Prognostic value of cardiac biomarkers and National Early Warning Score 2 in acute dyspnoea

Akershus Cardiac Examination 2 Study

The ACE 2 Study was a prospective single-centre cohort study investigating the diagnostic and prognostic properties of established and novel biomarkers in patients admitted with acute dyspnoea. Study details have previously been published in full.11 In short, the study was conducted at Akershus University Hospital from June 2009 to November 2010. Patients hospitalised due to acute dyspnoea were included if they were ≥18 years of age and able to provide informed consent. Dedicated study personnel screened for eligible patients during the daily morning briefings at the ED, and additional blood sampling was done within 24 hours. Patients with short life expectancy, as well as patients who had gone through major surgery, acute myocardial infarction, or coronary intervention within the last 2 weeks, were excluded. Of 468 patients hospitalised with acute dyspnoea, 314 patients were included in the final study cohort (figure 1). There was no patient or public involvement at the time the study was conducted (2009–2010).

Data collection

All patients completed a standardised questionnaire and clinical information was obtained directly from the ED physician. Additional data were collected from the hospital’s electronic records, including data collected prior to and/or after the current hospitalisation when available. Medical history was dichotomised, with all temporal patterns of atrial fibrillation (AF) being defined as AF, and previous myocardial infarction and/or coronary intervention being defined as coronary artery disease (CAD). Body mass index (BMI) was calculated as body weight/[height ×height] (kg/m²). Left ventricular ejection fraction (LVEF) and diastolic dysfunction were determined based on clinical routine transthoracic echocardiography. NEWS2 was retrospectively calculated based on vital parameters registered in the ED, and included data on respiratory rate, oxygen saturation, hypercapnic respiratory failure, need of supplemental oxygen, systolic blood pressure, heart rate, consciousness and body temperature. Variables were assigned a point value from 0 to 3 based on degree of abnormality and summed up. An aggregated score of 0–4 was assigned low risk, a score of 3 (full score) in any individual parameter was assigned low-medium risk, an aggregated score of 5–6 was assigned medium risk, and an aggregated score of 7 or more was assigned high risk, as described by The Royal College of Physicians of London.9

Adjudication of diagnosis and follow-up data

The index diagnosis of the hospitalisation was adjudicated by two senior physicians who had access to all data in the hospital’s electronic journal system. The adjudicators reviewed all relevant data individually and in cases of disagreement, the diagnosis was decided by consensus. HF diagnoses were based on ESC HF guidelines, requiring worsening of typical symptoms and clinical signs, and objective evidence of structural or functional myocardial abnormality, and subdivided according to LVEF. Patients with EF ≥50% and echocardiographic evidence of altered structure or dysfunction (including pathologic mitral E/A, E deceleration time, E/e’ and left atrial enlargement) were diagnosed with HF with preserved EF (HFpEF). In this study we did not include the intermediate group HF with mildly reduced EF (EF 40%–50%), and these patients were pooled with the group with EF <40% as HFrEF patients. We obtained survival status on 1 November 2012, from the hospital’s electronic records, which are monthly synchronised with Statistics Norway.

Biochemical measurements

Blood samples were collected within 24 hours of hospitalisation. Test tubes were centrifuged, and serum was immediately frozen and stored at −80°C. Measurements were performed without prior freeze-thaw cycles, ensuring excellent stability for both analytes as previously reported.12 13 hs-cTnT concentrations were measured with a high sensitivity Roche assay (Elecsys TnT hs stat, Roche Diagnostics, Penzberg, Germany) with a range of detection from 3 to 10 000 ng/L, a 10% coefficient of variation of 13 ng/L, and a 99th percentile in healthy individuals of 14 ng/L. Values below the limit of detection were assigned the value 3 ng/L. NT-proBNP was measured with the pro-BNP II assay from Roche (Roche Diagnostics, Basel, Switzerland), with a range of detection from 5 to 35.000 ng/L. Glomerular filtration rate was estimated with the Chronic Kidney Disease Epidemiology Collaboration equation. 14

