Presented at the Society for Academic Emergency Medicine annual meeting, Boston, MA, June 2011.
The authors acknowledge peer-reviewed funding from the Canadian Institutes of Health Research (MOP 82742). Dr. Stiell holds a Distinguished Professorship and University Health Research Chair from the University of Ottawa. Dr. Rowe is supported by the 21st Century Canada Research Chairs program through the Government of Canada (Ottawa, ON). Dr. Perry holds a New Investigator Award from the Canadian Institutes of Health Research and a Clinical Research Chair from the University of Ottawa. Dr. Aaron holds a Clinical Research Chair from the University of Ottawa. The authors have no further disclosures or conflicts of interest to report.
There are no validated guidelines to guide physicians with difficult disposition decisions for emergency department (ED) patients with heart failure (HF). The authors sought to develop a risk scoring system to identify HF patients at high risk for serious adverse events (SAEs).
This was a prospective cohort study at six large Canadian EDS that enrolled adult patients who presented with acute decompensated HF. Each patient was assessed for standardized clinical and laboratory variables as well as for SAEs defined as death, intubation, admission to a monitored unit, or relapse requiring admission. Adjusted odds ratios for predictors of SAEs were calculated by stepwise logistic regression.
In 559 visits, 38.1% resulted in patient admission. Of 65 (11.6%) SAE cases, 31 (47.7%) occurred in patients not initially admitted. The multivariate model and resultant Ottawa Heart Failure Risk Scale consists of 10 elements, and the risk of SAEs varied from 2.8% to 89.0%, with good calibration between observed and expected probabilities. Internal validation showed the risk scores to be very accurate across 1,000 replications using the bootstrap method. A threshold of 1, 2, or 3 total scores for admission would be associated with sensitivities of 95.2, 80.6, or 64.5%, respectively, all better than current practice.
Many HF patients are discharged home from the ED and then suffer SAEs or death. The authors have developed an accurate risk scoring system that could ultimately be used to stratify the risk of poor outcomes and to enable rational and safe disposition decisions.
Una Escala De Puntuación De Riesgo Para Identificar En El Servicio De Urgencias A Los Pacientes Con Insuficiencia Cardiaca Con Alto Riesgo De Presentar Eventos Adversos Graves
No existen guías clínicas validadas para orientar a los médicos en la dificíl decisión de la ubicación de los pacientes con insuficiencia cardiaca (IC) en el servicio de urgencias (SU). El objetivo fue desarrollar una escala de puntuación de riesgo para identificar a los pacientes con IC con alto riesgo de eventos adversos graves (EAG).
Estudio de cohorte prospectivo que incluyó pacientes adultos que presentaron IC aguda descompensada en seis SU canadienses. Se recogieron variables clínicas y de laboratorio estandarizadas en cada paciente así como de EAG, definidos como muerte, intubación, ingreso en una unidad monitorizada o recaída con necesidad de ingreso. Se calcularon las OR ajustadas para predecir EAG mediante una regresión logística de pasos hacia atrás.
De 559 consultas, el 38,1% finalizaron con el ingreso del paciente. De los 65 (11,6%) casos de EAG, 31 (47,7%) ocurrieron en los pacientes no ingresados inicialmente. El modelo multivariable y la escala Ottawa Heart Failure Risk Scale resultante constan de 10 elementos, y el riesgo de EAG varió de 2,8 a 89%, con buena correlación entre lo observado y las probabilidades esperadas. La validación interna mostró que las puntuaciones de riesgo eran muy certeras en las 1.000 réplicas usando el método bootstrap. Un umbral de 1, 2 ó 3 de puntuación total para ingreso estaría asociado con sensibilidades para detectar EAG del 95,2%, 80,6% y 64,5%, respectivamente, cualquiera de ellas mejor que la práctica actual.
Muchos de los pacientes con IC son dados de alta a casa desde el SU y después sufren EAG o muerte. Se ha desarrollado un sistema de puntuación de riesgo certero que podría ser usado para estratificar el riesgo de malos resultados y permitir decisiones de ubicación racionales y seguras.
