Patients were recruited at the Emergency Departments (EDs) in Christchurch Hospital, New Zealand (NZ; n = 479) and the National University Hospital, Singapore (SG; n = 475) according to matched recruitment criteria (12). Recruitment for the NZ cohort occurred from 2007 to 2019 and SG between 2010 and 2013. Shortness of breath as the primary complaint triggering presentation to the ED was the key inclusion criterion. Exclusion criteria included; under 18 years of age, shortness of breath related to trauma, patients with an acute ST-elevation myocardial infarction and those on haemodialysis for renal failure. This study was performed in compliance with the principles outlined in the Declaration of Helsinki. All patients gave written informed consent to participate in the studies. Study protocols were approved by the local ethics committee of the two institutions.

The adjudicated diagnosis of ADHF was made in accordance with the 2016 ESC taskforce guidelines (1), by two independent clinicians (an ED specialist and a cardiologist) with access to all clinical data whilst blinded to NT-proBNP and ERFE results. In NZ, disagreement between the two adjudicators was resolved by consensus and in SG, a third adjudicator gave a final diagnosis.

Assessment included a comprehensive medical history, routine biochemical analyses, chest radiography, ECG recordings and pulse oximetry. Venous blood samples drawn at presentation were taken into EDTA collection tubes, the plasma rapidly separated, snap frozen at −80°C and biobanked at until assay.

Human ERFE was measured using a commercially available ELISA kit (SKU# ERF-001, Intrinsic Lifesciences) and performed following the manufacturers’ instructions. All ERFE measurements were conducted at the Christchurch Heart Institute, New Zealand. The inter-assay coefficient of variation (CV) of low (5 ng/ml, 37%) and high (35 ng/ml, 36%) quality control samples, with intra-assay CVs at 28% and 25% were derived over 16 and 29 assays, respectively. NT-proBNP measurements were assessed using the commercially available Elecsys proBNP II assay, a chemiluminescent two-site assay conducted on the Roche Cobas e411 analyser (Roche Diagnostics GmbH, Mannheim, Germany). NT-proBNP had an inter-assay CV of 5.5% and 5.7% for the low (845 pg/ml) and high (4860 pg/ml) quality control samples, respectively. Haemoglobin measurements were determined at the core biochemistry laboratories of the respective institutions.

Values are expressed as median [interquartile ranges (IQRs)], or counts and percentages as appropriate. The Shapiro–Wilk test of normality was used to determine whether the data were normally distributed and where applicable, skewed data were normalised by log₁₀-transformation. Pearson's correlation coefficient was calculated on log₁₀ transformed variables to examine any relationships between ERFE concentrations and other continuous variables. T-tests were used for comparisons of continuous measures between the two cohorts. The prevalence of ADHF, AF and obesity were compared between the two cohorts using Chi-squared tests. Multiple linear regression analysis was used to identify any independent associations of demographic and clinical features to plasma ERFE concentrations within each cohort.

The overall discriminatory power of markers (ERFE and NT-proBNP) for the diagnosis of ADHF was analysed separately in NZ and SG patients and then further analysed after stratification for presence of AF and obesity. This was assessed using areas under the ROC curves (AUCs) derived from raw data for ERFE and NT-pro-BNP, and from binary logistic models using log₁₀ transformed values after adjustment for significant demographic and clinical risk factors within each cohort. The sensitivity, specificity, positive (PPV) and negative predictive values (NPV) and accuracy of ERFE and NT-proBNP for the diagnosis of ADHF were calculated using ROC-derived optimal values at the point closest to perfect sensitivity and specificity (13). Pair-wise comparisons were used to determine if ROC curves were statistically different from each other. All statistical assessments were made using SPSS v26 (IBM Corp, Armonk, NY) and graphical representation using SigmaPlot v14.5 (Systat Software, San Jose, CA). In all analyses, a p-value of <0.05 was considered statistically significant.

Data from a total of 954 patients were included in this analysis; (479 NZ and 475 SG). Characteristics for Western and Asian cohorts are presented in Table 1 and Supplementary Table S1. The NZ cohort consisted largely of NZ European patients (n = 430, 90%) with minorities of Māori (3%), Pacific Islander (0.8%) and other ethnicities (6%). The SG cohort was more ethnically diverse including Chinese (45%), Malay (29%), Indian (18%) and others (8%). As previously published (12), recruitment criteria were the same in both countries. As shown in Table 1, the two cohorts differed in age (younger in SG), renal function (reduced in NZ) and key elements of medical history. The adjudicated diagnosis of ADHF was higher in the NZ cohort (35.7%, n = 171) compared with SG cohort (24.2%, n = 115, p < 0.001). The overall prevalence of AF was more frequent in the NZ cohort (23.1%, n = 105) than the SG cohort (10.1%, n = 44, p < 0.001). Prevalence of obesity in Western patients (classified as BMI ≥ 30 kg/m²) (34%, n = 150) was lower than Asian patients (classified as BMI ≥ 27.5 kg/m²) (40%, n = 163, p = 0.039).

