An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Sean P. Collins, MD, MSc, Alan Storrow, MD, Dandan Liu, PhD, Cathy A. Jenkins, MS, Karen F. Miller, RN, BSN, MPA, Christy Kampe, MAcc, and Javed Butler, MD, MPH, MBA.
Author Information and AffiliationsDespite a relative reduction in the hospitalization rate for heart failure (HF), the actual number of HF hospitalizations remains >1 million annually. More than 80% of patients who are hospitalized are initially seen in the emergency department (ED). Importantly, the ED is the safety net for acute HF (AHF) care. Thus, it is the primary provider for vulnerable patients, including those with low socioeconomic status, minority populations, and those with poor health literacy. Hospitalized patients with HF are among those most frequently readmitted within 30 days; they have up to a 25% risk of readmission within a month and a 50% risk within 6 months of discharge.1 Events in patients with AHF discharged from the ED are reportedly much higher, and an even greater disparity exists between vulnerable and nonvulnerable patients. Self-care education and other strategies could improve this disparity gap. We proposed studying the impact of a self-care intervention on patients with AHF discharged from the ED.
We determined the impact of our self-care intervention on our primary outcome, a composite global rank prioritizing outcomes in the following order: cardiovascular (CV) death, HF event (first ED/clinic revisit or hospital admission), and Kansas City Cardiomyopathy Questionnaire (KCCQ) score at 90 days. We (1) determined the overall impact of our strategy regardless of vulnerable characteristics, (2) explored individual characteristics of vulnerability associated with the highest (and lowest) improvements from our intervention, and (3) assessed the reduction in disparities in our primary outcome between those with and without characteristics of vulnerability. Secondary end points included our primary composite outcome at 30 days, patient satisfaction, HF knowledge, and HF health status based on the KCCQ at 30 and 90 days.
Patients with AHF who were discharged after a brief ED stay at 15 sites were screened to ensure that they fulfilled our broad inclusion and exclusion criteria: history of HF, not on an outpatient inotrope infusion, systolic blood pressure (BP) >100 mm Hg, and no evidence of an acute coronary syndrome. They were randomly assigned 1:1 to structured usual ED discharge care vs a tailored intervention strategy that focused on self-care strategies over the subsequent 90 days. This intervention was conducted by study team coordinators, nurses, and paramedics trained under an identical protocol. We stratified by site and randomized at the patient level using random permuted blocks. Research staff who were blinded to intervention arm called all patients at 30 and 90 days after discharge to determine the primary and secondary outcomes. For our primary outcome, we analyzed the impact of our intervention on our global rank end point at 90 days. This outcome was also evaluated in our subset of vulnerable patients. Key secondary outcomes included the impact of our intervention on our global rank end point at 30 days and changes in the KCCQ and Dutch Heart Failure Knowledge Scale (DHFKS) scores at 30 and 90 days. We used means and medians for simple descriptive statistics, and the proportional hazards model for the global ranking outcome. Our models adjusted for traditional covariates of HF severity, including age, sex, systolic BP, prior ejection fraction (EF) (moderate/severe vs normal), and estimated glomerular filtration rate.
From October 28, 2015, to September 5, 2019, we randomly assigned 491 patients at 15 sites. Of these 491 patients, 245 were allocated to structured usual care, and 246 were allocated to our intervention arm. The overall median age was 63 years, 63% were African American, 36% were female, and 40% had a normal prior EF. Comorbidities such as diabetes, hypertension, and chronic kidney disease were prevalent and well balanced between the 2 arms. Our adjusted analysis for the global rank primary outcome showed that patients in the intervention arm were 10% less likely (hazard ratio [HR], 0.89; 95% CI, 0.73-1.10; P = .28) to have a worse global ranking over 90 days compared with patients in the usual care arm. Compared with patients in the structured usual care arm, patients in the intervention arm had a 4% (95% CI, ‒0.04 to 0.13; P = .29) overall lower rate of CV death and HF-related events over 90 days. The adjusted model for 90-day CV death and HF events found a 22% reduction in events in the intervention arm compared with the usual care arm (HR, 0.78; 95% CI, 0.57-1.06; P = .11). Similar differences were seen between the intervention arm and usual care arm among our vulnerable population. For our key secondary end point, compared with patients in the usual care arm, patients in the intervention arm were significantly less likely to have a worse global ranking over 30 days (HR, 0.80; 95% CI, 0.64-0.99; P = .04).
