ACADEMIC EMERGENCY MEDICINE 2011; 18:947–958 © 2011 by the Society for Academic Emergency Medicine
Objectives: Few tools exist that provide objective accurate prediction of short-term mortality risk in patients presenting with acute heart failure (AHF). The purpose was to describe the accuracy of several biomarkers for predicting short-term death rates in patients diagnosed with AHF in the emergency department (ED).
Methods: The Biomarkers in ACute Heart failure (BACH) trial was a prospective, 15-center, international study of patients presenting to the ED with nontraumatic dyspnea. Clinicians were blinded to all investigational markers, except troponin and natriuretic peptides, which used the local hospital reference range. For this secondary analysis, a core lab was used for all markers except troponin. This study evaluated patients diagnosed with AHF by the on-site emergency physician (EP).
Results: In the 1,641 BACH patients, 466 (28.4%) had an ED diagnosis of AHF, of whom 411 (88.2%) had a final diagnosis of AHF. In the ED-diagnosed HF patients, 59% were male, 69% had a HF history, and 19 (4.1%) died within 14 days of their ED visit. The area under the curve (AUC) for the 14-day mortality receiver operating characteristic (ROC) curve was 0.484 for brain natriuretic peptide (BNP), 0.586 for N-terminal pro-B-type natriuretic peptide (NT-proBNP), 0.755 for troponin (I or T), 0.742 for adrenomedullin (MR-proADM), and 0.803 for copeptin. In combination, MR-proADM and copeptin had the best 14-day mortality prediction (AUC = 0.818), versus all other markers.
Conclusions: MR-proADM and copeptin, alone or in combination, may provide superior short-term mortality prediction compared to natriuretic peptides and troponin. Presented results are explorative due to the limited number of events, but validation in larger trials seems promising.
In heart failure (HF) patients presenting with acute dyspnea, short-term risk assessment is important to identify candidates for aggressive therapy and to aid in accurate disposition decision-making. A recent study of more than 50,000 HF patients found a 4% 30-day mortality rate in those discharged from the emergency department (ED), nearly matching the 5.7% rate if hospitalized.1 Better HF risk stratification tools are clearly needed. Even when hospitalized, delayed therapy from failing to identify high-risk patients is associated with increased mortality, longer hospitalizations, and more invasive procedures.2–5
Beyond troponin,6 no biomarkers have proven short-term prognostic utility in acute HF. Natriuretic peptide testing, combined with clinical impression, is currently used to assess acute prognosis.2 Recommended as a prognostic adjunct by current acute HF guidelines,7,8 most natriuretic peptides studies report 30- and 90-day outcomes. This time horizon is less helpful in the acute setting. Decisions regarding the necessity of immediate hospitalization or intensive care unit (ICU) placement cannot be based on mortality risk 3 months into the future. Of more value for emergency physician (EP) decision-making would be determining 7- or 14-day risk, such that an intervention could be performed to potentially alter future adverse outcomes rates. Unfortunately, there are very limited data on short-term outcomes. We selected candidate markers that are known to predict longer-term outcomes and are associated with disease severity and hypothesized that the best markers for predicting short-term risk would be those reflecting hemodynamic status.
Adrenomedullin (ADM), a hemodynamically active vasodilatory peptide with potent hypotensive effects, may be a short-term outcome predictor. Expressed in many tissues,9 plasma levels are elevated in chronic HF10 and increase with disease severity.11,12 Recently available immunoassay platforms detecting the midregion of the biologically inactive peptide by-product of ADM synthesis (MR-proADM) may reflect clinical events, as the plasma concentration of MR-proADM is stoichiometrically related to active ADM.13,14
Another candidate marker for short-term outcome prediction in acute heart failure (AHF) is arginine vasopressin (AVP). In the setting of intravascular volume deficit, AVP prompts the retention of water and increases systemic vascular resistance by vasoconstriction. In chronic HF, elevated levels of AVP are associated with increased long-term mortality. Like ADM, antibodies directed at a portion of AVP’s precursor peptide, c-terminal proAVP (termed copeptin), provide assay results stoichiometrically related to the level of biologically active AVP.15,16
Because of the potential clinical need and value in identifying short-term mortality risk, our purpose was to determine which of several candidate markers so far not reported provide the most accurate short-term mortality prediction in patients presenting with AHF.
