B-type natriuretic peptides in chronic obstructive pulmonary disease: a systematic review

Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines were followed. The population of interest was patients with COPD receiving natriuretic peptide testing. The outcome of interest was NP, including: levels and proportion elevated in different COPD populations, stratified by COPD severity (stable disease, acute exacerbation (AECOPD), associated cor pulmonale); thresholds used to define abnormal; correlations between NP and measures of ventricular and pulmonary function; risk associated with NP; and accuracy of NP in diagnosing HF. All study designs including cohort, case-control and cross-sectional were accepted.

MEDLINE (from 1990), EMBASE (from 1990), and the Cochrane Library were searched to June 2015, limited to adult humans, without date or language restriction. Search terms were selected by consensus and iterative database queries. Medical Subject Headings (MeSH) and Emtree terms were identified from keyword mapping and published literature. COPD was identified using MeSH (pulmonary disease, chronic obstructive; bronchitis, chronic), Emtree (chronic obstructive lung disease; chronic bronchitis), and keywords. NP were identified using MeSH (natriuretic peptides), Emtree (brain natriuretic peptide), and keywords. Terms and keywords were combined according to the requirements of the database. The search strategy is outlined in Appendix 1. No review protocol was registered or published. The search identified 440 articles in Medline and EMBASE, totalling 276 records after duplicate removal (Fig. 1). Case reports, reviews and conference abstracts were excluded. Two reviewers (NH and AK) screened titles and abstracts (binary yes/no) with reconciliation through discussion. Studies fulfilling the participant, outcomes and study design criteria were included. Studies involving patients with different pulmonary diseases (as opposed to COPD) or only HF were excluded (Fig. 1). Variables of interest were decided a priori and expanded iteratively after pilot. Excel spreadsheets were employed as data extraction forms and populated directly by both reviewers (NH and AK). The following information was extracted: bibliographic details, sample size and number of centers, population, baseline characteristics and comorbidities, pulmonary function, NP outcomes.

In accordance with the Cochrane Collaboration and Institute of Medicine guidance, risk of bias in observational studies was assessed in selected components with empirical evidence and strong clinical or theoretical grounds. A quality scale was not utilized as many have limited development methodology, validation, arbitrary weightings and inconsistent relationships with effect sizes. 7 bias domains were selected (selection, misclassification, performance, detection, reporting, information and confounding), based on the Cochrane Collaboration Risk of Bias Tool and Handbook and Agency for Healthcare Research and Quality RTI Item Banks,[11‐13] Judgement of low, high or unclear risk of bias was assigned for each domain (Appendices 2, 3 and 4).

The evidence is presented as a narrative synthesis given the heterogeneous populations, diverse objectives and outcomes examined, varying assays and thresholds, and poorly defined confounding factors. Most importantly, the summary measures presented in many studies (median and ranges) require transformation for meta-analysis. We explored multiple transformation methods, [14‐16] all of which declined in accuracy with increasing skew and underestimated the variance by up to half. We identified 4 main groups (stable COPD/BNP, stable COPD/NT-proBNP, exacerbation COPD/BNP, exacerbation COPD/NT-proBNP). Median/IQR was more often reported in the exacerbation and NT-proBNP studies due to skewed distributions (Table 1). Thus transformation for meta-analysis would introduce major error into already large variances in a systematic manner.

Risk of bias in many domains was low with respect to measurement of NP. Studies were typically small, prospective, without interventions or exposures, cohort or cross-sectional in design, and measured NP in all patients using commercial validated assays. However, approximately 50% of studies exhibited selection bias, 20% lacked objective definition of COPD, and 40% failed to report sufficient information to facilitate interpretation of NP levels (e.g. presence of HF) (Appendices 3 and 4).

The most consistent association was between NP and pulmonary artery pressure, with correlation coefficients ranging from 0.28 to 0.68, typically being around 0.5 (Table 2). In most studies with echocardiography, NP elevation was associated with left ventricular ejection fraction (LVEF) among patients with stable and exacerbation of COPD, [25, 26, 32, 35, 37, 39] even in the absence of raised pulmonary artery pressures. Right ventricular function was rarely characterized, and then using a variety of measures including ejection fraction,[19] tricuspid annular plane systolic excursion (TAPSE), [37] right ventricular diameter and hypokinesia [29, 31]. Heterogeneity and small sample sizes limits interpretation.