Overall median ERFE levels were significantly higher in dyspneic NZ vs. SG patients (2.4 [IQR 0.9–6.1] ng/ml vs. 0.71 [IQR 0.19–2.9] ng/ml, p < 0.001). ERFE and NT-proBNP levels were not normally distributed in either population and were log₁₀ transformed prior to analyses. Plasma ERFE correlated with several clinical variables in both populations (Table 2), with the strongest associations being to NT-proBNP (NZ: r = 0.43, SG: r = 0.57, both p < 0.001), haemoglobin (NZ: r = −0.39, SG: r = −0.30, both p < 0.001) and to normal chest x-ray (CXRnorm) in SG (r = −0.44, p < 0.001).

Putative predictors were assessed for independent associations with plasma ERFE in multivariable analyses conducted separately in each cohort, incorporating age, gender, BMI, height, haemoglobin, eGFR, ejection fraction (NZ only), presence of AF on ECG, history of hypertension, myocardial infarction (MI), chronic heart failure (CHF), chronic obstructive pulmonary disease (COPD) and CAD (Table 3). In both countries, pulmonary rales and low haemoglobin levels were independent predictors of increased plasma ERFE and presence of a normal chest x-ray with lower levels of plasma ERFE. Independent associates of ERFE that differed between the two countries were BMI and AF in the NZ cohort, and presence of arrhythmia, orthopnoea and diabetes in the SG cohort. These models generated adjusted r² values of 0.27 and 0.29 for NZ and SG, respectively, indicating most (70%–75%) of inter-individual variation in ERFE is determined by other factors not measured within this study. Table 4 lists the variables remaining independently associated with the diagnosis of ADHF after multivariable analysis for the two cohorts separately. Plasma ERFE, NT-proBNP and the presence of interstitial oedema on chest x-ray (CXRinoed) were the only variables in common to both cohorts. Four further variables were significant in the NZ cohort (COPD, normal chest x-ray, paroxysmal nocturnal dyspnea and AF) with three others in SG (history of CHF, orthopnoea and pulmonary rales).

Median ERFE levels were elevated in ADHF in both cohorts compared with other diagnoses (NZ: 4.9 [2.4–11.0] ng/ml vs. 1.4 [0.6–3.2] ng/ml, p < 0.001 and SG: 4.2 [1.7–8.0] ng/ml vs. 0.4 [0.1–1.4] ng/ml, p = 0.021). Accordingly, logistic regression revealed ERFE to be independently associated with the diagnosis of ADHF in both the NZ and SG populations (both p = 0.003). This relationship was weaker than NT-proBNP which was more tightly associated with ADHF (p < 0.001 in both cohorts). Table 5 compares test performance characteristics for ERFE and NT-proBNP in the diagnosis of ADHF in the overall dyspneic population and the comorbid subgroups. In the discrimination of ADHF, ERFE performed strongly but was inferior to NT-proBNP. Both ERFE and NT-proBNP performed significantly better in the Asian cohort than the Western cohort (Figure 1 and Table 5): ERFE AUCs 0.84 vs. 0.75 (p = 0.006) and NT-proBNP AUCs 0.92 vs. 0.86 (p = 0.004). For ERFE, sensitivity and PPV were comparable between NZ and SG, respectively; whereas specificity, NPV and accuracy were better in SG compared to NZ (Table 5). NT-proBNP performed similarly to previously published data (12) in these two cohorts.

As AF is known to impair the performance of NT-proBNP in diagnosing ADHF, we assessed whether the diagnostic performance of ERFE in the presence of AF may be impacted. In both cohorts, median ERFE levels were elevated in AF compared with non-AF (NZ: 4.1 [1.9–10.3] ng/ml, n = 105 vs. 2.0 [0.7–4.8] ng/ml, n = 349, p < 0.001; and SG: 2.9 [1.3–5.4] ng/ml, n = 44 vs. 0.7 [0.2–2.8] ng/ml, n = 392, p < 0.001). In univariate analysis for the discrimination of ADHF in the presence of AF, AUCs in both countries are reduced and similar (NZ; 0.71, SG; 0.61, p = 0.331), and not inferior to that of NT-proBNP (NZ; p = 0.968, SG; p = 0.728) (Table 5). Specificity was comparable for ERFE in both countries, whereas PPV, NPV and accuracy were slightly better and sensitivity substantially better in NZ than SG.

As obesity has also been shown to reduce the discriminatory power of NT-proBNP in diagnosing ADHF, we explored the impact of BMI on the performance of ERFE. Median ERFE levels were similar between patients with obesity and non-obese patients in both cohorts (NZ: 2.7 [0.99–7.8] ng/ml, n = 148 vs. 2.3 [0.81–5.1] ng/ml, n = 302, p = 0.579 and SG: 0.81 [0.22–3.6] ng/ml, n = 163 vs. 0.69 [0.19–2.8] ng/ml, n = 241, p = 0.902). Univariate analysis for the discrimination of ADHF in patients with obesity revealed ERFE to perform slightly better in the Asian cohort (AUC 0.87, n = 164) compared with the Western cohort (AUC 0.79, n = 150), but this was not significant (p = 0.066). NT-proBNP was significantly better than ERFE at diagnosing ADHF in SG (AUC 0.94, p = 0.027) but not NZ (AUC 0.87, p = 0.073). For ERFE, sensitivity, PPV and accuracy were comparable between countries, whereas specificity and NPV trended higher in SG than NZ (Table 5). The AUCs for ERFE and NT-proBNP in the subgroup of patients with obesity, were similar to those of the overall dyspneic population for both countries.