The BACH (Biomarkers in the Assessment of Congestive HF) trial was a prospective, 15-center international, convenience sample study. This study was approved by the institutional review boards of all institutions.
Study Setting and Population
The study population consisted of 1,641 patients presenting to an ED with acute dyspnea from March 2007 through February 2008. Eligible patients had a chief complaint of shortness of breath, but were excluded if they were younger than 18 years old, were unable to provide informed consent, had posttraumatic shortness of breath, had an acute ST-segment elevation myocardial infarction, or were receiving hemodialysis. Enrollment into BACH had to occur while the patient was in the ED, before the attending physician saw any of the local laboratory brain natriuretic peptide (BNP) or N-terminal pro-B-type natriuretic peptide (NT-proBNP) results.
Clinical findings were extracted from the chart by trained research personnel. The final ED diagnosis was determined by the site’s EP, based on all available information, but blinded to biomarkers except for the locally performed B-type natriuretic peptides and troponins. For this secondary analysis we evaluated only patients diagnosed with AHF by the on-site EP and not based on the final diagnosis as reported in the primary manuscript.17 The evaluation of these biomarkers for 7-day outcome prediction was a predefined secondary endpoint of the BACH trial; however, because of limited outcomes, the time course was increased to 14 days. Final diagnoses were determined 90 days after discharge by two cardiologists blinded to all biomarker results, except the locally performed natriuretic peptides or troponins. Blood samples were collected immediately after arrival in plastic EDTA tubes, processed by personnel blinded to patient data, and stored at −70°C. MR-proADM, copeptin, MR-proANP, BNP, and NT-proBNP were measured in batches at the core lab of the University of Maryland. Troponin I or T was measured using routine tests available at the participating centers. For quantitative comparisons, troponin values were quantile-transformed by assay type to allow comparability.
MR-proADM and MR-proANP were measured with an automated sandwich immunofluorescent assay using TRACE technology (time resolved amplified cryptate emission) (BRAHMS MR-proADM KRYPTOR and BRAHMS MR-proANP KRYPTOR, BRAHMS GmbH, Hennigsdorf/Berlin, Germany). Copeptin was measured with a sandwich immunoluminometric assay (BRAHMS CTproAVP LIA, BRAHMS GmbH, Hennigsdorf, Germany). The ADM assay is described elsewhere.13,14 Used in several previous studies,15,16 it provided a limit of quantitation of 0.23 nmol/L, within-run imprecision of 1.9%, and between-run imprecision of 9.8%. The copeptin assay, also described in previous studies,16,18–20 provides a limit of quantitation of 0.4 pmol/L, within-run imprecision of <5%, and total imprecision of <10%. In 200 healthy individuals, median copeptin concentration was 3.7 pmol/L, with a 97.5 percentile of 16.4 pmol/L.18 Performance of MR-proANP included a limit of quantitation of 4.5 pmol/L, within-run imprecision of 1.2%, and total imprecision of 5.4%.
Brain natriuretic peptide was measured by Triage immunoassay reagents (Biosite, San Diego, CA) formatted for Beckman Coulter instrumentation (Brea, CA). Laboratory performance included a limit of quantitation of 5.0 ng/L, within-run imprecision of 1.5%, and total imprecision of 3.0%. NT-proBNP was measured by the ElecSys 2010 analyzer (Roche Diagnostics, Indianapolis, IN). Laboratory performance included a limit of quantitation of 10.0 ng/L, within-run imprecision of 1.5%, and total imprecision of 3.0%.
Follow-up was conducted by telephone interviews performed 7 days after ED presentation in discharged patients, and after 30 and 90 days in all patients, regardless of admission status. Follow-up at 14 days was derived by calculating back from the data received in the 30-day telephone interview.