Median 30-day changes from baseline to 30 days in KCCQ score were 9.5 points in the intervention arm and 5.7 in the usual care arm (P = .05). Median 90-day change in KCCQ score was 10.9 points in the intervention arm and 9.4 points in the structured usual care arm (P = .75). Based on our adjusted model, 30-day changes in KCCQ score for patients in the intervention arm were 5.49 points higher (95% CI, 1.25-9.72; P = .01) than for patients in the structured usual care arm. There were no significant differences in KCCQ score at 90 days (β = 2.7; P = .25). We used a linear regression model, with changes in the DHFKS score as the outcome while adjusting for the baseline DHFKS score. The adjusted difference in DHFKS score between the intervention arm and the structured usual care arm was 1.22 points (95% CI, 0.34-2.09; P = .007) at 30 days and 1.94 points (95% CI, 1.02-2.87; P < .001) at 90 days.
We successfully completed the first randomized study of patients with AHF discharged from the ED and collected 30-day and 90-day event rates. There were no significant differences between arms in our primary 90-day outcome. At 30 days, our intervention resulted in a significant improvement in our primary global rank outcome. Importantly, as a result of our intervention, significant differences in patient-centered outcomes, such as KCCQ score and HF knowledge, were seen at 30 and 90 days. A variety of health care providers successfully delivered our intervention strategy, suggesting that this strategy would be readily amenable to rapid dissemination and implementation.
There are several limitations of this work. First, we had projected CV death, hospital admission, and ED revisit event rates of 62% in the usual care arm2-5 but experienced only a 36% rate, drastically limiting our power to detect differences between usual care and the intervention. Second, participant withdrawal in the intervention arm was greater than in the usual care arm, which suggests that certain patients may be more amenable to self-care coaching and that our results may be most applicable to this group. Third, our overall accrual rate was slower than expected, resulting in an extension of study duration; a discussion with PCORI resulted in a target sample size adjustment and a change in our primary outcome. Temporal changes in admission patterns at certain sites were largely responsible for this occurrence. Finally, the consent rate was only 56%, suggesting that patients may still be hesitant to allow virtual or in-person home visits. The receptivity to telehealth may change as a result of the COVID-19 virus pandemic.
Heart failure (HF) is common, with 6.2 million adults affected in the United States; approximately 915 000 new cases are diagnosed each year.6 HF is characterized by frequent hospitalizations, with 83% of patients with HF admitted at least once and 54% admitted ≥3 times following diagnosis.7 In the population receiving Medicare, HF is the leading hospital discharge diagnosis.6 Mortality rates range from 4% to 10% in the inpatient setting and approach 40% at 1 year posthospitalization.6,8,9 Despite a relative reduction in the hospitalization rate for HF, the actual number of HF hospitalizations remains >1 million annually, a figure that is expected to worsen significantly with the aging US population and the growing prevalence of HF. More than 80% of patients hospitalized for acute HF (AHF) are initially seen in the emergency department (ED),10 but not all those seen in the ED for AHF need to be admitted. Because disposition decisions rest largely with ED providers, the ED continues to play a significant role in the management of patients with HF and in avoiding unnecessary hospitalizations. Identifying safe transition and discharge strategies is a critical unmet need in this cohort.
Importantly, the ED also serves as the safety net for AHF care and the primary provider for uninsured patients. Over the past 5 years, >50% of AHF care occurred at safety-net hospitals. Our analysis of the Nationwide Emergency Department Sample database suggests that patients with AHF who do not have insurance (Figure 1) were significantly more likely to be discharged from the ED than those with insurance (15.4% vs 25.4%; P < .001), even after adjusting for potential confounders (adjusted odds ratio [OR], 0.64; 95% CI, 0.60-0.68; P < .001).10 However, when admitted, uninsured patients were more likely (21.6% vs 12.6%; P < .001) to undergo diagnostic tests or procedures, such as echocardiography, cardiac or pulmonary artery catheterization, pacemaker or cardioverter/defibrillator implantations, or stress testing. Further, African American patients were more likely to present to the ED for AHF instead of being directly admitted to the hospital.11 In summary, the ED is the primary access point for minority populations and for health care interaction among those who lack access to primary care.