Acute decompensated heart failure (HF) is a common and serious condition that is the leading cause for hospital admissions for seniors in the United States and Canada.[1-5] An estimated one million HF patients seek emergency department (ED) care annually in the United States. HF often coexists with other chronic diseases (such as diabetes mellitus, chronic obstructive pulmonary disease [COPD], and coronary artery disease), which further contribute to hospitalization and prolonged lengths of stay. Patients who are admitted to hospital for HF have high readmission rates in the months following initial discharge. Moreover, patients discharged from the ED following treatment for acute exacerbation of HF have high rates of relapse and mortality.[8, 9] High-quality evidence to assist clinicians in ED decision-making is lacking, and only recently have clinical guidelines focused on the management of acute HF.[10, 11]
An important challenge facing emergency physicians when treating HF patients is deciding on disposition, whether to admit with or without monitoring, or whether to discharge, with or without early follow-up. It is not desirable to admit all patients, as many will respond to therapy in the ED and will not benefit from hospitalization. A small but important number of these patients have serious adverse events (SAEs), i.e., they die, require intensive care therapy, or suffer morbidity such as myocardial infarction (MI). Many other patients are discharged after prolonged ED management only to suffer a SAE or return to receive additional ED treatment and/or be admitted. The issue of disposition decisions is an important question because in many jurisdictions there is a shortage of hospital beds and many EDs are overcrowded. There are, however, no widely accepted or validated guidelines to aid with these difficult decisions.
The overall goal of this study was to develop a risk scoring system to guide the disposition decisions of physicians for ED patients with acute dyspnea secondary to HF. In particular, this risk scale is intended to be highly sensitive for predicting the potential for development of SAEs among such patients. Ultimately, this scale should improve and standardize disposition practices for these patients, diminishing both unnecessary admissions and unsafe discharge decisions.
We conducted a prospective observational cohort study. The study protocol was approved by the research ethics boards at each center; the boards at three hospitals determined that written informed consent was required, whereas those at the other three sites waived the need for written consent for this observational study.
Study Setting and Population
The study was conducted in six Canadian teaching hospital sites in Ottawa, Toronto, Kingston, Montreal, and Edmonton, with a combined annual ED volume of approximately 350,000 patient visits. We included a convenience sample of adults ≥ 50 years of age who presented with acute shortness of breath secondary to exacerbations of chronic HF or new-onset HF. We used pragmatic criteria for the diagnosis of HF as recommended by the working group on HF of the European Society of Cardiology.[12-14] Patients must have had appropriate symptoms (shortness of breath or fatigue) with clinical signs of fluid retention (pulmonary or peripheral) in the presence of an underlying abnormality of cardiac structure or function. If doubt remained, a beneficial response to treatment (for example, a brisk diuresis accompanied by substantial improvement in breathlessness) was also considered. While B-type natriuretic peptide (BNP) levels are not always used in these Canadian EDs for patients with dyspnea, when available, we reviewed BNP values to confirm values compatible with a diagnosis of HF.
We excluded patients who were obviously too ill to be considered for discharge within the 2- to 15-hour ED treatment study window or who were otherwise unsuitable for the study because of: 1) resting oxygen saturation < 85% on room air or after being on their usual home oxygen setting for 20 minutes on ED arrival; 2) heart rate greater than or equal to 120 beats/min on arrival; 3) systolic blood pressure < 85 mm Hg on arrival; 4) confusion, disorientation, or dementia; 5) ischemic chest pain requiring treatment with nitrates on arrival; 6) acute ST-segment elevation on electrocardiogram (ECG) on arrival; 7) terminal status—death expected within weeks from chronic illness; 8) from nursing home or chronic care facility; 9) enrolled into the study in previous 2 months; or 10) on chronic hemodialysis.
Patient assessments were made by registered respiratory therapists or registered nurses who were on duty at variable times depending on the site. The target assessment period was 4 to 8 hours after initial ED treatment, but patients could be considered for enrollment at 2 to 15 hours after treatment. The research assistants were trained by means of lectures and practical demonstrations to assess all variables in a uniform manner. A standardized description of each assessment was provided and the research assistants recorded their findings on data collection sheets. There was ongoing evaluation of the quality of the patient assessments by a central study nurse coordinator who provided regular feedback to the sites.
The variables assessed in the study were chosen by the investigators based on their clinical experience and reports in the literature.[15-26] As shown in Tables 1-3, we collected data from history, general examination, laboratory tests, and the 3-minute walk test. For the 3-minute walk test, patients were asked to walk at their own pace in the ED for a fixed period of 3 minutes, regardless of the distance covered. Patients could use their normal walking aids (e.g., cane or walker), but could not be physically supported by another person. Patients used no supplementary oxygen or their normal home oxygen flow level. The same model of recording pulse oximeter (Criticare 504DXP, Criticare Systems, Inc., Waukesha, WI) was used at all sites to record and continuously measure heart rate and oxygen saturation levels. Not all patients were well enough to undertake the walk test and some were unable to complete the entire 3 minutes. BNP samples were drawn from enrolled patients and then processed in batch by The Ottawa Hospital (NT-ProBNP with Roche Elecsys 2010 system, Hoffmann-La Roche, Ltd., Basel, Switzerland). At two sites, we used the N-terminal prohormone-BNP (NT-ProBNP) values provided by the hospital laboratories for clinical use, also analyzed by the Roche system.
