The reference standard for the final diagnosis made by experts, such as cardiologists and emergency physicians, refers to all available patients’ information, including clinical features and response to treatment.
The difference between CPE and ARDS in ED and ICU settings commonly comprise the combined results for physical examination, echocardiogram, and invasive evaluation (e.g., PAWP). In echocardiogram, ejection fraction (EF), left ventricular end-diastolic dimension, and diameter of the inferior vena cava are often evaluated. However, the inter-observer agreement for the diagnosis can be low unless these examinations are performed by expert sonographers, including cardiologists. Moreover, in recent years, PAWP has been found to provide inaccurate clinical estimation [26], and there is no clear evidence showing its benefits[27, 28]. The trend was identified based on the Berlin Definition, in which low PAWP (<18 mmHg) is no longer required for the assessment of ARDS[1].
Target conditions
The target condition is CPE, which should have causes different from those of acute respiratory failure during the initial treatment. Acute respiratory failure with bilateral lung infiltrates on chest radiography or CT scan is common in the ED and ICU settings. In this review, CPE was defined as bilateral lung infiltrates on radiography or CT scan based on the reference standard (18). The timing of the diagnosis ranges from the early stage to the late stage of illness, such as during hospital discharge.
Clinical settings
The clinical settings will be in the ED and ICU.
Search methods used to identify studies
Electronic searches
An electronic search strategy has been developed in collaboration with librarians. To identify all prospective, retrospective, or randomized controlled trials, we will search MEDLINE (via PubMed; from 1966 to the present) and Cochrane Central Register of Controlled Trials in the Cochrane Library. We will search for ongoing and unpublished studies at www.clinicaltrials.gov and the International Clinical Trials Platform (www.who.int/ctrp/en). There are no limits regarding language and publication date for this review. We have outlined the search strategy in Appendix 1. Moreover, the reference lists of relevant articles will be hand searched.
Data collection and analysis
Selection of studies
Two or more authors will independently screen all articles identified using our search strategy based on the inclusion criteria of this review. Screening will be a two-step process (initial title/abstract screening and full-text screening). Disagreements among reviewers will be resolved via a consensus or third-party reviewer. After the full-text screening, a list of excluded studies with reasons will be provided in the appendix of the final report.
Data extraction and management
Two or more authors will develop the data extraction sheet with the following information:
- Study characteristics: author, year of publication, country where the study was conducted, design, sample size, clinical settings, and funding source.
- Population characteristics: inclusion/exclusion criteria, number of dropouts with reason, and demographic characteristics of the participants (such as age and sex).
- Index test: timing of sampling, method of examination, time to result, and name of the person who conducted the test.
- Reference standard: method of examination, time to result, and name of the person who performed the examination.
- Information regarding quality assessment items based on the Quality Assessment of Diagnostics-Accuracy Studies 2 (QUADAS-2) assessment system[15].
- Outcomes: Based on the information in the 2 × 2 table, we will assess diagnostic accuracy parameters, such as TP, FP, TN, and FN values.
Assessment of methodological quality
Two or more investigators will independently evaluate and report the risk of bias using the QUADAS-2 tool[29]. We will assess four domains for the risk of bias, which are as follows: patient selection, index test, reference test, and flow and timing. Moreover, applicability concerning the first three domains will be evaluated. For each domain, we will respond to the questions with a Yes/No/Unclear answer, and the risk of bias will be considered as Low/High/Unclear.
A statistical assessment of publication bias will not be performed. There is no evidence of publication bias in the systematic reviews of diagnostic accuracy, and the methods used in assessing publication bias are not reliable when applied to diagnostic accuracy studies.
The sources of bias in diagnostic accuracy studies include those related to the patients (spectrum bias and selection bias), index test (information bias), reference test (misclassification bias, partial verification bias, differential verification bias, incorporation bias, disease progression bias, and information bias), and data analysis (excluded data bias).
Statistical analysis and data synthesis
We will individually analyze BNP and NT-proBNP. In the included studies, the reference standard (final diagnosis made by experts) will have dichotomous outcomes, and the index tests will have thresholds at which the diagnostic accuracy parameters will be calculated. For all studies, we will establish 2 × 2 tables (multiple tables for a study with multiple thresholds) with data on TP, FP, FN, and TN values in each study. The diagnostic odd ratio will be also calculated, which is a measure of the discriminative power of a test that has been considered a good indicator of test performance[30, 31]. We will use forest plots with 95% confidence intervals (CIs) to assess the sensitivity and specificity in each study. To visually assess the correlation between both indices, the summary of the ROC curve will be plotted when the studies have different cutoffs or the results were presented as circles in the ROC space when the studies had a similar cutoff for reporting sensitivity versus 1-specificity. We expect that the included studies will use different threshold cutoffs for the assessment of sensitivity and specificity because no consensus has been established as to the optimal threshold cutoff of BNP or NT-proBNP for the diagnosis of CPE. In the meta-analysis, we will use a hierarchical summary receiver operating characteristic model to pool data and to estimate and summarize the receiver operating characteristic curve when the studies use different cutoffs. A bivariate model will also be used when the studies use similar cutoffs.
All analyses will be performed using the STATA, SAS (SAS Institute Inc., Cary, NC, the USA), or Review Manager 5 software (Cochrane Collaboration, London, the United Kingdom).
Assessment of heterogeneity
Heterogeneity was assessed using the I2 statistical method, with I2 > 50% or p-value < 0.05 indicating significant heterogeneity. We want to perform subgroup analyses if the following data are available: age (elderly/adult) and past medical history (left ventricular dysfunction or chronic renal insufficiency)
Sensitivity analyses
We will assess for robustness by excluding studies with a high risk of bias.
Assessment of reporting bias
We will not assess publication or reporting bias as there are no accepted method that can be used for its evaluation in a meta-analysis of diagnostic test accuracy studies[32].