Hospital Admission for AHF by Insurance Status.
To optimize care and reduce ED and hospital revisits, there has been significant emphasis on improving transitions at the time of hospital discharge for patients with HF. Unfortunately, the results of those programs suggest that they may not be associated with improved quality of care. Although these programs have decreased readmissions, concern has been raised that mortality increased over the same period.12,13 Further, efforts to improve transitions in care have been almost exclusively directed at hospitalized patients; individuals with AHF who are discharged from the ED miss the benefits of transitional care initiatives. Ensuring optimal transitions of care for patients with AHF discharged from the ED is a critical unmet need. Data show that patients with AHF discharged from the ED receive suboptimal guideline-directed medical therapy (GDMT), suggesting that interventions to improve AHF transitions are needed in the ED setting.14 This is particularly true for patients with poor access to health care, many of whom have characteristics of vulnerability. By default, the ED is often the sole or primary provider of AHF care for these patients.10 Event rates after ED discharge have been reported to be high, and a disparity exists between vulnerable and nonvulnerable patients.
Patients hospitalized for HF are among those most frequently readmitted within 30 days; they have up to a 25% risk of readmission within a month and a 50% risk within 6 months of discharge.1 Previous studies suggest that patients with AHF discharged from the ED have 20% to 30% higher readmission rates than those discharged from the inpatient setting.2-4 Theoretically, ED patients who are hospitalized are more ill than those who are deemed safe for ED discharge. This paradoxical finding of higher readmission risk among discharged ED patients further highlights the need for structured discharge planning. Recent studies suggest that there are diverging risks between 30-day mortality and readmission,15,16 implying that this readmission paradox may be explained by nonphysiologic variables and postdischarge management rather than by inpatient treatment and testing.17 Moreover, a lack of completion of follow-up visits after ED discharge has been associated with increased adverse events (AEs) and is higher in those with low education.4,18-20
Patients with HF who effectively engage in self-care interventions have lower hospitalization rates and mortality risk,21-23 better quality of life (QOL), higher functional status, and decreased health care costs.24,25 A review of 3166 patients with HF involved in various self-care interventions found that most reported significantly higher levels of HF knowledge and self-care behaviors, including regular weighing, medication adherence, and dietary sodium restriction. These are all behaviors we aimed to improve with our intervention.26 Many ED patients with AHF have characteristics of vulnerability, putting them at increased risk of self-care barriers. For example, poor health literacy may lead to problems taking medication and reading HF instructional information about the importance of recognizing signs and symptoms of worsening HF. The preponderance of evidence suggests that HF self-care interventions are beneficial in stable outpatients.27 Whether similar interventions will work for vulnerable patients with AHF slated for discharge from the ED, many of whom have significant barriers to self-care, is unknown.3,10,28 We have pilot tested self-care interventions that include disease and dietary education, teach-back, and facilitated access to medication prescriptions and follow-up care designed to specifically bridge this evidence gap.29-31
Patients with AHF discharged from the ED afford a unique opportunity to test ED-based transitional care interventions to combat self-care barriers. Because these barriers disproportionally affect vulnerable populations, including those with low health literacy, low socioeconomic status (SES), and no health insurance, as well as minority populations, we hypothesized our interventions would have greater impact in this cohort and thereby reduce disparities in ED revisits and hospitalizations. A carefully designed randomized trial comparing usual care with our transition nurse-coordinator (TNC) strategy (Figure 2) in vulnerable patients with AHF discharged from the ED would (1) bridge the evidence gaps that exist in current HF guidelines and provide prospective, randomized trial data to support important guideline revisions in future versions that address many of these critical issues; and (2) determine the impact of this strategy on important patient-centered outcomes, including patient satisfaction, HF knowledge, and HF QOL. To achieve this outcome, we proposed the following specific aims.
