The n e w e ng l a n d j o u r na l of m e dic i n e

Original Article

Trial of an Intervention to Improve Acute

Heart Failure Outcomes
D.S. Lee, S.E. Straus, M.E. Farkouh, P.C. Austin, M. Taljaard, A. Chong, C. Fahim,
S. Poon, P. Cram, S. Smith, R.S. McKelvie, L. Porepa, M. Hartleib, P. Mitoff,
R.M. Iwanochko, A. MacDougall, S. Shadowitz, H. Abrams, E. Elbarasi, J. Fang,
J.A. Udell, M.J. Schull, S. Mak, and H.J. Ross, for the COACH Trial Investigators*​​

A BS T R AC T

BACKGROUND
The authors’ full names, academic de- Patients with acute heart failure are frequently or systematically hospitalized, often
grees, and affiliations are listed in the Ap- because the risk of adverse events is uncertain and the options for rapid follow-up
pendix. Dr. Lee can be contacted at
­dlee@​­ices​.­on​.­ca or at ICES, 2075 Bay- are inadequate. Whether the use of a strategy to support clinicians in making
view Ave., Rm. G-106, Toronto, ON M4N decisions about discharging or admitting patients, coupled with rapid follow-up in
3M5, Canada. an outpatient clinic, would affect outcomes remains uncertain.
*The complete list of COACH (Compari-
son of Outcomes and Access to Care METHODS
for Heart Failure) Trial Investigators is
provided in the Supplementary Appen-
In a stepped-wedge, cluster-randomized trial conducted in Ontario, Canada, we
dix, available at NEJM.org. randomly assigned 10 hospitals to staggered start dates for one-way crossover
This article was published on November 5,
from the control phase (usual care) to the intervention phase, which involved the
2022, at NEJM.org. use of a point-of-care algorithm to stratify patients with acute heart failure accord-
N Engl J Med 2023;388:22-32.
ing to the risk of death. During the intervention phase, low-risk patients were
DOI: 10.1056/NEJMoa2211680 discharged early (in ≤3 days) and received standardized outpatient care, and high-
Copyright © 2022 Massachusetts Medical Society. risk patients were admitted to the hospital. The coprimary outcomes were a com-
posite of death from any cause or hospitalization for cardiovascular causes within
30 days after presentation and the composite outcome within 20 months.

RESULTS
A total of 5452 patients were enrolled in the trial (2972 during the control phase
and 2480 during the intervention phase). Within 30 days, death from any cause or
hospitalization for cardiovascular causes occurred in 301 patients (12.1%) who
were enrolled during the intervention phase and in 430 patients (14.5%) who were
enrolled during the control phase (adjusted hazard ratio, 0.88; 95% confidence
interval [CI], 0.78 to 0.99; P = 0.04). Within 20 months, the cumulative incidence
of primary-outcome events was 54.4% (95% CI, 48.6 to 59.9) among patients who
were enrolled during the intervention phase and 56.2% (95% CI, 54.2 to 58.1)
among patients who were enrolled during the control phase (adjusted hazard ratio,
0.95; 95% CI, 0.92 to 0.99). Fewer than six deaths or hospitalizations for any cause
occurred in low- or intermediate-risk patients before the first outpatient visit
within 30 days after discharge.

CONCLUSIONS
Among patients with acute heart failure who were seeking emergency care, the use
of a hospital-based strategy to support clinical decision making and rapid follow-
up led to a lower risk of the composite of death from any cause or hospitalization
for cardiovascular causes within 30 days than usual care. (Funded by the Ontario
SPOR Support Unit and others; COACH ClinicalTrials.gov number, NCT02674438.)

22 n engl j med 388;1  nejm.org  January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

H
eart failure affects approximate- ducted in 10 hospitals in Ontario, Canada. A list
ly 26 million people worldwide and ex- of participating sites is provided in the Supple-
erts substantial pressure on health care mentary Appendix, available with the full text of
A Quick Take
systems as a result of its high levels of associ- this article at NEJM.org. Because the trial evalu- is available at
ated morbidity and resource use.1,2 A key indicator ated a hospital-level intervention that required NEJM.org
of the burden on patients with heart failure and the involvement of multiple disciplines, random-
on health systems is hospitalization; heart fail- ization was performed at the hospital level. The
ure leads to more than 1 million hospital admis- stepped-wedge, cluster-randomized design was
sions and readmissions annually in the United chosen to facilitate recruitment and implementa-
States alone.3 The 30-day risk of readmission and tion of the intervention at all participating sites.11
30-day mortality associated with heart failure have Each hospital was randomly assigned to a
not decreased substantively over time.4,5 Conse- sequence that indicated the period (cluster) dur-
quently, new approaches to improve clinical deci- ing which crossover from the control phase to
sion making and care are needed if health sys- the intervention phase would occur. There were
tems are to improve outcomes for patients with five sequences and six periods, with each period
heart failure. having a duration of 4 months (Fig. S1 in the
The first point of medical contact for patients Supplementary Appendix). A covariate-constrained
with acute heart failure is often the emergency randomization method12 was used to allow bal-
department, where physicians have traditionally ance between patients who were enrolled during
relied on clinical judgment to decide whether to the intervention phase and those who were en-
discharge or admit their patients.6 However, when rolled during the control phase with regard to
such decisions are based on clinical judgment, status as a teaching hospital (there were five aca-
some high-risk patients are discharged directly demic hospitals and five community hospitals)
from the emergency department and then have and the annual volume of patients with heart fail-
early adverse events, including death.7 Conversely, ure.10 An independent statistician implemented the
some low-risk patients are admitted when they randomization scheme.13 All research personnel
could have been discharged and monitored in an and hospitals were unaware of the hospital se-
outpatient clinic.7 One of the barriers to early quence assignments until 4 months before im-
discharge of low-risk patients is the lack of access plementation of an intervention; at that point,
to transitional care, which could include follow- staff were informed about the intervention, and
up outpatient care with a cardiac specialist during the implementation team performed training be-
the vulnerable period after discharge; the lack of fore crossover.
access to transitional care may contribute to a Some of the material presented here is based
high risk of readmission.8,9 on data and information compiled and provided
In the Comparison of Outcomes and Access by the Canadian Institute for Health Informa-
to Care for Heart Failure (COACH) trial, we as- tion and the Ontario Ministry of Health. Details
sessed a strategy to support clinical decision mak- regarding the trial design are provided in the
ing, which involved the use of a tool for objective protocol, available at NEJM.org.
risk stratification in the emergency department,
combined with the provision of standardized Trial Population
transitional care when indicated for patients with Patients were eligible for enrollment in the trial
heart failure. Our objective was to determine if they were at least 18 years of age and pre-
whether this strategy led to better clinical out- sented to the emergency department with acute
comes than usual care, which involved the use of heart failure. The clinical diagnosis of heart
clinical judgment and estimation of risk for deci- failure was verified with data from the hospital
sion making, as well as nonstandardized routine record and was based on the emergency depart-
follow-up after discharge. ment face sheet showing a primary diagnosis
code of I50 from the International Classification of
Diseases, 10th Clinical Canadian Modification.10
Me thods
The following patients were excluded: pa-
Trial Design tients who did not have a clinical diagnosis of
The COACH trial was a cross-sectional, stepped- heart failure according to Framingham Heart
wedge, cluster-randomized trial10,11 that was con- Study criteria or had a B-type natriuretic peptide