Table 1. Patient Characteristics for 559 HF Patient Visits
N = 559
Canadian Triage Acuity Scale (CTAS) ranges from 1 (most urgent) to 5 (least urgent).
Duration of respiratory distress (hours), mean (±SD)
Home oxygen (%)
Smoker (pack-years), mean (±SD) (n = 20 and 187)
Current cardiac medications (%)
Calcium channel blockers
Current respiratory medications (%)
Caretaker at home (%) (n = 304)
Able to drink fluids in ED (%) (n = 419)
The primary outcome was SAE: death from any cause within 30 days of the ED visit, or any of the following within 14 days of the index ED visit, regardless of whether initially admitted: 1) Admission to a critical care or acute monitoring unit where the patient is too ill to ambulate; this excludes ambulatory telemetry units. 2) Endotracheal intubation or need for noninvasive ventilation after hospital admission, unless on noninvasive ventilation at home. 3) MI, as defined by international consensus standards. Either one of the following criteria satisfies the diagnosis for an acute, evolving, or recent MI: i) Typical rise and gradual fall of troponin with at least one of the following: a) ischemic symptoms, b) development of pathologic Q waves on the ECG, c) ECG changes indicative of ischemia, or d) coronary artery intervention (e.g., coronary angioplasty). ii) Pathologic findings of an acute MI. 4) Major procedure defined as coronary artery bypass graft, percutaneous coronary intervention, other cardiac surgery, or new hemodialysis. 5) Relapse and hospital admission for patients who were discharged on the initial ED visit, defined as a return to the ED for any related medical problem within 14 days followed by admission to hospital; relapse to the ED without associated admission was not considered a SAE. Of note, Canadian EDs generally make disposition decisions within 12 hours of arrival and do not use observation units.
Assessment of the primary outcome measure was made by the investigators, blinded to the patient status for the predictor variables, from these source documents: 1) ED health records, 2) hospital health records, 3) computerized hospital patient tracking and record system, and 4) review of provincial death records. Patients were not contacted by telephone.
The association between the primary outcome, SAE, and variables from the history, physical examination, and investigations was assessed by the appropriate univariate analyses. Continuous variables were also categorized using the most discriminative cut points. We imputed the values for NT-proBNP for 45% of cases using a multiple imputation procedure and assuming that missingness was random.[29-31] NT-proBNP was frequently missing because it was only gathered for research purposes at four of six sites, not as part of routine laboratory investigations, and required additional written patient consent. Logistic regression with stepwise selection was conducted for those variables found to be associated with SAE on univariate analysis (p < 0.05), as well as for clinically sensible interaction terms. Analyses were conducted by visit rather than individual patient. Using accepted approaches, a scale was created by rounding up the lowest likelihood ratio (LR) beta coefficient to one and then multiplying the other coefficients by the same factor and then rounding to the nearest whole number. The classification performance of the final score categories was internally validated for sensitivity, specificity, positive LR, and negative LR across 1,000 replications using the bootstrap method. We estimated that we would need between 60 and 100 cases positive for the primary outcome to conduct stable and valid multivariate analyses.
Table 1 shows the demographic and clinical characteristics of the 559 eligible patient visits from September 2007 to April 2010. Patients at the six participating sites had a mean (±SD) age of 76.0 (±10.6) years and an overall admission rate of 38.1%. Another 835 eligible patient visits at the study sites were not enrolled, primarily because patients presented when research staff were not available. The characteristics of these patients are very similar to those of the patients enrolled (Data Supplement S1, available as supporting information in the online version of this paper). Overall there were 65 (11.6%) SAEs and, of concern, 31 (47.7%) of these occurred in the 346 patients not admitted on the initial ED visit (Table 2). Also, 4 of the 13 study patient deaths (30.8%) occurred within 30 days among patients initially discharged home from the ED.
Univariate and Multivariate Data Analyses
Tables 3 and 4 show the association between the primary outcome, SAE, and variables from the history, physical examination, and investigations. Some continuous variables were further categorized using the most discriminative cut points (Table 5). Overall, 22 of these univariate associations were statistically significant.