Statistical analysis

Continuous variables are presented as median (IQR) and compared with the Mann-Whitney U. Dichotomous variables are presented as absolute numbers (percentage) and compared with the χ² test. We calculated correlations with the Spearman rank method. Univariate and multivariable linear regression were used to assess association between explanatory variables and NEWS2, whereas univariate and multivariable Cox regression were used to identify predictors of all-cause mortality. In both cases, we used backward selection on all the significant variables from univariate regression to establish the final explanatory model, analogously to the strategy of previous studies from this cohort.11 We base this approach on the purposeful selection of covariates for models with dichotomised outcomes as described by Hosmer and Lemeshow.15 The prognostic accuracy of NEWS2 was assessed as a continuous variable, as a categorical variable with the four recommended clinical risk groups (1) low risk; (2) low-medium risk; (3) medium risk; (4) high risk, and as a dichotomous variable (low risk vs low-medium, medium, and high risk). Kaplan-Meier survival curves were stratified by NEWS2 clinical risk, and differences in survival rates were tested by the log-rank test. The accuracy for biomarkers and NEWS2 to predict mortality were assessed by receiver operating characteristic curve (ROC) analysis with the area under the curve (AUC). For 11 of the 314 patients, respiratory rate was missing when calculating NEWS2. For these cases, we used multiple imputation with predictive mean matching for the five closest neighbours with five imputations per missing variable. All biomarkers were transformed by the natural logarithm before analyses due to a right-skewed distribution. We used two-tailed tests and considered p<0.05 significant. Kappa statistics were used to calculate consistency between the adjudicators. All analyses were performed with Stata (V.16, StataCorp). We used the TRIPOD checklist when writing our report.16

Patient characteristics

In total, 314 patients were included in the current study. Forty-eight per cent were women, median age was 73 (Q1–3, 63–81) years, median BMI 26 (22–30) kg/m² and 27% were current smokers (table 1).Thirty-two per cent had a history of HF and 49% of chronic obstructive pulmonary disease (COPD). In total, 143 of 314 hospitalised patients (46%) were adjudicated with AHF, with a consensus rate of 95% (κ=0.897 (0.848–0.946)). AHF patients were older, more often male and had more comorbidities than patients with non-HF-related dyspnoea. HFpEF (n=52) patients were more often older females with de novo HF and had lower prevalence of established CAD compared with HFrEF patients (n=91) (online supplemental table 1). We report discharge medication for the HF patients in online supplemental table 2.

NEWS2 and cardiac biomarkers on hospital admission

The NEWS2 ranged from 0 to 12 with median 5 (Q1–3, 3–7). In adjusted analysis, New York Heart Association class IV, history of COPD and higher C-reactive protein and hs-cTnT concentrations were associated with increasing NEWS2 (online supplemental table 3). The distribution of vital parameters contributing to the total NEWS2 were mostly similar between the different causes of acute dyspnoea; AHF, AECOPD or other (table 1), and between HFpEF and HFrEF (online supplemental table 4). Patients with AHF had higher peripheral oxygen saturation (94% vs 91%, p<0.001) and lower temperature (36.8℃ vs 37.2℃, p<0.001), compared with patients with AECOPD. We found no significant differences for respiratory rate, supplemental oxygen, systolic blood pressure, heart rate or consciousness. In total, patients with AHF had lower NEWS2 (5.1), compared with AECOPD patients (5.9; p=0.039). Among AHF patients, the HFpEF group had lower peripheral oxygen saturation and higher temperature (online supplemental table 4).

The median concentration of NT-proBNP was 1926 ng/L (Q1–3, 446–5468) for the total cohort, 3600 (1601–8396) for AHF and 379 ng/L (171–1008) for AECOPD. With a similar pattern, hs-cTnT was 27 ng/L (16-53) in the total cohort, and 38 ng/L (22-75) and 18 ng/L (10-28) for AHF and AECOPD, respectively. The concentrations and prognostic accuracies of hs-cTnT and NT-proBNP in this cohort have previously been reported.17 hs-cTnT correlated with respiratory rate, the need of supplemental oxygen, systolic blood pressure and total NEWS2, whereas NT-proBNP correlated with respiratory rate, peripheral oxygen saturation, the need of supplemental oxygen, systolic blood pressure and temperature (online supplemental table 5). The correlation coefficient between NEWS2 and hs-cTnT was 0.30 (p<0.001) and the correlation coefficient between NEWS2 and NT-proBNP was 0.20 (p<0.001).

