Diagnostic accuracy of point-of-care natriuretic peptide testing for chronic heart failure in ambulatory care: systematic review and meta-analysis

Abstract Objective To assess the diagnostic accuracy of point-of-care natriuretic peptide tests in patients with chronic heart failure, with a focus on the ambulatory care setting. Design Systematic review and meta-analysis. Data sources Ovid Medline, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects, Embase, Health Technology Assessment Database, Science Citation Index, and Conference Proceedings Citation Index until 31 March 2017. Study selection Eligible studies evaluated point-of-care natriuretic peptide testing (B-type natriuretic peptide (BNP) or N terminal fragment pro B-type natriuretic peptide (NTproBNP)) against any relevant reference standard, including echocardiography, clinical examination, or combinations of these, in humans. Studies were excluded if reported data were insufficient to construct 2×2 tables. No language restrictions were applied. Results 42 publications of 39 individual studies met the inclusion criteria and 40 publications of 37 studies were included in the analysis. Of the 37 studies, 30 evaluated BNP point-of-care testing and seven evaluated NTproBNP testing. 15 studies were done in ambulatory care settings in populations with a low prevalence of chronic heart failure. Five studies were done in primary care. At thresholds >100 pg/mL, the sensitivity of BNP, measured with the point-of-care index device Triage, was generally high and was 0.95 (95% confidence interval 0.90 to 0.98) at 100 pg/mL. At thresholds <100 pg/mL, sensitivity ranged from 0.46 to 0.97 and specificity from 0.31 to 0.98. Primary care studies that used NTproBNP testing reported a sensitivity of 0.99 (0.57 to 1.00) and specificity of 0.60 (0.44 to 0.74) at 135 pg/mL. No statistically significant difference in diagnostic accuracy was found between point-of-care BNP and NTproBNP tests. Conclusions Given the lack of studies in primary care, the paucity of NTproBNP data, and potential methodological limitations in these studies, large scale trials in primary care are needed to assess the role of point-of-care natriuretic peptide testing and clarify appropriate thresholds to improve care of patients with suspected or chronic heart failure.

Patients with unstable angina or whose dyspnea was clearly not secondary to congestive heart failure (for example, those with trauma or cardiac tamponade) and those with acute myocardial infarction or renal failure. Mak 2004 Referrals for echocardiographic evaluation of cardiac structure and function Patients whose referral was to assess valvular disease, to determine whether vegetation was present, or to rule out a cardiac cause of stroke.
Patients with known significant valvular disease referred for other reasons or found to have significant valvular disease Monfort 2015 Patients with stable CHF who were followed at the cardiac rehabilitation unit at Lariboisière University Hospital None stated.

Morrison 2002
Patients presenting to urgent care with dyspnea, with shortness of breath either at rest, with exertion or upon lying down, as a prominent complaint. Other associated symptoms could be edema, weight gain, cough or wheezing.
Patients whose dyspnea was clearly not secondary to CHF (knife wounds, trauma and cardiac tamponade   Figure S4. Sensitivity analysis: Summary Receiver-Operating-Characteristics plots for B type natriuretic peptide and N-Terminal pro-B type natriuretic peptide, compared with clinical assessment, for cross-sectional/cohort/randomised controlled trial studies of populations in ambulatory settings, selecting the lowest threshold for each study.
Based on data for 17 studies; 4 NT-proBNP and 13 BNP studies. Size of symbol indicates study size. There were insufficient studies to draw meaningful prediction and confidence regions for NT-proBNP. BNP -Triage, black circle; NT-proBNP -Cardiac Reader, red triangle.