Values are expressed as means and standard deviations (SDs), medians and interquartile ranges (IQRs), or counts and percentages, as appropriate. All biomarkers (except quantile-transformed troponin) were nonnormally distributed and therefore log-transformed before analysis. To test whether MR-proADM or copeptin was superior to BNP or NT-proBNP for prognosis, log-transformed values of all biomarkers were evaluated in Cox proportional hazard regression models. To test for differences in the predictive value of MR-proADM beyond that of BNP and NT-proBNP, likelihood ratio chi-square test for nested models assessed whether MR-proADM adds predictive value to a clinical model with BNP or NT-proBNP and vice versa. Because of limited mortality at 14 days, we set multivariate Cox models (also used to compare the biomarkers to other clinical findings) for short term prediction to include two variables. We combined MR-proADM and copeptin with each clinical covariate and biomarker that was significant in univariate analysis, to demonstrate that the two new biomarkers were independent of and superior to already known clinical variables. The predictive value of each model was assessed by the model likelihood ratio chi-square statistic and C index. The C index was bootstrap corrected for models using multiple biomarkers. Area under time-dependent receiver operating characteristic (ROC) curves were estimated from censored survival data using the Kaplan-Meier method.21 Missing data were not replaced. All statistical analyses were performed using R version 2.5.1 (http://www.r-project.org) and SPSS version 16.0 (SPSS Inc., Chicago, IL).
BACH enrolled 1,641 ED patients with acute shortness of breath; the ED diagnoses was AHF in 466 (28.4%). While 411 (88.2%) patients had a final adjudicated criterion standard diagnosis of AHF, our secondary analysis used the 466 patients diagnosed in the ED with AHF. There were 157 patients diagnosed as AHF by the criterion standard adjudicated diagnosis, but non-AHF by the EP. When this occurred, the EP diagnosed pneumonia (29), ACS (24), arrhythmia (19), pulmonary embolism (1), and “other” (94); 10 patients had more than one diagnosis and were therefore counted twice. Of the 466 patients initially diagnosed in the ED with AHF, the majority (395, 84.8%) were hospitalized at the index visit. In this subgroup, 13 (2.8%), 19 (4.1%), 31 (6.7%), and 53 (11.4%) died within 7, 14, 30, and 90 days after presentation, respectively. One patient died within 14 days after initially being discharged from the ED.
The descriptive characteristics of the overall BACH patients are presented in detail elsewhere.17 The clinical findings in this HF subset of BACH were of limited assistance in prospectively identifying patients who would suffer short term mortality (see Table 1). Although patients who did not survive 14 days were older, with lower pulse oximetry, more likely to present with wheezing, and less likely to have a history of hyperlipidemia, there were no significant differences between survivors and the occurrence of short-term death in respect to all other aspects of the physical examination, recent history, or past medical history.
|Variables||All Patients (n = 466)||14-day Survival (n = 447)||14-day Mortality (n = 19)||p-value|
|Age (years)||70.8 ± 14||70.4 ± 14.1||79.1 ± 8.9||0.0065|
|Male sex||273 (58.6)||264 (59.1)||9 (47.4)||0.3469|
|White||361 (77.6)||343 (76.9)||18 (94.7)||0.1989|
|Black||92 (19.8)||91 (20.4)||1 (5.3)|
|Other||12 (2.6)||12 (2.7)||0 (0)|
|Smoking||95 (21.1)||89 (20.5)||6 (37.5)||0.1183|
|Wheezing||85 (19.5)||81 (19.4)||4 (23.5)||0.7539|
|Cough||223 (49.1)||216 (49.7)||7 (36.8)||0.3504|