Predictors of long-term mortality

One-hundred and fourteen patients (36%) died during a median follow-up time of 823 days (Q1–3, 471–998). NEWS2 as a continuous variable, hs-cTnT and NT-proBNP all predicted mortality in unadjusted analysis. hs-cTnT (HR 1.35, 95% CI 1.12 to 1.61) and NEWS2 (HR 1.08, 95% CI 1.02 to 1.16) were also significant in multivariable analysis, while the association was attenuated and no longer significant for NT-proBNP in the final multivariable model (table 2).

By stratifying patients according to NEWS2 clinical risk groups, we found that the lowest risk group had the highest survival rates (figure 2A; p<0.001 by the log rank test). The association between elevated NEWS2 clinical risk and survival was significant in both unadjusted (HR 2.74, 95% CI 1.66 to 4.51 for elevated compared with low clinical risk, p for trend <0.001) and adjusted analyses (HR 2.11, 95% CI 1.28 to 3.48, p for trend=0.034) (online supplemental table 6). NEWS2 predicted mortality independent of cardiac biomarker concentrations, and patients with elevated NEWS2 risk grade in combination with concentrations of hs-cTnT (figure 2B) and/or NT-proBNP (figure 2C) above the median (n=149) had the highest mortality rate during follow-up, with a HR of 5.64 (95% CI 2.45 to 12.97), compared with those with low NEWS2 clinical risk and cardiac biomarkers below the median (n=50) (online supplemental table 6). Among the variables contributing to NEWS2, higher respiratory rate (HR 1.07, 95% CI 1.04 to 1.09) and lower systolic blood pressure (HR 0.99, 95% CI 0.98 to 0.99) predicted mortality in adjusted analyses (online supplemental table 6). The ROC-AUC of NEWS2 to predict mortality was 0.64 (95% CI 0.58 to 0.71), which was not statistically different from hs-cTnT 0.69 (95% CI 0.64 to 0.75, p for comparison=0.22) or NT-proBNP at 0.67 (95% CI 0.61 to 0.73, p for comparison=0.64).

Among the subgroup of patients with AHF, 66 patients (46%) died. In unadjusted analysis for this subgroup, NT-proBNP (HR 1.53, 95% CI 1.24 to 1.89), hs-cTnT (HR 1.37, 95% CI 1.10 to 1.71) and NEWS2 (HR 1.12, 95% CI 1.03 to 1.22) predicted mortality (online supplemental table 7). Among these, only NT-proBNP still predicted mortality in multivariable analysis (HR 1.46, 95% CI 1.17 to 1.83). Stratifying patients by NEWS2 clinical risk also predicted mortality for the subgroup with AHF (p=0.018 by the log-rank test, (online supplemental figure). Adding NEWS2 to both biomarkers resulted in a higher AUC for mortality compared with the two biomarkers alone for the total cohort (AUC 0.72, 95% CI 0.66 to 0.78 vs 0.70, 95% CI 0.64 to 0.75, p for comparison=0.042), and for the subpopulation with AHF (AUC 0.70, 95% CI 0.60 to 0.78 vs 0.67, 95% CI 0.58 to 0.76, p for comparison=0.014) (online supplemental table 8). The prognostic accuracy of NEWS2 was higher for HFpEF compared with HFrEF (AUC 0.80, 95% CI 0.68 to 0.93 vs 0.56 95% CI 0.43 to 0.68, p for comparison=0.005) (online supplemental table 8).

Strengths and limitations

In this study, we used an adjudication committee consisting of two independent experienced physicians to set the index diagnoses according to current guidelines, which provided an excellent consensus between the adjudicators. All relevant data were collected in a uniform matter by dedicated study personnel, and hs-cTnT and NT-proBNP concentrations were measured at a core laboratory as a batch to avoid variation in lab calibration over time. As limitations, this study had moderate sample size and a single-centre design. Respiratory rate, which was one of the variables used to calculate NEWS2, was missing in 11 of 314 patients. This was handled by imputing missing values based on the other NEWS2 variables and outcome, and imputations did not significantly change any results compared with analyses of original datasets. Even though adjudication consensus was excellent in this study, some of the non-AHF patients in the study never underwent echocardiography, which could have contributed to underdiagnosis of AHF.