Appendix 1. Research protocol
Systematic review: Accuracy of point-of-care B-type natriuretic peptide testing in the diagnosis and monitoring of chronic heart failure Introduction Target condition: An estimated 800,000 people in the UK currently have heart failure with an estimate of over 250 000 new cases of heart failure in the UK every year. Incidence increases with age and is highest in adults over 75 [1]. The ageing population and improved survival of individuals with ischaemic heart disease are likely to lead to a continuing rise in the prevalence of heart failure. Overall a general practitioner (GP) with a patient population of 2000 will care for approximately 40-50 patients with heart failure and see 2-3 new cases each year.
Since heart failure may be reversible in the early stages it is important that heart failure is diagnosed as quickly as possible. However, because it has a low incidence, GPs are unlikely to have sufficient experience to identify more subtle presentations of heart failure. For example, whilst heart failure is frequently diagnosed by GPs, it is only confirmed by echocardiography in approximately a third of cases [2]. A recent health technology assessment of the use of B-type natriuretic peptide (BNP) in the diagnosis of heart failure, compared with ECG and echocardiography found that a normal ECG can be used to exclude a diagnosis of heart failure but it is relatively non-specific. Whilst ECG is a sensitive test when performed by cardiologists (sensitivity = 89%; 95% CI 77%-95%) [3], its sensitivity was much lower (53%) when carried out by GPs [4]. It is for this reason that the most recent NICE guidelines on chronic heart failure (CHF) have refined the diagnostic algorithm for heart failure, replacing ECG with serum BNP measurement [5]. POC BNP testing would enable GPs to rapidly refer the appropriate patients or, if CHF can be excluded, investigate alternative causes of dyspnoea.

Recent National Institute for Health and Clinical Excellence (NICE) and European Society for
Cardiology (ESC) guidelines on the initial diagnosis of CHF and referral for echocardiography recommend the use of BNP in combination with clinical assessment [5,6]. NICE guidelines recommend measurement of serum natriuretic peptides in patients with suspected heart failure without previous myocardial infarction (MI), and those with previous MI should be referred for an urgent echocardiogram [5]. Timely diagnosis and treatment of CHF may result in improved patient outcomes as has been demonstrated for acute decompensated heart failure (ADHF) [7]. Although several hospital laboratories carry out BNP testing, few return results within a day. Point-of-care (POC) BNP testing can considerably reduce turnaround time and could lead to earlier initial treatment, more timely referral and less uncertainty for patients. Moreover, using POC BNP levels to quickly rule out heart failure, could allow more rapid initiation of investigation of other causes of dyspnoea.
Index test: B-type natriuretic peptide is produced from heart muscle cells as a pro-hormone (proBNP) and released into the cardiovascular system in response to ventricular dilation and pressure overload [8].
The pro-hormone is split by a protease and secreted as the physiologically active C-terminal fragment (BNP) and the inactive N-terminal fragment (NT-proBNP). Several POC BNP testing devices are on the market, which either measure BNP or NT-proBNP (the latter has a longer half-life), for example: BNP: Importantly, BNP levels have been found to vary with age, gender and certain diseases (e.g. renal failure) [8]. It has been suggested that higher cut-off values are used when individuals are >75 years, female or in renal failure.

Rationale
Systematic reviews have been compiled on the role and accuracy of BNP and NT-proBNP in the diagnosis of chronic heart failure [3,12,13,15], however none have focussed specifically on the accuracy of point-of-care testing, particularly in the primary care setting. As mentioned above, several devices are currently on the market, however a systematic review on the evidence for the accuracy of the devices is currently lacking.

Clinical question
In patients presenting to primary care with suspected or confirmed chronic heart failure, what is the accuracy of point-of-care (POC) BNP devices compared to a reference standard (including laboratory-based tests, echocardiography or clinical examination or combinations of these) for the diagnosis and monitoring of chronic heart failure.

Therefore:
Population: patients with suspected chronic heart failure, or patients with chronic heart failure currently being monitored. The setting of particular interest is Primary Care, but patients from other settings (hospital, specialist clinic) will also be included

Index test: POC/near patient/bedside BNP tests
Reference standard: laboratory-based BNP tests or echocardiography or clinical examination or combinations of these.
Target condition: chronic heart failure

Objective
Compiling a systematic review and meta-analysis (if appropriate) of studies that investigate the diagnostic accuracy of POC BNP tests in patients with dyspnoea/suspected heart failure or with a diagnosis of heart failure. While the review will include studies from any setting, we will also seek to summarise the evidence in primary care, if possible.