Randomization and Patient Flow.
To estimate the reduction in disparities in time to first ED/clinic revisit or hospital admission for AHF or cardiovascular (CV) death in vulnerable patients associated with our TNC strategy.
We hypothesized our TNC intervention would mitigate disparities in this composite outcome in vulnerable patients. We approached patients with AHF who were to be discharged from the ED. Our ED data suggest 70% of these patients would represent vulnerable populations (minority groups, low SES, and low health literacy).10,32,33 This approach provided adequate power to (1) determine the overall impact of our strategy regardless of characteristics of vulnerability, (2) explore individual characteristics of vulnerability associated with the highest (and lowest) improvements from our intervention, and 3) assess the reduction in disparities in our primary outcome between those with and without characteristics of vulnerability.
To (1) compare patient satisfaction, HF knowledge, and QOL between usual care and our TNC strategy in 3 focused analyses between those with and without characteristics of vulnerability; (2) explore individual characteristics of vulnerability associated with the highest (and lowest) improvements from our intervention; and (3) determine the differences between arms in all patients with AHF (independent of characteristics of vulnerability) discharged from the ED.
We hypothesized that our intervention would mitigate disparities in patient satisfaction, HF knowledge, and QOL. We would have adequate power to detect a 5-point difference in the Kansas City Cardiomyopathy Questionnaire (KCCQ) score assessing HF health status.
The groundwork was laid for the patient-engagement elements of the trial by a group of patients, caregivers, stakeholders, and advocates called the Citizen Scientist Task Force, commissioned by the American Heart Association's (AHA's) Healthcare Quality and Advocacy department to uphold patient-centeredness. Stakeholders include policy makers, administrators, and lay people. We aimed to incorporate the voices of patients at different stages in their journeys and their experience with the health care system. Therefore, the patient-engagement pyramid was the first step in designing this study (Figure 3). The multitier levels of engagement enable patients to participate at different points in their health journeys. The design was a collaborative effort among AHA staff members; clinical volunteers; and patients, caregivers, and stakeholders. They felt that the design best represented the ways we aimed to involve patients in the GUIDED-HF study.
Multitier Levels of Stakeholder Engagement.
The bottom tier of the pyramid is tailored for the passive consumer, someone who is interested in CV issues but perhaps not plugged into a health system or advocacy group. These patients may be earlier along in the health continuum. This audience may include patients, caregivers, or stakeholders from the medical community or advocacy groups. We wanted to represent the views of these individuals, and our study design created a way to incorporate their voices. Using the AHA digital channels on Facebook and Twitter, we hosted a “Highlight on HF” week. Results were reported on a campaign scorecard, enabling the team to assess how HF affects the public and which priorities were highest among followers. Conversations about HF between the AHA and the public resulted in the following:
The response to these questions was extremely robust, surpassing the participation that the AHA received with routine posts within the same period. The unusually high number of commenters (n = 297; 591% above monthly mean) suggested HF is a topic that elicits strong feedback and a desire among the public to engage further. The unusually high number of text comments generated by this conversation made it the top-volume brand post of the month in October 2014 by the AHA National Center. In other words, in this month, AHF was the number 1 topic of interest.
The Community Engagement Studio (CES) creates a safe haven and democratic space for patients and community experts to review and provide immediate feedback to the investigator about the patients' priorities given their unique backgrounds, experiences, and values. Vanderbilt University Medical Center (VUMC) has extensive experience using CES to obtain granular, patient-level feedback to inform study design and implementation.34-37 This is a valuable way to invite individuals who may not have internet access or high literacy to have meaningful, important roles on the team. The GUIDED-HF team at VUMC hosted a CES and included patient co-investigators on study team calls to obtain their input. The CES elevates community members as experts—people with extensive knowledge about the community of interest—and recognizes that these researchers offer unique expertise gained through personal experience. We made specific changes to our proposal based on feedback from the CES. When asked about the impact of HF on her life, a patient from our CES at VUMC said, “I want to do everything possible to understand HF, control my HF symptoms and not go back to the hospital; I am still paying for my 4-day stay from 2011.” As a result of this input, we incorporated out-of-pocket expenses into our data collection as a secondary end point.