n engl j med 388;1  nejm.org  January 5, 2023 23

The New England Journal of Medicine
Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

level that indicated that the diagnosis was un- Trial Intervention
likely (i.e., a B-type natriuretic peptide level of The trial intervention consisted of a strategy to
<100 pg per milliliter or an N-terminal pro–B-type support clinicians in making decisions about dis-
natriuretic peptide level of <300 pg per millili- charging or admitting patients who presented to
ter)10; patients who had end-stage disease or were the emergency department with acute heart fail-
receiving palliative care, because they were not ure. A previously derived and validated point-of-
eligible for risk estimation with the use of the care tool for risk stratification (EHMRG30-ST)14,15,17
Emergency Heart Failure Mortality Risk Grade was made available to clinicians (Tables S3 through
for 30-day mortality (EHMRG30-ST)14,15; patients S6). The EHMRG30-ST tool was housed on a cen-
whose data could not be linked to databases tral server at ICES and was made accessible to
because of an invalid health-card number; pa- Internet portal addresses at participating hospi-
tients who were unable to attend visits in an out- tals only after crossover to the intervention
patient clinic (i.e., those who had limited mobility phase. During the intervention phase, hospital
or dementia or were nursing-home residents), staff could use a secure Web interface to access
because greater burden would be placed on the the EHMRG30-ST tool in order ascertain wheth-
patient if care was shifted to the outpatient set- er patients had a low, intermediate, or high risk
ting as part of participation in the trial; and of death within 7 days or within 30 days (Fig. S2).
patients who did not have a permanent home Patients who had a low risk of both death
address, those who were not residents of On- within 7 days and death within 30 days were rec-
tario, Canada, and those who left the hospital ommended to be discharged early and to receive
against medical advice, because of logistic dif- standardized transitional care, and patients who
ficulties with scheduling multiple outpatient had a high risk were recommended to be admit-
visits (Table S1). Research ethics approval was ted to the hospital. Clinicians were advised to
obtained at all participating hospitals, and the use clinical judgment for patients who had an
requirement for informed consent was waived to intermediate risk, but the general guidance was
allow for the inclusion of patients regardless of to admit patients who had an intermediate-to-
language and other potential barriers to trial high risk and to consider early discharge for
participation. patients who had a low-to-intermediate risk.
Trained chart abstractors, who were unaware Early discharge was defined as either discharge
of the hospital sequence assignments, obtained directly from the emergency department or dis-
detailed clinical information from hospital re- charge after an observation period in the hospi-
cords — including demographic characteristics, tal of up to 3 days. Patients who were discharged
presenting signs and symptoms, medical histo- early were given access to standardized transi-
ry, left ventricular function, and laboratory test tional care in the Rapid Ambulatory Program for
results and biomarker levels — and confirmed Investigation and Diagnosis of Heart Failure
the clinical diagnosis of heart failure with data (RAPID-HF) clinic. The RAPID-HF clinic was
from hospital records for all trial participants. staffed by a nurse and supervised by a cardiolo-
With the use of government-issued health-card gist, and the clinic provided outpatient care for
numbers, trial participants were linked to ad- up to 30 days after discharge from the emer-
ministrative databases that provided informa- gency department or hospital, as described pre-
tion about hospitalizations (Canadian Institute viously (Tables S7 through S9 and Fig. S3).10,18
for Health Information Discharge Abstract Data-
base), emergency department visits (National Trial Outcomes
Ambulatory Care Reporting System), physician Two coprimary outcomes were specified (Table
claims (Table S2), and vital status for all resi- S10). The first coprimary outcome was a com-
dents of the province (Registered Persons Data- posite of death from any cause or hospitalization
base).16 These data sets were linked with the use for cardiovascular causes within 30 days after
of unique encoded identifiers and analyzed at presentation to the emergency department, eval-
ICES (formerly the Institute for Clinical Evalua- uated in a time-to-event analysis (i.e., early out-
tive Sciences). The use of data in this project was come).10 The second coprimary outcome was the
authorized under section 45 of the Ontario Per- composite outcome within 20 months (600 days)
sonal Health Information Protection Act. after presentation (i.e., extended outcome). The