The relationship between NP and FEV₁ or PaO₂ was inconsistent. Similar to the evidence stratifying by COPD severity, the smaller studies observed significant correlations between NP and both FEV₁ or PaO₂. However, correlation coefficients in the two largest studies of 208 and 80 patients were not significant (respectively FEV₁ r = 0.104 and PaO₂ 0.115; FEV₁ r = 0.06 and PaO₂ 0.14). A modest significant association was observed between NP and troponin in three studies (r = 0.34 to 0.50) [38, 44, 45].

Different strategies have been employed to define ‘abnormal’ (Table 3): ROC curve analysis to balance accuracy in predicting specific outcomes; measuring central tendency and dispersion of normal controls (e.g. mean ± 2 SD); manufacturer recommendation; existing publications or investigator selection. The proportion of patients with elevated NP according to these heterogeneous thresholds ranged from 15 to 71% in stable patients, and 16% to 60% during exacerbation. Five studies employed receiver operating curve analysis to determine optimal thresholds for detecting left ventricular dysfunction [24, 32, 39, 43, 46]. However, only one of these studies actually reported the prevalence of an elevated level according to these thresholds (approximately 50% in stable patients) [24]. Moreover, identical thresholds in different studies yielded very different frequencies of elevation. NT-proBNP >125 pg/ml occurred in 23% and 51% of stable patients in two studies [24, 30]. Likewise, NT-proBNP >125 pg/ml occurred in 16%, 27% and 44% of AECOPD in three studies [37, 38, 44].

Natriuretic peptides were always significantly elevated in patients with COPD and concurrent HF or LVSD compared to those without (Table 1). However, very few studies examined predictive accuracy to identify HF or LVSD, with just a single study in patients with stable COPD (Table 4) [24]. Four natriuretic peptide assays produced comparable results in 200 stable elderly patients with a clinical diagnosis of COPD. Each test excluded HF with reasonable accuracy (all negative predictive values above 0.85, with positive predictive values approximately 0.4). In three studies of patients with AECOPD, NP demonstrated high negative predictive values (0.80 to 0.98) to exclude left ventricular dysfunction applying thresholds exceeding the manufacturers’ guidance (Table 4) [32, 46, 47]. However, as in the stable population the positive predictive values were relatively low. Two studies also assessed ability to detect systolic and diastolic dysfunction separately [24, 47]. The receiver operating characteristic areas and overall accuracy in the latter were lower though remained acceptable.

We identified 12 studies (6 stable and 6 exacerbation of COPD) reporting the association between NP and prognosis, in which the prognostic significance of elevation was inconsistent (Table 5). Among stable patients, the association between NP and survival over 1 to 4 years failed to remain significant after multivariable adjustment in 3 studies [25, 35, 48]. However, NT-proBNP >500 pg/ml predicted one year mortality in 144 patients with predominantly mild to moderate COPD and preserved LVEF (>40%) undergoing major vascular surgery (adjusted HR 7.7 [95% 1.6–37.4]) [49]. NT-proBNP was also associated with all-cause mortality in a larger cohort of 220 elderly men with COPD (adjusted HR 1.61 [1.27–2.06]), although 26% of that cohort had documented HF [50].

In patients with AECOPD, NP independently predicted short term outcomes including intensive care unit admission, [33] inpatient and 30 day mortality [44, 51]. Median BNP was also significantly higher in failed (inpatient death or early re-hospitalisation) compared to successful discharges following AECOPD hospitalization (median (IQR) 261 (59–555) vs 49 (24–104) pg/ml) [36]. The relationship with longer term survival was less certain. Natriuretic peptides failed to predict mortality at 1 and 2 years in 244 and 208 consecutive patients hospitalized or presenting to the emergency department with exacerbation [33, 44]. However, elevated NP were independently associated with increased mortality at 6 months, 1 year and nearly 2 years in three subsequent studies (respectively HR 4.2, OR 3.3 and HR 3.2) [34, 45, 52].

Myocardial stretch in either ventricle consequent to volume or pressure overload increases NP levels [53]. Causes include heart failure with reduced and preserved ejection, [54, 55] right ventricular failure, [56] pulmonary emboli, [41, 57] acute coronary syndromes, [58, 59] valvular heart disease, [60] and arrhythmias [61]. Advancing age and renal dysfunction are also associated with elevated NT-proBNP concentrations [62]. Many of these factors are present in stable COPD and common non-infective precipitants of exacerbation [32, 41, 63]. The presence and extent of each factor varies significantly from patient to patient, and is largely independent of COPD severity or acute right ventricular dysfunction. Thus NP levels are higher during acute exacerbation or chronic decompensation (cor pulmonale) than stable disease, and exhibit significant variability with skewed distributions.