Data from the first layers helped the research team focus on the primary and secondary aims of the study. We invited members of the Citizen Scientist Task Force to help write and direct the study; we also provided a chance for them to review the proposal iteratively. The trial had 2 named co-investigators, each with different backgrounds and priorities as well as discrete tasks best fit to their talents. Our citizen-scientists were 1 patient (Chad Robicheaux: congenital heart disease, longtime AHA volunteer) and 1 HF caregiver (Cheryl Walsh). They are part of the Steering Committee, which continues to meet regularly both by phone and webinar. Further, our citizen-scientists were co-authors on 3 manuscripts throughout the study.38-40 These manuscripts required in-person meetings as well as conference calls and numerous electronic reviews of the manuscript draft. One co-investigator, for example, would rather use texting and phone conversation to discuss aspects of the study and provide feedback, so Sean P. Collins, MD, MSc (principal investigator [PI]), called her personally when she was unable to provide feedback via track changes in the traditional format. Doing so ensured that her views and suggestions were not left out.
The study team continued to engage with the citizen-scientists on an ongoing basis via the monthly and bimonthly Steering Committee conference calls. These calls continued after enrollment had ended and during draft final research report and manuscript preparation. Combined with the intense engagement during preparation, submission, and revision of the 3 manuscripts, our study has benefited from tremendous citizen-scientist engagement. This engagement continued during preparation of the primary and secondary manuscripts from the study as well as dissemination of study results via nontraditional channels to ensure patient and caregiver awareness.
Our plans for sharing the data and resources developed as part of this project are described below. Briefly, we will comply with the PCORI 2018 Policy for Data Management and Data Sharing, available at https://www.pcori.org/about-us/governance/policy-data-management-and-data-sharing. Procedures are planned for the preparation, management, and sharing of the full data package, including deidentified analyzable data set, full protocol, metadata, data dictionary, full statistical analysis plan, and analytic code. The trial is registered on ClincialTrials.gov; the registration is updated annually and provides the current status of the trial, including publications. The full data package will be maintained at VUMC for at least 7 years following acceptance by PCORI of this Final Research Report (ie, through 2027).
We expected all study co-investigators, both our patients and professionals, to help make key decisions about disseminating findings in traditional and nontraditional channels. The joint team worked on publications for peer review, abstracts, and materials for conferences. Likewise, we will put equal focus on nontraditional channels, such as the AHA's 4.2 million followers on social media sites such as Facebook, Twitter, YouTube, LinkedIn, Pinterest, Instagram, and blogs. Other nontraditional media the AHA will assist with include a consumer-facing newsletter and other AHA websites, such as the AHA Support Network.
The GUIDED-HF study team fully supports public dissemination of the results and public access to trial data. Assuming that no restrictions are imposed by institutional policies, the local IRBs, or local, state, or federal laws and regulations (including the HIPAA Privacy Rule), we will follow the February 2003 US Department of Health and Human Services data policy (revised June 27, 2005). We will share deidentified data in an analyzable data set in accordance with the HIPAA Privacy Rule [45 Code of Federal Regulations § 164.514(b)]: https://www.govinfo.gov/content/pkg/CFR-2017-title45-vol1/pdf/CFR-2017-title45-vol1-sec164-514.pdf.
As a general policy, all data were deidentified (ie, stripped of all protected health information in compliance with the HIPAA Privacy Rule). Specifically, we removed the following elements from the data set: (1) names; (2) geographic information, including city, state, and zip code (location codes were used instead); (3) elements of dates, such as those for birth, hospital admission and discharge, and death; (4) telephone and fax numbers; (5) email addresses; (6) Social Security numbers; (7) medical record and prescription numbers; (8) health plan beneficiary numbers; (9) account numbers; (10) certificate or license numbers; (11) any vehicle identifier or serial number, including license plate number; (12) any device identifier or serial number; (13) uniform resource locators (URLs); (14) Internet Protocol (IP) addresses; (15) any biometric identifiers, including fingerprints or voice prints; (16) full-face photographic images or any comparable images; and (17) any other unique identifying numbers, characteristics, or codes consisting of any segments of the previously listed identifiers, including dates (eg, visit date, admission date, use date, birthdate). In addition, we expressed all dates as the number of days since enrollment and added a variable that stores just the year of enrollment. For information about geographical location (eg, county) that would jeopardize identity because of a small number of patients within a particular vicinity, we used aggregation algorithms. If there were too few subjects with specific characteristics at a site, then the site was combined with another site. Situations may occur for which no predefined solution exists. In these cases, the Biostatistics Data Management Center (BDMC) would work with the GUIDED-HF Executive Committee to define an acceptable reporting solution.