24 n engl j med 388;1  nejm.org  January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

coprimary outcomes in our trial differed from stepped-wedge design with a sample of 10 hos-
the original outcomes (death or hospitalization pitals (2 hospitals randomly assigned to each of
for any cause) that were proposed when the trial the five sequences) and a mean of 145 patients
was conceived during application for funding, per period (a total of 8700 patients over the six
because in preparation for the trial, we found that periods, including patients at hospitals in the
among patients who had a shorter hospital ad- control phase and those at hospitals in the inter-
mission, the risk of hospitalization for cardio- vention phase in each period) provided the trial
vascular causes was increased during the early with 85% power to detect a clinically important
period after discharge, whereas no relationship relative difference of 10%, at a two-sided alpha
with the risk of hospitalization for any cause level of 0.05. Our calculation was performed
during that period was present.19 under the assumption of an intracluster correla-
Hospitalizations for cardiovascular causes that tion coefficient of 0.01 (based on the analysis of
occurred after the index presentation (and after ICES data) and with the use of a discrete expo-
the index hospital admission if indicated), were nential decay model to allow for 20% decay in
not elective, and were longer than 1 day were the strength of the correlation per period. No
counted as primary-outcome events. A hospital- adjustment was made for cluster attrition because
ization could not be counted as both an index the risk of attrition was extremely low and all
event and a primary-outcome event. If a hospi- outcome data were to be collected routinely, re-
talization or death occurred during the interven- gardless of any dropout.
tion phase but the index presentation occurred The coprimary and secondary outcomes were
during the control phase, the primary-outcome analyzed with the use of Cox proportional-
event was attributed to the control phase. hazards models, with adjustment for period to
Secondary outcomes included each of the fol- account for any secular trend and with robust vari-
lowing outcomes (evaluated in time-to-event anal- ance estimation to account for clustering according
yses): hospitalization for cardiovascular causes, to hospital.10 We entered time as steps (with
hospitalization for heart failure, and death from each step being a 4-month interval of time) as
any cause. Because the EHMRG30-ST tool can- fixed effects into the model to account for peri-
not be used to estimate risk in patients who are od effects. The analyses were also adjusted for
receiving palliative care, and because patients status as a teaching hospital and the annual
with heart failure may opt for palliative care dur- volume of patients with heart failure, the factors
ing extended follow-up, mortality data were ob- used in covariate-constrained randomization.10
tained until patients were transitioned to pallia- Binary outcomes were compared with the use of
tive care (e.g., symptom control or comfort care) logistic-regression models, generalized estimation
in accordance with previously published and equation methods to account for within-hospital
validated methods (Table S11).20 A secondary clustering, the small-sample correction method
outcome that was chosen by the patient advisory of Kauermann and Carroll (because the number
panel was a composite of the first nonelective of sites in the trial was not large), and further
emergency department visit, death from any cause, adjustment for period, status as a teaching hos-
or hospitalization for cardiovascular causes, evalu- pital, and the annual volume of patients with
ated in a time-to-event analysis.21 Serious adverse heart failure.22 Cumulative-incidence curves from
events were defined as death or hospitalization adjusted and unadjusted analyses of outcomes
for either cardiovascular causes or any cause that were constructed as appropriate. Censoring of
occurred in low- or intermediate-risk patients data occurred at 30 days or 20 months or on the
after early discharge and before the first outpa- last day of the study follow-up period (January
tient visit. 15, 2019). For the secondary outcome of death
from any cause, data were censored when a pa-
Statistical Analysis tient entered palliative care.
Power calculations were based on the analysis of The statistical analysis began on December 1,
ICES data regarding patients with heart failure, 2021, when the administrative data linkages for
which showed a baseline risk of primary-outcome follow-up outcomes first became available. How-
events of 65% at 1 year after presentation to the ever, our statistical analysis plan did not include
emergency department. We calculated that a adjustment for the type I error rate for two copri-

n engl j med 388;1  nejm.org  January 5, 2023 25

The New England Journal of Medicine
Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

10,596 Patients underwent screening

in an emergency department

4659 Were screened during 5937 Were screened during

the intervention phase the control phase

919 Were excluded 1778 Were excluded

126 Were receiving palliative 678 Were receiving palliative
care care
679 Did not have heart 992 Did not have heart
failure failure
114 Could not be linked to 108 Could not be linked to
databases databases

3740 Were eligible for inclusion in the trial 4159 Were eligible for inclusion in the trial

1260 Were excluded 1187 Were excluded

503 Had limited mobility 414 Had limited mobility
498 Had dementia 467 Had dementia
233 Resided in a nursing 285 Resided in a nursing
home or had no fixed home or had no fixed
address address
13 Left against medical advice 15 Left against medical advice
13 Were nonresidents 6 Were nonresidents

2480 Were enrolled during 2972 Were enrolled during

the intervention phase the control phase

Figure 1. Patient Enrollment According to Hospital Phase.

mary outcomes and did not specify alpha alloca- matched with respect to baseline characteristics
tion. Therefore, a P value is reported for the (Table 1 and Table S12); there was a high percent-
30-day coprimary outcome only. The remainder age of patients who had a history of heart failure
of the data are presented as point estimates and in both groups (Table S13). The trial population
95% confidence intervals. A two-sided P value of was representative of patients with heart failure
less than 0.05 was considered to indicate signifi- in the general population; the distributions ac-
cance. The widths of the 95% confidence inter- cording to age, sex, and race or ethnic group
vals were not adjusted for multiple comparisons were similar to those observed in population-
and should not be used to infer definitive effects based studies (Table S14).5,23 No patients were no
of the intervention. lost to follow-up.