Table 4. Univariate Correlation with SAEs for Variables from Physical Examination and Investigations for 559 HF Patient Visits
Walk test 1 minute postwalk highest HR ≥ 110 (n = 43 and 432)
Walk test 1 minute postwalk SaO2 <90% (n = 44 and 432)
Table 6 shows the multivariate logistic regression model that determined independent predictors for SAEs. This model was developed on a data set of 507 cases without missing values. This model has a nonsignificant Hosmer-Lemeshow chi-square goodness-of-fit statistic (p = 0.75) and an area under the receiver operating characteristic (ROC) curve of 0.77 (95% confidence interval [CI] = 0.71 to 0.84). Variables not significant in the final model included age, disposition status (admitted or discharged), and several interaction terms. We conducted a sensitivity analysis by performing multivariate analysis without NT-proBNP and found that the other nine variables remained in the model with very similar coefficients but a lower area under the ROC curve (0.75) (Data Supplement S2, available as supporting information in the online version of this paper).
Table 6. Independent Predictors of SAEs as Determined by Stepwise Logistic Regression Analysis for HF Patientsa
Hosmer-Lemeshow Goodness-of-fit p-value = 0.753.
Area under ROC curve = 0.774; 95% CI = 0.706 to 0.843.
Heart rate ≥ 110 beats/min during 3-minute walk test
Ottawa Heart Failure Risk Scale
The Ottawa Heart Failure Risk Scale (Figure 1) consists of 10 elements from history, examination, or investigations and has a maximum score of 15. We found that the risk of SAE varied from 2.8%, for a score of 0, to 89.0%, for a score of 9. Figure 2 shows good calibration between observed and expected probability of SAE up to a score of 6, beyond which there is variability due to small numbers (Hosmer-Lemeshow goodness-of-fit p = 0.950).
Table 7 shows the classification performance and expected admission proportions for the Ottawa Heart Failure Risk Scale compared to current practice at the six study hospitals. The scale could improve upon the sensitivity of current practice, which only sees 52.0% of patients with SAE being admitted on the first ED visit. For example, choosing total point score thresholds of 1, 2, or 3 as a decision for admission would be associated with sensitivities of 95.2, 80.6, or 64.5%, respectively. These theoretical admission threshold levels would result in admission rates of 80.9, 50.5, or 27.4%, respectively, compared to the current admission rate of 38.1% at the study sites.
Table 7. Classification Performance and Expected Admission Proportions for the Ottawa Heart Failure Risk Scale Compared to Current Practice at the Six Study Hospital Sites
The classification performance of the final score categories was validated using the sensitivity, specificity, positive LR, and negative LR. This internal validation showed the risk scores to be very accurate across 1,000 replications using the bootstrap method (Data Supplement S3, available as supporting information in the online version of this paper).
Interpretation of Results
We found a relatively high frequency of SAEs among patients who present to the ED with HF, a relatively low hospital admission rate, and a disturbingly high proportion of poor outcomes among patients discharged home from the ED. We were able to establish the strong association of some 22 clinical and laboratory predictors with the development of SAEs, and multivariate analyses resulted in a concise and accurate model of independent risk factors for SAEs. We then developed the 10-element Ottawa Heart Failure Risk Scale, which provides an estimate of the risk of poor outcomes. All criteria are sensible and some are unique, such as the 3-minute walk test and quantitative assessment of NT-proBNP. Physicians at sites that do not have immediate NT-proBNP values available can use the scale without this item. The scale does not require sophisticated imaging or expensive testing to assist decision-making. Use of this scale could assist in the identification of patients most at risk for adverse outcomes and who are most in need of admission or early follow-up. We expect that this risk scale, once fully validated, could be widely used and will improve both hospital admission practices and the safety of ED management decisions.
In the past, virtually all patients with acute HF were hospitalized whereas today in Canada, many of these patients are seen in the ED and subsequently discharged home. There is a lack of high-quality evidence, and the few studies to develop admission guidelines for HF have important limitations, particularly the use of large administrative databases rather than prospective and specific ED data collection.[15, 16, 35, 36] These studies did not evaluate patients after ED management and did not incorporate a functional measure, such as a walk test. None of these guidelines have been prospectively validated or implemented into practice. Lee et al. developed a scoring model to predict mortality for HF patients already admitted to Canadian hospitals, but did not include patients discharged from the ED or their response to therapy in the ED. Recent studies have evaluated the value of troponin, ECG QRS duration, and precipitating factors in predicting outcomes of hospitalized HF patients.[37-39]
The serum marker BNP is being used more often for the diagnosis of patients with HF.[40-42] BNP is a cardiac neurohormone released by the ventricles, and higher levels are associated with increased mortality and morbidity for patients with HF.[43-45] Clinical trials have shown limited effect of routine BNP measurement on patient outcomes,[46-48] its value for improving care in the ED is unclear, and its use is not widespread in many countries.[50, 51] Our study found that a very high NT-proBNP level is an independent predictor of SAE for HF patients in the ED.