|Weight gain||110 (27.4)||108 (27.7)||2 (18.2)||0.7344|
|Orthopnea||281 (62.9)||268 (62.5)||13 (72.2)||0.4649|
|Dyspnea at rest||241 (52.5)||230 (52.2)||11 (61.1)||0.4822|
|Heart rate (beats/min)||88 ± 25||88 ± 25||96 ± 30||0.1672|
|Systolic BP (mm Hg)||141 ± 31||142 ± 31||129 ± 35||0.0549|
|Diastolic BP (mm Hg)||83 ± 19||83 ± 18||80 ± 21||0.5308|
|BMI (kg/m2)||29.0 ± 8.2||29.1 ± 8.2||26.4 ± 6||0.2070|
|Temperature (°C)||36.6 ± 0.6||36.6 ± 0.6||36.8 ± 0.6||0.2829|
|Pulse oximetry (%)||94.7 ± 5.9||94.8 ± 5.7||90.9 ± 8||0.0253|
|Respiratory rate||21.9 ± 6||21.8 ± 6||24.2 ± 6.2||0.0751|
|Rales||256 (55.4)||243 (54.9)||13 (68.4)||0.3463|
|Wheezing||69 (15)||61 (13.9)||8 (42.1)||0.0034|
|S3||33 (7.5)||33 (7.8)||0 (0)||0.3836|
|Murmur||129 (28.6)||125 (28.9)||4 (22.2)||0.7904|
|Elevated JVP||170 (40)||160 (39.1)||10 (62.5)||0.0715|
|Edema||289 (62.8)||279 (63.3)||10 (52.6)||0.3447|
|Ascites||26 (5.8)||25 (5.8)||1 (5.9)||1.0000|
|Arrhythmia||191 (42.8)||180 (42.2)||11 (57.9)||0.2359|
|Asthma||34 (7.5)||34 (7.8)||0 (0)||0.3839|
|CRI||134 (30)||128 (29.8)||6 (35.3)||0.5989|
|HF||321 (69.8)||310 (70)||11 (64.7)||0.6019|
|CAD||221 (49.9)||210 (49.4)||11 (61.1)||0.3487|
|COPD||100 (22.2)||94 (21.7)||6 (33.3)||0.2510|
|DM||179 (38.7)||172 (38.7)||7 (38.9)||1.0000|
|Hyperlipidemia||206 (47.9)||204 (49.5)||2 (11.1)||0.0012|
|Hypertension||365 (79.9)||352 (80.2)||13 (72.2)||0.3785|
|MI||143 (32.2)||138 (32.4)||5 (27.8)||0.8006|
|Pneumonia||68 (15.7)||65 (15.7)||3 (17.6)||0.7388|
|Pulmonary embolism||29 (6.4)||27 (6.2)||2 (11.1)||0.3215|
|CABG||77 (16.8)||76 (17.3)||1 (5.6)||0.3322|
|Angioplasty/stent||89 (19.7)||85 (19.5)||4 (23.5)||0.7549|
|Stroke/CVA||56 (12.3)||52 (11.9)||4 (21.1)||0.2734|
|Pacemaker/ICD||91 (19.8)||88 (20)||3 (15.8)||1.0000|
|Prosthetic valve||23 (5.1)||22 (5)||1 (5.6)||0.6141|
|Echocardiography||248 (53.1)||239 (53.5)||9 (47.4)||0.6445|
|EF (%)||40 ± 17||40 ± 17||38 ± 13||0.5808|
|ECG normal or unchanged||83 (17.8)||79 (17.7)||4 (21.1)||0.7584|
|X-ray with edema (alveolar or interstitial)||131 (28.1)||122 (27.3)||9 (47.4)||0.0688|
|Oxygen||180 (38.6)||169 (37.8)||11 (57.9)||0.1552|
|IV vasodilators (nitroglycerin, nitroprusside, nesiritide)||128 (27.5)||121 (27.1)||7 (36.8)||0.6370|
|IV inotropes (dopamine, milrinone, norepinephrine)||6 (1.3)||6 (1.3)||0 (0)||1.0000|
|Furosemide||303 (65)||289 (64.7)||14 (73.7)||0.4732|
|Pulmonary artery catheter||12 (2.6)||12 (2.7)||0 (0)||1.0000|
|Admitted to ICU||37 (7.9)||32 (7.2)||5 (26.3)||0.0123|
|Admitted to hospital||395 (84.8)||377 (84.3)||18 (94.7)||0.3324|
While it would be expected that patients with greater severity of illness and higher risk of short-term death would receive more aggressive therapy, this is not what we found. With one exception (a 26% vs. 7.2% ICU admission rate), we note that ED treatment was similar between survivors and those who were dead within 14 days. The fact that nearly three-fourths of patients destined for early death were not even admitted to the ICU suggests an opportunity for more aggressive care. Further, of the 19 patients who died within 14 days of presentation, one was discharged from the hospital, and four were admitted to the observation unit, suggesting a disconnect between clinical presentation and subsequent adverse events.
In the ED-diagnosed acute HF cohort, valid measurements of MR-proADM and copeptin were obtained in 99.6% (two failures) and 98.7% (six failures), respectively. BNP, NT-proBNP, and MR-proANP results were obtained in 99.6% (two failures), 98.7% (six failures), and 99.6% (two failures), respectively. Troponin (T or I), measured on site, was available in 426 (91.4%) patients.