Patient Risk Groups, Early Discharge,

R e sult s and Outpatient Care
Patients During the control phase, hospitals did not have
Of the 10,596 patients who presented to the access to the EHMRG30-ST tool, and risk scores
emergency department with acute heart failure, could not be determined by hospital staff. For
7899 were eligible for inclusion in the trial. After this phase, risk scores could be calculated after
the exclusion of those who had reasons that lim- completion of the trial, during the analysis
ited them from attending visits in the RAPID-HF stage, for 2919 of 2972 patients (98.2%). Of the
clinic, 5452 patients were enrolled in the trial 2919 patients, 531 (18.2%) had a low risk, 824
(2972 during the control phase and 2480 during (28.2%) had an intermediate risk, and 1564
the intervention phase) (Fig. 1). The patients who (53.6%) had a high risk. During the intervention
were enrolled during the two phases were well phase, risk scores could be calculated with the

26 n engl j med 388;1  nejm.org  January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

Table 1. Characteristics of the Patients at Baseline.*

Intervention Control
Characteristic (N = 2480) (N = 2972)
Median age (IQR) — yr 78 (68–85) 78 (68–85)
Female sex — no. (%) 1125 (45.4) 1336 (45.0)
Hypertension — no. (%) 1810 (73.0) 2289 (77.0)
Diabetes — no. (%) 960 (38.7) 1227 (41.3)
Previous myocardial infarction — no. (%) 377 (15.2) 526 (17.7)
Chronic atrial fibrillation or flutter — no. (%) 952 (38.4) 1012 (34.1)
Stroke or transient ischemic attack — no. (%) 321 (12.9) 403 (13.6)
Severe mitral or aortic valvular heart disease — no. (%) 331 (13.3) 466 (15.7)
Cancer — no. (%) 128 (5.2) 113 (3.8)
Previous diagnosis of heart failure — no./total no. ≥40 yr of age (%) 1551/2447 (63.4) 1913/2928 (65.3)
Median no. of previous hospitalizations for heart failure since onset (IQR) 0 (0–0) 0 (0–1)
Presenting signs
Median systolic blood pressure (IQR) — mm Hg 138 (121–158) 135 (116–155)
Median heart rate (IQR) — beats/min 85 (71–103) 88 (73–107)
Median oxygen saturation (IQR) — % 95 (92–97) 94 (89–96)
Medications before presentation — no. (%)
Renin–angiotensin system inhibitor† 1293 (52.1) 1538 (51.7)
Beta-blocker 1538 (62.0) 1793 (60.3)
Mineralocorticoid-receptor inhibitor 263 (10.6) 331 (11.1)
Sodium–glucose cotransporter 2 inhibitor 69 (2.8) 38 (1.3)
Furosemide 1279 (51.6) 1614 (54.3)
Metolazone 53 (2.1) 58 (2.0)
Laboratory test results in emergency department
Median eGFR (IQR) — ml/min/1.73 m2 56.5 (38.6–76.5) 55.8 (37.1–76.1)
Median potassium concentration (IQR) — mEq/liter 4.0 (4.0–5.0) 4.0 (4.0–5.0)
Elevated troponin level — no./total no. (%)‡ 879/2388 (37) 1152/2887 (40)
Left ventricular ejection fraction — no. (%)
≤40%: Heart failure with reduced ejection fraction 684 (27.6) 818 (27.5)
41–49%: Heart failure with middle-range ejection fraction 231 (9.3) 291 (9.8)
≥50%: Heart failure with preserved ejection fraction 953 (38.4) 1039 (35.0)
Not assessed 612 (24.7) 824 (27.7)

* Percentages may not total 100 because of rounding. The term eGFR denotes estimated glomerular filtration rate, and
IQR interquartile range.
† Renin–angiotensin system inhibitors include angiotensin-converting–enzyme inhibitors, angiotensin-receptor blockers,
and sacubitril–valsartan.
‡ An elevated troponin level was defined as a level higher than the upper limit of the normal range used for the detection
of myocardial injury or used for non–high-sensitivity assays or a level more than three times the 99th percentile used
for high-sensitivity assays.

use of the EHMRG30-ST tool for 2442 of 2480 Among high-risk patients, early discharge
patients (98.5%). Of the 2442 patients, 575 (i.e., discharge directly from the emergency de-
(23.5%) had a low risk, 783 (32.1%) had an inter- partment or discharge after a hospitalization of
mediate risk, and 1084 (44.4%) had a high risk. ≤3 days) occurred in 423 of 1564 patients (27.0%)

n engl j med 388;1  nejm.org  January 5, 2023 27

The New England Journal of Medicine
Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

among those enrolled during the intervention

100 phase and 12 days (interquartile range, 5 to 29)
90 among those enrolled during the control phase
80 (adjusted hazard ratio, 1.38; 95% CI, 1.08 to 1.76).
Percentage of Patients Who For intermediate-risk patients who were dis-
Were Discharged Early
70
57 58 charged early, the median time to the first out-
60
51 patient visit was 4 days (interquartile range, 2 to
50 44 7) among those enrolled during the intervention
40 phase and 5 days (interquartile range, 3 to 9)
30 27 among those enrolled during the control phase.
20
19 The median time to the first outpatient visit with
either an internist or a cardiologist was 7 days
10
(interquartile range, 3 to 20) and 9 days (inter-
0 quartile range, 4 to 27) among those enrolled
Co n