We are concerned by the high rate of SAEs among HF patients discharged from the ED. The Ottawa Heart Failure Risk Scale has the potential to significantly improve patient safety by helping to ensure that physicians admit or arrange early follow-up for patients most at risk for poor outcomes. While Canadian hospitals would struggle with admitting 80% of HF patients as is the case for U.S. hospitals, we believe that even a modest increase in admission (e.g., to 50% from the current 38%) would very likely lead to safer management practices. More important than increasing the admission rate is ensuring admission of the correct patients, i.e., those at highest risk of a poor outcome. Alternately, the scale could be used to identify at-risk patients who should have guaranteed early follow-up if they are not admitted, perhaps in specialized HF clinics.
We believe that the study has significant strengths, including multicenter and rigorous prospective collection of real-time clinical data, as well as the unique use of the 3-minute walk test and quantitative NT-proBNP values. One issue is the inclusion of both admitted and discharged patients, which we believe is the correct methodologic approach. Admission status may confound the likelihood of an SAE occurring, in that some admitted patients will not suffer SAEs because they receive more intensive therapy and that the same patients might have suffered SAEs if they had been discharged home. Our objectives are to develop a decision tool that ensures appropriate admission of high risk patients (i.e., sensitivity) as well as minimizing admission of low risk patients (i.e., specificity). We can only do this by evaluating admitted patients. We also note that admitted status was not significantly associated with SAE in the multivariate model.
We believe that use of the “monitored unit” (not an ambulatory telemetry unit) as a criterion for SAE is an important outcome that almost always reflects severity of illness. Such patients would likely suffer significant morbidity if they had been discharged, as they were bed-ridden and required constant vital sign monitoring. This new risk scale should be prospectively validated in a new set of ED patients with HF. In particular, future studies should attempt to obtain quantitative NT-proBNP values on all patients. We suggest that the scale can be used effectively with or without NT-proBNP values, depending on the availability of this test. We chose to impute missing values for NT-proBNP but not for other variables missing a small number of values. While we had relatively few outcomes relative to the number of predictors, internal validation with bootstrapping demonstrated that the estimate and classification performance for risk scores were stable. We chose not to adjust the analysis despite the fact that some patients were enrolled more than once. We were unable to enroll a large number of eligible patients because they presented outside of normal business hours. We could detect neither selection bias nor threat to the validity of our findings. Finally, it is possible, although unlikely, that some SAEs were not identified in patients who returned to different hospitals. If such events did occur, we do not believe that this would have had a significant effect on the identification of variables strongly associated with the primary outcome.
Many heart failure patients are discharged home from the ED and then suffer severe adverse events or death. We have developed an accurate and easy to use Ottawa Heart Failure Risk Scale that can be used to stratify the risk of poor outcomes for ED patients with heart failure and to enable rational and safe disposition decisions. With appropriate validation, this scale could ultimately benefit both patients and health care systems by ensuring appropriate admissions, targeting those who need early follow-up, and diminishing unnecessary hospitalizations.
The authors thank the invaluable assistance of the following individuals from the study sites: Jewish General Hospital (Montreal, Quebec)—Antoinette Colacone, RN, Chris Tselios, BSc, Yanka Mihaylova; Kingston General Hospital (Kingston, Ontario)—Katherine Bowes, RN, Jane Reid, RN; Mount Sinai Hospital (Toronto, Ontario)—Michelle Loftus, RN; Ottawa Hospital, Civic and General Campuses (Ottawa, Ontario)—Sara Dorrington, RRT, James Acheson, RRT, Michael Gatta, RRT, Michelle Saulnier, RRT, Krista Bangs, RRT, Christine Belisle, RRT, Vanessa Lawlor, RRT; and University of Alberta Hospital (Edmonton, Alberta)—Debbie Boyko, RN, Dion Pasichnyk. In addition, we thank the following individuals from the coordinating center in Ottawa, Ontario: Ottawa Hospital Research Institute—Tami Clavet, RN, Jennifer Brinkhurst, Connor Sheehan, Angela Marcantonio, Sheryl Domingo, My-Linh Tran; Department of Pathology and Laboratory Medicine, Ottawa Hospital—Carol Ann Jodouin; University of Ottawa Heart Institute—Li Chen. Finally we are grateful to the many physicians, nurses, and respiratory therapists at the study hospitals for their very helpful assistance with patient identification and data collection.