Median and interquartile biomarker ranges in patients diagnosed with acute HF are reported in Table 2. Spearman’s correlation coefficients between MR-proADM versus BNP and MR-proADM versus NT-proBNP were 0.42 and 0.56, respectively, and were 0.83 with BNP versus NT-proBNP. Spearman’s correlation coefficients were 0.64 and 0.27 for MR-proADM versus copeptin and versus troponins, respectively.
|Variables||N||All||14-day Survival||14-day Mortality||p-value|
|BUN (mg/dL)||426||24.5 (17.0–36.1)||24.0 (17.0–35.9)||32.8 (29.1–58.6)||0.004|
|Creatinine (mg/dL)||460||1.2 (1.0–1.6)||1.2 (1.0–1.6)||1.6 (1.1–2.1)||0.042|
|BNP (pg/mL)||464||764 (402–1415)||766 (401–1413)||694 (463–1476)||0.818|
|NT-proBNP (pg/mL)||460||5165 (2332–10096)||5114 (2296–9989)||8014 (3309–14897)||0.207|
|MR-proANP (pmol/L)||464||419 (282–619)||417 (282–614)||550 (371–865)||0.099|
|Copeptin (pmol/L)||459||26 (11–53)||24 (11–49)||135 (56–272)||<0.001|
|MR-proADM (nmol/L)||464||1.4 (1.0–2.1)||1.4 (1.0–2.0)||2.4 (1.6–4.7)||<0.001|
|I||185||0.02 (0.01–0.04)||0.02 (0.01–0.04)||0.06 (0.04–0.14)||<0.001|
|T||244||0.05 (0.03–0.09)||0.05 (0.03–0.09)||0.12 (0.07–0.30)|
Marker performance for outcome prediction was evaluated by Cox regression, chi-square, C index, and time-dependent area under the curve (AUC) and is reviewed in Table 3. Natriuretic peptides performed worse or only marginally better than chance for predicting 14-day mortality. By time-dependent AUC, the combination of MR-proADM and copeptin was superior to all other markers evaluated at all time points, although MR-proADM, copeptin, and troponin all performed well individually for short-term mortality prediction.
|Marker||14-day Follow up (n = 19 events)||90-day Follow-up (n = 53 events)|
|Cox regression||ROC analysis, AUC (95% CI)||Cox regression||ROC analysis, AUC (95% CI)|
|chi-square||p-value||C index||chi-square||p-value||C index|
|BNP (log10 pg/mL)||0.1||0.768||0.513||0.484 (0.438–0.531)||12.5||<0.001||0.636||0.650 (0.605–0.694)|
|NT-proBNP (log10 pg/mL)||1.8||0.179||0.586||0.586 (0.539–0.631)||25.6||<0.001||0.693||0.707 (0.663–0.748)|
|BUN (log10 mg/dL)||2.1||0.147||0.697||0.701 (0.655–0.744)||16.1||<0.001||0.664||0.674 (0.627–0.718)|
|MR-proANP (log10 pmol/L)||3.5||0.061||0.610||0.612 (0.566–0.656)||31.4||<0.001||0.703||0.718 (0.675–0.759)|
|Systolic blood pressure (mm Hg)||3.7||0.056||0.628||0.630 (0.584–0.674)||16.8||<0.001||0.657||0.666 (0.621–0.709)|
|Pulse oximetry (%)||4.5||0.033||0.646||0.651 (0.605–0.694)||9.9||0.002||0.585||0.590 (0.543–0.636)|
|Creatinine (log10 mg/dL)||5.5||0.019||0.635||0.638 (0.592–0.682)||20.6||<0.001||0.649||0.659 (0.614–0.703)|
|Age, years||8.1||0.004||0.680||0.684 (0.640–0.726)||5.8||0.016||0.611||0.613 (0.567–0.658)|
|Troponin (T or I, quantile-transformed)||12.3||<0.001||0.751||0.755 (0.711–0.795)||20.9||<0.001||0.706||0.714 (0.668–0.756)|
|MR-proADM (log10 nmol/L)||18.6||<0.001||0.738||0.742 (0.699–0.781)||48.2||<0.001||0.733||0.744 (0.702–0.783)|
|Copeptin (log10 pmol/L)||29.1||<0.001||0.798||0.803 (0.763–0.838)||32.4||<0.001||0.705||0.710 (0.666–0.751)|
|Copeptin and MR-proADM (both log10)||33.1||<0.001||0.808*||0.818 (0.779–0.852)||53.2||<0.001||0.743*||0.760 (0.718–0.798)|
Due to the limited number of events within the first 7 days (n = 13), multivariable evaluations were performed for 14 and 90 days only. For 14- and 90-day mortality prediction, the combination of MR-proADM and copeptin was superior to each single marker alone (all p < 0.05). For 14-day mortality prediction, the contribution of copeptin was larger than that of MR-proADM, although by 90-day follow up MR-proADM became superior.