Co n

Co n

l
ro

ro

ro
during the intervention phase and the control
tio

tio

tio
nt

nt

nt
en

en

en
rv

rv

rv phase, respectively.
te

te

te
In

In

In

Low Risk Intermediate Risk High Risk For high-risk patients who were discharged
early, the median time to the first outpatient
Figure 2. Early Discharge According to Risk Group. visit was 5 days (interquartile range, 3 to 9)
Early discharge was defined as either discharge directly among those enrolled during the intervention
from the emergency department or discharge after a phase and 5 days (interquartile range, 3 to 11)
hospitalization of up to 3 days.
among those enrolled during the control phase.
The median time to the first outpatient visit with
either an internist or a cardiologist was 11 days
who were enrolled during the control phase and (interquartile range, 4 to 27) and 13 days (inter-
in 207 of 1084 patients (19.1%) who were en- quartile range, 5 to 33) among those enrolled
rolled during the intervention phase (Fig. 2 and during the intervention phase and the control
Fig. S4). Among intermediate-risk patients, early phase, respectively. Additional results regarding
discharge occurred in 364 of 824 patients (44.2%) outpatient visits with stratification according to
and in 397 of 783 patients (50.7%) who were risk group are shown in Table S15.
enrolled during the control phase and the inter-
vention phase, respectively. Among low-risk pa- Outcomes within 30 Days
tients, early discharge occurred in 309 of 531 Within 30 days, death from any cause or hospi-
patients (58.2%) and in 328 of 575 patients talization for cardiovascular causes occurred in
(57.0%) who were enrolled during the control 301 patients (12.1%) who were enrolled during
phase and the intervention phase, respectively. the intervention phase and in 430 patients (14.5%)
Early discharge was most likely to occur on the who were enrolled during the control phase (ad-
first day after presentation, and trends for dis- justed hazard ratio, 0.88; 95% CI, 0.78 to 0.99;
charge directly from the emergency department P = 0.04). The risk of hospitalization for cardio-
were consistent with trends for early discharge vascular causes appeared to be lower during the
(Fig. S5). intervention phase than during the control phase
For low-risk patients who were discharged (adjusted hazard ratio, 0.85; 95% CI, 0.74 to
early, the median time to the first outpatient visit 0.98), as did the risk of hospitalization for heart
was 4 days (interquartile range, 2 to 7) among failure (adjusted hazard ratio, 0.81; 95% CI, 0.69
those enrolled during the intervention phase and to 0.95). Results for other secondary outcomes
4 days (interquartile range, 3 to 8) among those within 30 days are shown in Table 2.
enrolled during the control phase (adjusted haz- A post hoc analysis of primary-outcome
ard ratio, 1.10; 95% confidence interval [CI], 0.94 events within 30 days with stratification accord-
to 1.28). In addition, the median time to the first ing to risk group was performed. Death from any
outpatient visit with either an internist or a car- cause or hospitalization for cardiovascular causes
diologist was 6 days (interquartile range, 3 to 12) occurred in 36 low-risk patients (6.3%) enrolled

28 n engl j med 388;1  nejm.org  January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

Table 2. Primary and Secondary Outcomes within 30 Days.

Adjusted
Intervention Control Hazard Ratio
Outcome (N = 2480) (N = 2972) (95% CI) P Value
Primary outcome
Composite of death from any cause or hospitalization 301 (12.1) 430 (14.5) 0.88 (0.78–0.99) 0.04
for cardiovascular causes — no. (%)
Secondary outcomes
Hospitalization for cardiovascular causes — no./total 190/2343 (8.1) 294/2775 (10.6) 0.85 (0.74–0.98) —
no. (%)
Hospitalization for heart failure — no./total no. (%) 142/2343 (6.1) 222/2775 (8.0) 0.81 (0.69–0.95) —
Death from any cause — no. (%) 147 (5.9) 196 (6.6) 0.94 (0.74–1.19) —
Composite of emergency department visit, death from 687 (27.7) 851 (28.6) 0.97 (0.85–1.11) —
any cause, or hospitalization for cardiovascular
causes — no. (%)

during the intervention phase and in 36 low-risk Serious Adverse Events

patients (6.8%) enrolled during the control phase Among patients who were enrolled during the
(hazard ratio, 0.88; 95% CI, 0.55 to 1.39); in 67 intervention phase, the risk of serious adverse
intermediate-risk patients (8.6%) enrolled dur- events occurring before the first outpatient visit
ing the intervention phase and in 100 intermedi- within 30 days after discharge did not appear to
ate-risk patients (12.1%) enrolled during the be higher than the risk among patients who
control phase (hazard ratio, 0.66; 95% CI, 0.43 were enrolled during the control phase. Specifi-
to 1.01); and in 190 high-risk patients (17.5%) cally, no deaths or hospitalizations for any cause
enrolled during the intervention phase and in occurred in low- or intermediate-risk patients
285 high-risk patients (18.2%) enrolled during before the first outpatient visit within 7 days, and
the control phase (hazard ratio, 1.05; 95% CI, fewer than six events occurred within 30 days
0.91 to 1.21) (Table S17). (Table S18).

Outcomes within 20 Months

Discussion
Within 20 months, the cumulative incidence of
death from any cause or hospitalization for car- In this trial of a strategy for patients with heart
diovascular causes was 54.4% (95% CI, 48.6 to failure, the use of a validated point-of-care tool
59.9) among patients who were enrolled during for risk stratification in the emergency depart-
the intervention phase and 56.2% (95% CI, 54.2 ment to support clinicians in making decisions
to 58.1) among patients who were enrolled dur- about discharging or admitting patients, com-
ing the control phase (adjusted hazard ratio, 0.95; bined with the provision of standardized transi-
95% CI, 0.92 to 0.99). The intervention appeared tional care, led to a 12% lower risk of death
to be associated with lower risks of hospitaliza- from any cause or hospitalization for cardiovas-
tion for cardiovascular causes and hospitalization cular causes within 30 days after presentation
for heart failure (Table S16). The median follow- than usual care. The intervention also led to a
up time was 280 days (interquartile range, 82 to lower risk of death from any cause or hospitaliza-
520) among patients enrolled during the control tion for cardiovascular causes within 20 months.
phase and 144 days (interquartile range, 64 to The risks of hospitalization for cardiovascular
286) among patients enrolled during the inter- causes and hospitalization for heart failure after
vention phase. Cumulative-incidence curves from the index presentation also appeared to be lower
adjusted and unadjusted analyses of early and among patients who were enrolled during the
extended outcomes are shown in Figure 3, and intervention phase than among those who were
curves for secondary outcomes are shown in Fig- enrolled during the control phase.
ures S6, S7, and S8. Not all patients who present to the emergency

n engl j med 388;1  nejm.org  January 5, 2023 29

The New England Journal of Medicine
Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