Methods and Data Collection:
Inclusion criteria: 1. Study designs: To include cross-sectional, longitudinal and case-control designs; both prospective and retrospective. Randomised controlled trials, cohort and case-control studies will be included.
2. Index test and Reference standard: Studies comparing POC-BNP and/or POC NT-proBNP tests with any reference standard (including but not limited to laboratory-based tests, echocardiography or clinical examination or combinations of these). Studies comparing different POC tests with each other may also be considered, but will need to include a comparison with a laboratory or clinical reference standard. The index test will be any portable test that can be performed at the point of care that does not require laboratory services. Large bench-top devices will be excluded.
3. Outcomes: Studies reporting accuracy measures, such as sensitivity, specificity, likelihood ratios and predictive values, Bland-Altman plots etc. will be included.

Participants and setting: Studies with participants with suspected or confirmed Chronic Heart
Failure of any age in all settings will be included. No restrictions will be made regarding population numbers.
5. Study quality will not be a basis for exclusion.
6. Language: Any language. The search strategy will be developed in consultation with a healthcare librarian experienced with supporting systematic reviews. Multiple electronic databases will be searched including, but not confined to, MEDLINE, EMBASE, the Cochrane library, TRIP database and the Science Citation Index.
Free text and MeSH headings will be used. The search may use relevant filters, but in order to maximise sensitivity, will not be limited to these. The reference lists of relevant studies will be examined and additional tools such as the "related articles" feature in PubMed will also be used to identify relevant publications. Clinical trials registries will be searched and manufacturers of POC BNP/NT-proBNP devices will be contacted for information on any unpublished studies.

Study Selection:
Search results will be screened by two reviewers independently. Full text of relevant studies will be obtained and assessed for inclusion by each reviewer. Disagreements will be resolved by discussion or referral to a third reviewer. Study identification will be summarised in a PRISMA flow diagram.

Data extraction:
A data extraction form will be developed and data extraction will be performed independently by 2 reviewers, cross-checked and disagreements will be resolved by discussion or referral to a third reviewer.
Extracted data will include: Quality assessment: QUADAS2 (the Revised Tool for Quality Assessment of Diagnostic Accuracy Studies; [16]) will be used to assess the study quality of each included study and will be reported in table or graphics form.
Data analysis:

Summary Tables
Summary tables will detail the patient sample, study design, the test under evaluation, the thresholds at which accuracy has been reported, and the comparator.

Meta-analysis
For each test, RevMan will be used to produce paired forest plots to explore the between-study variability of sensitivity and specificity across the included studies. For each study estimate of sensitivity and specificity, corresponding 95% confidence intervals will be shown to illustrate the uncertainty related to each study estimate. If accuracy has been reported at multiple common thresholds, forest plots will be sub-grouped on threshold.
Bivariate meta-analysis methods will be used to generate pooled estimates of sensitivity and specificity where sufficient data is available for each test. Hierarchical summary ROC meta-analysis methods will be used to produce summary ROC curves with corresponding 95% confidence region and prediction region. We will use the xtmelogit command in Stata for these analyses and feed parameters directly into Revman to produce Cochrane-standardised output.

Investigations of heterogeneity
Two approaches will be used to explore the sources of between-study heterogeneity: 1) adjusting for possible sources by adding them as covariates to the bivariate model 2) carrying out sub-group analyses (e.g. subgroup analysis excluding studies using a case-control design, which is known to introduce bias). The latter will only be carried out if there is sufficient data available and sub-group specific pooled estimates are thought to be of clinical relevance.

Sensitivity analyses
If there appear to be any outliers in the data, these studies will be removed from the analysis to evaluate the impact on the overall pooled estimates. 11 Appendix 3. Variables in data extraction sheet Authors' analysis: Basic conclusions of study; our general comments