Both MR-proADM and copeptin individually added significant value to all other biomarkers from Table 2, including troponin (all p < 0.01), for predicting 14-day mortality. Outcome prediction was significantly improved by adding either MR-proADM or copeptin to a Cox model with either age (added chi-square = 15.8, p < 0.0001; and chi-square = 24.2, p < 0.0001, respectively), systolic blood pressure (added chi-square = 15.6, p < 0.0001; and chi-square = 30.6, p < 0.0001, respectively), blood urea nitrogen (BUN; added chi-square = 12.2, p < 0.0005; and chi-square = 23.6, p < 0.0001, respectively), pulse oximetry (added chi-square = 15.2, p < 0.0001; and chi-square = 26.1, p < 0.0001, respectively), or creatinine (added chi-square = 12.6, p = 0.0004; and chi-square = 23.2, p < 0.0001, respectively). Therefore, both biomarkers provide independent information for short-term prediction to the known clinical findings on admission and in addition to other commonly used biomarkers.
Figure 1 demonstrates the temporal relation of the time-dependent AUC, reflecting biomarker prognostic accuracy. The combination of MR-proADM and copeptin provided the best accuracy for short- and long-term mortality prediction. However, while this combination is good for 14-day outcomes, it deteriorates over the subsequent 90 days. In opposite fashion, natriuretic peptides, with poor initial mortality prediction, improve with time, but are never superior to the combination of MR-proADM and copeptin or MR-proADM alone.
MR-proADM, copeptin, BNP, and NT-proBNP levels, stratified by time of death, are shown in Figure 2. For all markers, values are lowest in survivors. MR-proADM and copeptin values are greatest for deaths within 14 days (significantly higher than survivors, post hoc p < 0.01), while for BNP and NT-proBNP values are greatest for later deaths (death after Day 14, values significantly larger than for survivors, post hoc p < 0.01). Figure 3 shows time of death as related to quartiles of MR-proADM, copeptin, BNP, and NT-proBNP. Because neither BNP nor NT-proBNP provided added value on top of MR-proADM at 90-day follow-up,17 we had no reason to believe that adding MR-proANP (or BNP/NT-proBNP) would improve short-term prediction and thus do not present these data.
For all biomarkers presented, patients above the third quartile are at considerably higher risk than those below. Based on this cutoff (third quartile cutoff at 2.1 nmol/L), the sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio for MR-proADM predicting 14-day mortality are 63.2%, 76.9%, 2.7, and 0.5, respectively, and are 79.0%, 77.3%, 3.5, and 0.3, respectively, for copeptin (third quartile cutoff at 52.1 pmol/L). In comparison, the best natriuretic peptide, MR-proANP, achieved 36.8%, 75.5%, 1.5, and 0.8, respectively (third quartile cutoff at 618 pmol/L). For both MR-proADM and copeptin ROC, optimized cutoffs were 2.15 nmol/L and 54.2 pmol/L, respectively, and gave slightly higher specificities (78.9 and 79.6%), while leaving sensitivity unchanged.