A Adjusted Analysis of Composite Primary Outcome within 30 Days B Unadjusted Analysis of Composite Primary Outcome within 30 Days
100 20 100 20
90 Hazard ratio, 0.88 (95% CI, 0.78–0.99)
90
P=0.04

Unadjusted Cumulative Incidence (%)

Control
Adjusted Cumulative Incidence (%)

80 15 Control 80 15

70 70
60 10 10
60
Intervention
50 50 Intervention
5 5
40 40

30 30
0 0
20 0 5 10 15 20 25 30 20 0 5 10 15 20 25 30
10 10

0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Days after Presentation Days after Presentation

C Adjusted Analysis of Composite Primary Outcome within D Unadjusted Analysis of Composite Primary Outcome within
20 Months 20 Months
100 100

90 90
Unadjusted Cumulative Incidence (%)
Adjusted Cumulative Incidence (%)

80 Hazard ratio, 0.95 (95% CI, 0.92–0.99) 80

70 70

60 Control 60 Control

50 50
Intervention
40 40 Intervention

30 30

20 20

10 10

0 0
0 60 120 180 240 300 360 420 480 540 600 0 60 120 180 240 300 360 420 480 540 600
Days after Presentation Days after Presentation

Figure 3. Cumulative Incidence of the Composite of Death from Any Cause or Hospitalization for Cardiovascular Causes.
The first coprimary outcome was a composite of death from any cause or hospitalization for cardiovascular causes within 30 days after
presentation to the emergency department, and the second coprimary outcome was the composite outcome within 20 months after pre-
sentation. Cumulative-incidence curves from adjusted and unadjusted analyses of the first coprimary outcome (Panels A and B, respec-
tively) and from adjusted and unadjusted analyses of the second coprimary outcome (Panels C and D, respectively) are shown. Shading
indicates the 95% confidence interval. Insets show the same data on an enlarged y axis.

department with heart failure require hospital- patients, and may improve efficiency by reducing
ization.24 A barrier to the selection of patients admission of lower-risk patients.7 In our trial, the
for early hospital discharge is the inability to ac- strategy of risk assessment with a validated algo-
curately predict the occurrence of adverse events rithm, combined with the provision of transi-
on the basis of physician risk estimation alone.25 tional care, was associated with a lower risk of
The ability to prognosticate more accurately hospitalization among patients with heart failure.
may enable physicians to make informed deci- Transitional care for patients with acute heart
sions about appropriate care settings, may en- failure has had varied degrees of success,26-28 but
hance safety by reducing discharge of high-risk it may be an important intervention nonetheless.

30 n engl j med 388;1 nejm.org January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

Feltner et al. reported that the provision of tran- pects of the complex intervention had the great-
sitional care reduced the 30-day risk of readmis- est effect. For example, the RAPID-HF clinic was
sion, whereas primary care visits and educational managed by a nurse and supervised by a cardiolo-
programs alone tended to increase the risk of gist, so the effect that was attributable to care by
hospitalization.29 In the Patient-Centered Care a cardiac specialist could not be ascertained. Sec-
Transitions in Heart Failure study, a peridischarge ond, a health system–based trial of a complex in-
intervention that consisted of patient education, tervention is subject to the “learning curve” of the
a structured hospital-discharge summary, a fam- health care team. Although acceleration of the
ily-physician appointment, and home visits by a learning process was made possible by a nurse
nurse did not reduce the risks of readmission for navigator, benefits could be underestimated in the
any cause, emergency department visits, or death.30 evaluation of early outcomes, because care pro-
The involvement of specialists31 and the rapidity cesses become more efficient over time. Third,
of outpatient follow-up32 may be important, since race or ethnic group was not known for more than
readmissions can occur within the first few days half the participants in each group. Therefore, we
after hospital discharge.33 cannot make any conclusions about the effect of
Our intervention was complex and multi- race or ethnic group on outcomes. Finally, not all
pronged, with several potential mechanisms of patients could be linked to data on follow-up out-
benefit. Physicians tend to underestimate risk for comes, and the success of the linkage appeared to
those with the worst prognosis,17 which could lead be marginally higher among patients who were
to inadvertent discharge of high-risk patients.7 enrolled during the control phase.
In addition, earlier follow-up may be advanta- For patients with acute heart failure who pre-
geous, because readmissions occur frequently sented to the emergency department, the system-
after shorter index hospital admissions.19 atic use of a point-of-care tool to support clinical
Our findings support the concept that not all decision making, coupled with rapid follow-up
patients who present to the emergency depart- in an outpatient clinic, led to a lower risk of
ment with heart failure require hospitalization. death from any cause or hospitalization for car-
If a patient is not assessed as having a high risk diovascular causes within 30 days after presen-
with the use of a broadly validated method, and tation than usual care. Implementation of this
if a structured outpatient clinic is available for approach across health systems may provide a
rapid follow-up, then this lower-risk patient may pathway for early and safe discharge from the
be discharged directly from the emergency depart- hospital or emergency department and improved
ment or discharged early after hospital admission. patient outcomes.
Prospective testing before implementation is im- Supported by the Ontario SPOR (Strategy for Patient-Oriented
portant, and pragmatic randomized trials repre- Research) Support Unit, the Ted Rogers Centre for Heart Re-
sent an approach for this form of evaluation.34,35 search, the Peter Munk Cardiac Centre, a Foundation Grant
(FDN 148446) from the Canadian Institutes of Health Research,
Policymakers may consider appropriate discharge and ICES, which is funded by an annual grant from the Ontario
with rapid follow-up to be a viable alternative to Ministry of Health and the Ministry of Long-Term Care.
hospitalization in some situations. Disclosure forms provided by the authors are available with
the full text of this article at NEJM.org.
This trial has some limitations. First, the trial A data sharing statement provided by the authors is available
design limited our ability to determine which as- with the full text of this article at NEJM.org.