By systematically searching and aggregating individual studies, our review highlights several new and consistent observations which suggest NP release is multifactorial with limited direct relationship to COPD. First, NP levels are increased even in some patients with mild COPD without arterial hypoxaemia, severe pulmonary hypertension or right ventricular dysfunction. Second, levels are stable or exhibit only a minor gradient with increasing COPD severity. Third, the magnitude of the correlation coefficients (r) suggests only approximately 25% to 50% of the variance (r ²) in NP is attributable to any single variable. Moreover, correlation between left and right ventricular function is likewise modest (LVEF and TAPSE r = 0.46 in one study), [37] indicating only around 20% of the variance in function of either ventricle is explained by the function of the other.

Individual studies have concluded that NP may be useful in risk stratifying patients with COPD [34, 44, 49]. However, the overall literature has not previously been summarized. The association with longer term outcomes was inconsistent in both stable and exacerbation populations. Our findings highlight many of the challenges in developing biomarker strategies: relatively small sample sizes; variable performance in heterogeneous populations; and failure to replicate findings from derivation to validation cohorts [7]. At present there is insufficient evidence to recommend routine risk stratification using NP.

The more consistent prediction of early outcomes following exacerbations suggests that NP are more strongly associated with acute pathologies rather than COPD itself [33, 44, 51]. The precise causes remains unclear, as risk associated with many acute events improves with time e.g. HF, PE. Nevertheless, unrecognised LVSD undoubtedly underpins many adverse outcomes. While NP levels were typically modest, [44] up to one fifth of patients with AECOPD had marked elevation indicative of probable left heart failure (although acute right ventricular strain remains possible) [45]. Moreover, the significant unadjusted association between NT-proBNP and mortality in one study was nulled after adjustment for LVEF and valvular disease [25]. This hypothesis is further supported by the high prevalence of unrecognised heart failure in imaging and autopsy studies, [64] and the improved outcomes associated with angiotensin converting enzyme inhibitors and beta-blockers in observational COPD studies [65, 66].

Natriuretic peptides exhibit lower diagnostic accuracy for HF in COPD than in populations with acute dyspnoea, [67, 68] due to greater overlap of NP distributions in the respective states to be distinguished: levels are elevated in stable and exacerbation of COPD, and lower in stable compared to acute HF. The threshold providing adequate sensitivity and negative predictive value must generate sufficiently few false positives to integrate into systems of care, be cost-effective, and improve outcomes. However, the positive predictive values in the 3 stable or exacerbation populations we identified ranged from 0.4 to 0.47. This compares unfavourably with a recent meta-analysis of NP in the acute care setting, which reported positive predictive values ranging from 0.67 and 0.64 for BNP and NT-proBNP respectively at the guideline recommended lower thresholds, rising to 0.85 and 0.80 respectively for mid-range values [69]. The resulting increase in false positive results will increase demand on imaging services to confirm or refute the diagnosis.

To improve generalizability and interpretation, future studies should use validated assays in consecutive patients, and standardized definitions for COPD, HF and comorbidities. Detailed cardiovascular profiles and imaging are needed to systematically define pathologies contributing to NP elevation. Levels should be reported using guideline and manufacturer recommended thresholds, for both the overall population and stratified according to presence or absence of predictors of NP elevation, particularly left ventricular dysfunction. Larger studies examining cause-specific outcomes are needed. Integrating NP with clinical variables and simple investigations such as electrocardiograms should be evaluated to reduce false positive results and develop cost-effective screening strategies. The goal of improving outcomes is particularly challenged by the inconsistent prognostic implications of NP in COPD in studies to date. The greatest incremental prognostic and therapeutic value is likely in populations with unrecognized heart failure and cardiovascular disease amenable to treatment [34, 45, 70].

Most of the identified studies were single centre with limited numbers of patients and endpoints. The patient populations, assays and cutoffs for NP, and definitions of LVSD and HF were heterogeneous. No study systematically defined causes of NP elevation, and the proportion amenable to therapy e.g. arrhythmia, ischaemia, LVSD, pulmonary emboli. These comorbidities will strongly influence every outcome examined, from symptoms to prognosis. The causes of death in relation to NP elevation also require clarification.