Table 4 shows individual treatment and biomarker profiles for the 19 patients who died within 14 days after presentation. Only five (26.3%) were admitted to the ICU. Of the 14 not sent to the ICU, 11 (78.6%) had MR-proADM and/or copeptin concentrations above the third quartile, while only four (6%) were high in BNP (NT-proBNP), again defined as concentrations larger than the third quartile. While the one patient discharged (Patient 16) had high copeptin and low MR-proADM, another patient (admitted to the ICU, Patient 10) had high MR-proADM and low copeptin. Troponin correlated in most cases with BNP/NT-proBNP, except for one patient (Patient 14), who showed nonelevated values for all other biomarkers.
|Variables||Patients Not Surviving 14 Days After Admission|
|ED treatment data|
|Systolic BP (mm Hg)||102||88||110||114||117||151||113||160||157||117||93||204||130||190||90||90||155||160||104|
|Pulse oximetry (%)||92||85||96||99||97||95||96||100||70||100||88||81||98||95||93||88||84||80||91|
|Central laboratory data|
|Local laboratory data|
|Troponin I (ng/mL)||0.68||—||—||—||—||—||—||—||—||—||0.12||—||0.06||0.43||—||0.08||0.02||0.17||—|
|Troponin T (ng/mL)||—||0.27||0.05||0.14||0.06||0.12||—||0.04||2.99||0.02||0.04||0.01||—||—||—||—||—||—||0.15|
|Days until death||1||7||6||9||0||8||13||2||2||6||12||11||0||2||1||4||6||9||5|
|Readmission for HF||No||No||No||No||No||No||No||No||No||No||No||No||No||No||No||Yes||No||No||No|
We demonstrate that MR-proADM, copeptin, and their combination provide superior short-term mortality prediction (defined as death within 14 days of presentation) not evident on clinical exam, and compared to troponin and natriuretic peptides, in patients diagnosed with acute HF in the ED. Our results, as illustrated in Figure 2, suggest that these new markers provide unique and additive information versus natriuretic peptides and troponin. That these markers provide information that is not clinically evident is supported by the fact that the majority of patients identified at high risk of short-term mortality by copeptin and ADM did not receive care that was significantly different than survivors.
We do note that patients more likely to suffer a 14-day mortality tended to be older, with lower pulse oximetry measures and higher rates of wheezing. However, when these findings of increased mortality risk were present, they were not associated with unique therapeutic interventions by the EP. Most importantly, the accuracy for a lower pulse oximetry measure in predicting short-term mortality is significantly inferior compared to either ADM or copeptin.
While this was not an interventional trial, and the consequence of more aggressive care for the high-risk cohort is unknown, our data demonstrate that these high-risk patients did not receive maximum treatment (only one-fourth were admitted to the ICU), suggesting room for improvement if they had been identified as high-risk patients. It is important to note that while elevated copeptin or ADM is associated with short-term mortality risk, a low level does not allow for immediate discharge. Mortality is not the only outcome of importance, and the patient’s presenting symptoms must be considered. AHF treatment should not depend solely on copeptin or MR-proADM, as these are hemodynamic markers that are not specific for HF. However, a high concentration of either of these markers appears to be an indication that something is considerably wrong with the patient. Ultimately, these HF markers may be similar to troponin, where a marker predicting short-term death supersedes the clinical impression.
We also report that testing with natriuretic peptides does not identify patients at risk for short-term mortality. Although of significant clinical relevance, short-term mortality prediction is rarely reported as a biomarker endpoint, as the norm is to report 30- and 90-day mortality. The value of 14-day mortality prediction is its potential for identifying patients in whom immediate intervention or ICU admission would be reasonable. Accurate clinical decision-making in the acute environment requires knowledge of potential outcomes in a much shorter time frame than 90 days. By identifying patients at short-term mortality risk, hospitalization and aggressive intervention may decrease the high 30-day outpatient mortality rate reported to occur when discharged after an ED visit.1 With the exception of troponin,6 no commonly available prognostic HF marker accurately predicts short-term mortality.
Identification of short-term outcomes has important implications for HF research. Most therapeutic HF trials completed in the past decade have been unable to show mortality reduction as a primary endpoint. This may be the consequence of poor early risk stratification. An inability to identify a high-risk cohort from the generalized HF population results in the requirement for prohibitively large trials and makes it difficult to determine the effect of any potential life-preserving therapy. If patients at high short-term mortality risk are prospectively selected, adequate power for interventional studies may be obtainable with much smaller trials. Furthermore, as the risk–benefit relationship of interventions changes when patients are identified as being high risk for death, invasive options become a more reasonable course (e.g., left ventricular assist device placement).