Appendix
The authors’ full names and academic degrees are as follows: Douglas S. Lee, M.D., Ph.D., Sharon E. Straus, M.D., Michael E. Farkouh,
M.D., Peter C. Austin, Ph.D., Monica Taljaard, Ph.D., Alice Chong, B.Sc., Christine Fahim, Ph.D., Stephanie Poon, M.D., Peter Cram,
M.D., Stuart Smith, M.D., Robert S. McKelvie, M.D., Ph.D., Liane Porepa, M.D., Michael Hartleib, M.D., Peter Mitoff, M.D., Robert M.
Iwanochko, M.D., Andrea MacDougall, M.D., Steven Shadowitz, M.D., C.M., Howard Abrams, M.D., Esam Elbarasi, M.B., Ch.B., Jim-
ing Fang, Ph.D., Jacob A. Udell, M.D., M.P.H., Michael J. Schull, M.D., Susanna Mak, M.D., Ph.D., and Heather J. Ross, M.D., M.H.Sc.
The authors’ affiliations are as follows: the University of Toronto (D.S.L., S.E.S., M.E.F., P.C.A., S.P., P.C., R.M.I., S. Shadowitz, H.A.,
J.A.U., M.J.S., S.M., H.J.R.), the Ted Rogers Centre for Heart Research and the Peter Munk Cardiac Centre, University Health Network
(D.S.L., M.E.F., J.A.U., H.J.R.), ICES (formerly the Institute for Clinical Evaluative Sciences) (D.S.L., P.C.A., A.C., P.C., J.F., J.A.U.,
M.J.S.), St. Michael’s Hospital and Li Ka Shing Knowledge Institute, Unity Health (S.E.S., C.F.), the Divisions of Cardiology (S.P.) and
General Internal Medicine (S. Shadowitz) and the Department of Emergency Services and Sunnybrook Research Institute (M.J.S.), Sun-
nybrook Health Sciences Centre, the Division of Cardiology, St. Joseph’s Hospital (P.M.), the Division of Cardiology, Toronto Western
Hospital (R.M.I.), the Division of General Internal Medicine, Toronto General Hospital (H.A.), the Division of Cardiology, Women’s
College Hospital (J.A.U.), and the Division of Cardiology, Sinai Health (S.M.), Toronto, the Clinical Epidemiology Program, Ottawa

n engl j med 388;1  nejm.org  January 5, 2023 31

The New England Journal of Medicine
Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.
 Intervention to Improve Acute Heart Failure Outcomes

Hospital Research Institute, School of Epidemiology and Public Health, University of Ottawa, Ottawa (M.T.), the Division of Cardiology,
London Health Sciences Centre (S. Smith), Western University (S. Smith, R.S.M.), and the Division of Cardiology, St. Joseph’s Health
Care (R.S.M.), London, the Division of Cardiology, Southlake Regional Health Centre, Newmarket (L.P.), the Division of Cardiology,
Peterborough Regional Health Centre, Peterborough (M.H.), the Division of Cardiology, Thunder Bay Regional Health Sciences Centre,
Thunder Bay (A.M.), and the Division of Cardiology, William Osler Health System, Brampton (E.E.) — all in Ontario, Canada; and the
Department of Medicine, University of Texas Medical Branch, Galveston (P.C.).

References
1. Savarese G, Lund LH. Global public of a tuberculosis preventive therapy inter- intervene? Curr Heart Fail Rep 2015;​12:​
health burden of heart failure. Card Fail vention. Clin Trials 2007;​4:​190-9. 309-17.
Rev 2017;​3:​7-11. 13. Greene EJ. CCR_SW for SAS. GitHub, 25. Lee DS, Lee JS, Schull MJ, Grimshaw
2. Dharmarajan K, Wang Y, Lin Z, et al. May 10, 2017 (https://github​.­com/​­ejgreene/​ JM, Austin PC, Tu JV. Design and rationale
Association of changing hospital read- ­ccrsw​-­sas). for the Acute Congestive Heart Failure Ur-
mission rates with mortality rates after 14. Lee DS, Stitt A, Austin PC, et al. Pre- gent Care Evaluation: the ACUTE study.
hospital discharge. JAMA 2017;​318:​270-8. diction of heart failure mortality in emer- Am Heart J 2016;​181:​60-5.
3. Tsao CW, Aday AW, Almarzooq ZI, et gent care: a cohort study. Ann Intern Med 26. Freund Y, Cachanado M, Delannoy Q,
al. Heart disease and stroke statistics — 2012;​156:​767-75. et al. Effect of an emergency department
2022 update: a report from the American 15. Greig D, Austin PC, Zhou L, et al. care bundle on 30-day hospital discharge
Heart Association. Circulation 2022;​145(8):​ Ischemic electrocardiographic abnormal- and survival among elderly patients with
e153-e639. ities and prognosis in decompensated acute heart failure: the ELISABETH ran-
4. Khera R, Dharmarajan K, Wang Y, et heart failure. Circ Heart Fail 2014;​7:​986- domized clinical trial. JAMA 2020;​ 324:​
al. Association of the hospital readmis- 93. 1948-56.
sions reduction program with mortality 16. Tu JV, Chu A, Donovan LR, et al. The 27. O’Connor CM, Starling RC, Hernan-
during and after hospitalization for acute Cardiovascular Health in Ambulatory dez AF, et al. Effect of nesiritide in pa-
myocardial infarction, heart failure, and Care Research Team (CANHEART): using tients with acute decompensated heart
pneumonia. JAMA Netw Open 2018;​1(15):​ big data to measure and improve cardio- failure. N Engl J Med 2011;​365:​32-43.
e182777. vascular health and healthcare services. 28. Metra M, Teerlink JR, Cotter G, et al.
5. Ko DT, Khera R, Lau G, et al. Read- Circ Cardiovasc Qual Outcomes 2015;​8:​ Effects of serelaxin in patients with acute
mission and mortality after hospitaliza- 204-12. heart failure. N Engl J Med 2019;​381:​716-
tion for myocardial infarction and 17. Lee DS, Lee JS, Schull MJ, et al. Pro- 26.
heart failure. J Am Coll Cardiol 2020;​75:​ spective validation of the emergency heart 29. Feltner C, Jones CD, Cené CW, et al.
736-46. failure mortality risk grade for acute heart Transitional care interventions to prevent
6. Lam CSP, Ferreira JP, Pfarr E, et al. failure. Circulation 2019;​139:​1146-56. readmissions for persons with heart fail-
Regional and ethnic influences on the re- 18. Ezekowitz JA, O’Meara E, McDonald ure: a systematic review and meta-analy-
sponse to empagliflozin in patients with MA, et al. 2017 comprehensive update of sis. Ann Intern Med 2014;​160:​774-84.
heart failure and a reduced ejection frac- the Canadian Cardiovascular Society 30. Van Spall HGC, Lee SF, Xie F, et al.
tion: the EMPEROR-Reduced trial. Eur guidelines for the management of heart Effect of patient-centered transitional
Heart J 2021;​42:​4442-51. failure. Can J Cardiol 2017;​33:​1342-433. care services on clinical outcomes in pa-
7. Lee DS, Schull MJ, Alter DA, et al. 19. Sud M, Yu B, Wijeysundera HC, et al. tients hospitalized for heart failure: the
Early deaths in patients with heart failure Associations between short or long PACT-HF randomized clinical trial. JAMA
discharged from the emergency depart- length of stay and 30-day readmission 2019;​321:​753-61.
ment: a population-based analysis. Circ and mortality in hospitalized patients 31. Lee DS, Stukel TA, Austin PC, et al.
Heart Fail 2010;​3:​228-35. with heart failure. JACC Heart Fail 2017;​5:​ Improved outcomes with early collabora-
8. Greene SJ, Fonarow GC, Vadugana- 578-88. tive care of ambulatory heart failure pa-
than M, Khan SS, Butler J, Gheorghiade 20. Tanuseputro P, Budhwani S, Bai YQ, tients discharged from the emergency
M. The vulnerable phase after hospital- Wodchis WP. Palliative care delivery department. Circulation 2010;​122:​1806-
ization for heart failure. Nat Rev Cardiol across health sectors: a population-level 14.
2015;​12:​220-9. observational study. Palliat Med 2017;​31:​ 32. Atzema CL, Austin PC, Yu B, et al. Ef-
9. Bueno H, Ross JS, Wang Y, et al. 247-57. fect of early physician follow-up on mor-
Trends in length of stay and short-term 21. Lee DS, Armstrong R, Mohamed S. tality and subsequent hospital admissions
outcomes among Medicare patients hos- Patient engagement in a trial testing a after emergency care for heart failure: a
pitalized for heart failure, 1993–2006. new strategy of care for acute heart fail- retrospective cohort study. CMAJ 2018;​
JAMA 2010;​303:​2141-7. ure. CMAJ 2018;​190:​Suppl:​S34-S36. 190(50):​E1468-E1477.
10. Lee DS, Straus SE, Austin PC, et al. 22. Li F, Hughes JP, Hemming K, Taljaard 33. Dharmarajan K, Hsieh AF, Lin Z, et al.
Rationale and design of the comparison M, Melnick ER, Heagerty PJ. Mixed-effects Diagnoses and timing of 30-day readmis-
of outcomes and access to care for heart models for the design and analysis of sions after hospitalization for heart fail-
failure (COACH) trial: a stepped wedge stepped wedge cluster randomized trials: ure, acute myocardial infarction, or pneu-
cluster randomized trial. Am Heart J an overview. Stat Methods Med Res 2021;​ monia. JAMA 2013;​309:​355-63.
2021;​240:​1-10. 30:​612-39. 34. Reilly BM, Evans AT. Translating clin-
11. Hemming K, Taljaard M, McKenzie 23. Samsky MD, Ambrosy AP, Youngson ical research into clinical practice: impact
JE, et al. Reporting of stepped wedge clus- E, et al. Trends in readmissions and of using prediction rules to make deci-
ter randomised trials: extension of the length of stay for patients hospitalized sions. Ann Intern Med 2006;​144:​201-9.
CONSORT 2010 statement with explana- with heart failure in Canada and the Unit- 35. Ahmad T, Desai NR, Yamamoto Y, et
tion and elaboration. BMJ 2018;​ 363:​ ed States. JAMA Cardiol 2019;​4:​444-53. al. Alerting clinicians to 1-Year mortality
k1614. 24. Dunbar-Yaffe R, Stitt A, Lee JJ, Mo- risk in patients hospitalized with heart
12. Moulton LH, Golub JE, Durovni B, et hamed S, Lee DS. Assessing risk and pre- failure: the REVEAL-HF randomized clin-
al. Statistical design of THRio: a phased venting 30-day readmissions in decom- ical trial. JAMA Cardiol 2022;​7:​905-12.
implementation clinic-randomized study pensated heart failure: opportunity to Copyright © 2022 Massachusetts Medical Society.

32 n engl j med 388;1  nejm.org  January 5, 2023

The New England Journal of Medicine

Downloaded from nejm.org by Guilherme Carelli on February 25, 2023. For personal use only. No other uses without permission.
Copyright © 2023 Massachusetts Medical Society. All rights reserved.