We found that the ability of biomarkers to predict outcomes is not static, but rather changes over time. In BACH, biomarkers with excellent initial prognostic ability (MR-proADM and copeptin) became slightly less accurate. Conversely, natriuretic peptides, initially poor acute prognostic markers, improve their performance during follow-up. We suggest that the “regression to the mean” of prognostic value is the result of the biologic characteristics of the tested markers. It is not surprising that markers reflecting hemodynamic status or myocardial necrosis have better short-term prognostic value (e.g., MR-proADM, copeptin, and troponin) and that this prognostic value would deteriorate with resolution of the acute insult. Analogous to trauma, if the patient initially survives, his or her risk of death decreases over time.
A very limited number of parameters (only presenting blood pressure and renal function) have consistently demonstrated acute mortality prognostic ability.22 Conversely, in our analysis natriuretic peptides demonstrated minimal ability for predicting short-term outcomes. While natriuretic peptides may reflect chronic myocardial stress, their relationship does not tightly track with invasively measured hemodynamics (i.e., pulmonary artery occlusion pressure). This is supported by the fact that in acute HF, natriuretic peptide levels have only a weak relationship to pulmonary artery catheter measures of hemodynamics.23
Other limitations in our study may have resulted in the natriuretic peptide’s poor prediction of short-term outcomes. BACH was an “all comers” dyspnea trial, allowing many “real world” pathologies to be enrolled, despite the fact that this secondary analysis focused on patients diagnosed with acute HF. By including patients with obesity and renal insufficiency, conditions known to confound natriuretic peptide interpretation, we may have negatively affected their performance. However, in the management of HF, clinical confounders are common. Nearly 50% of Americans are obese, and one-third of HF patients have significant renal dysfunction. To eliminate these confounders would exclude the very patients for whom short-term mortality risk prediction may be critical. Ultimately this suggests that the BACH data are robust and will perform well in the clinical arena where confounders are common and frequently incompletely defined.
It should be noted that MR-proADM, copeptin, and the natriuretic peptides predominantly identified patients at risk for cardiovascular mortality. This is clinically relevant, as noncardiovascular deaths occurred in patients with lower marker concentrations, suggesting that the absence of elevation does not equate to a low risk of short-term all-cause mortality. Ultimately, the value of these markers is the prognostic information they provide when elevated, not when they are found in the normal range.
In our analysis, previously described mortality predictors were not helpful. For example, an ADHERE analysis found elevated BUN and creatinine, and low initial blood pressure, were associated with an acute mortality rate exceeding 20%.22 In BACH, we found these to be less valuable. Troponin showed prognostic power, but it was not part of a primary or secondary endpoint of the trial, of which the main focus was the comparison to natriuretic peptides. Therefore, troponin values are only available if measured at the local laboratory, and the results on troponin may be underestimated due to the fact that we did not have a standardized measurement available. Although we cannot comment on the interaction between presenting blood pressure and interventions (as physicians may have implemented therapy in response to abnormal vital signs), in our analysis the rate of chronic renal failure was nearly identical in patients surviving or dying within 14 days. Furthermore, our 14-day mortality rate of 4.1% is nearly twice that of the low-risk group in this ADHERE analysis.22
Finally, the effect of therapy on patients identified as high risk for short-term mortality should be considered. With the exception of locally available troponin and natriuretic peptide tests, the treating physicians were blinded to the BACH biomarkers. Thus, any value of therapeutic intervention in response to the tested biomarkers cannot not be ascertained. Another limitation is that troponin was measured locally, using different assays standardized by quantile transformation. Although reflective of real world practice, comparison to other studies may be complicated and represents a limitation in the troponin portion of this analysis. Also, we may have underestimated the predictive value of troponin via a classification bias, where patients with initially elevated troponins were excluded by receiving a criterion standard diagnosis of non-ST-segment elevation myocardial infarction, rather than acute HF. Last, the number of events within 14 days after presentation to the ED is fortunately small (n = 19), which limits the ability to evaluate the biomarkers in multivariable models; thus, the models may be overfit for the data presented, resulting in a potential decrease in external validity.
Determination of ED-diagnosed heart failure patients at high risk of short-term mortality identifies a cohort for whom hospitalization and aggressive therapy would be reasonable. Both MR-proADM and copeptin, and their combination, demonstrate superior 14-day mortality prediction in our population, compared to other markers, including troponin and natriuretic peptides. However, it should be noted that our results are explorative due to the limited number of events and the need for validation in larger trials.