As the final analysis nears completion and investigators contact us, we will provide a description of the regulatory and contracting approvals required, the role of the original study team in reviewing the data use request, and stipulations for publishing results. Requests for data will be presented to the Publications and Presentations Committee for review, whose members will determine data release. Reasons for not allowing access will generally be limited to when the data are insufficient to support the new study question. In situations where multiple investigators request access for similar study questions, we will coordinate among investigators to minimize duplication of effort. If a third-party data request is received before these data are deposited in a repository, we will notify PCORI to seek potential funds for the transfer of data to a PCORI-designated repository. If the data are already deposited in a repository, such requests will be directed primarily to the repository.
Our Clinical Coordinating Center (CCC) and BDMC team members fully support the public dissemination of the results and public access to our data as delineated above. We further support dissemination of the tools and workflows that will be developed because of our work. Regarding dissemination of trial results, we registered the trial on ClincialTrials.gov before enrolling the first patient. We updated the registration annually and posted the main results on ClinicalTrials.gov. The CCC is responsible for posting the information, and the BDMC will provide the data and results needed. Our informed consent document included an explicit statement advising participants that the study and its results will be made available on ClinicalTrials.gov. We note that VUMC has strict policies and procedures in place to ensure that registration and reporting of trial results occurs as required.
We submitted and published a manuscript38 describing the design within 6 months of the first patient being enrolled, and we submitted a manuscript describing the trial results within 6 to 12 months of the completion of data collection. Beyond publication, the study team presented the design, approach, and findings at national meetings and will post information on the study website and via institutional media relations offices. We expect our approach to data coordination and project management to be both replicable and scalable at low cost. We will document our procedures, including the code-supporting system interfaces, and make this information publicly available.
Our comparative effectiveness research study was designed initially to compare the improvement in HF disparities in vulnerable populations; however, with PCORI's approval, we modified the proposal to look at our primary outcome (ie, the impact of our intervention on our global rank outcome at 90 days) in the overall population. As our secondary outcome, we evaluated the improvement in disparities that our intervention provided in vulnerable populations. This modification was needed because the study confronted unanticipated recruitment obstacles that prevented us from enrolling 700 patients as originally proposed in the application to PCORI. We have outlined the protocol changes below and provide the remainder of the text in this section based on those changes.
Our proposal was a multicenter, stratified, randomized, 1:1 trial comparing standardized ED discharge (referred to as the structured usual care arm) vs a tailored discharge plan (referred to as the intervention arm) using a TNC and GUIDED-HF recommendations in patients with AHF discharged from the ED. We stratified by site and randomized at the patient level. The randomization module for each site was designed using Research Electronic Data Capture (REDCap) software to ensure patient-level randomization via blocks of variable size, which ensured balanced randomization at each site.
Patients were identified during an ED evaluation for AHF. The information collected included their stay in the ED and ED-based observation as they received continued evaluation and treatment for AHF. All study sites had trained ED personnel to screen and approach potential candidates. These sites were academic medical centers serving a largely urban population as well as 1 Veterans Affairs ED.
Patients who presented to the ED with signs or symptoms of AHF were screened for possible inclusion in the study (Table 1). Entry criteria for ED patients included those (1) in whom AHF was diagnosed; (2) whom ED physicians intended to discharge either directly from the ED or after a period of observation; and (3) who were low risk and would not have difficulties complying with the protocol because of psychiatric disease, dementia, or distance from domicile to the enrolling institution, which would make the TNC home visit problematic.
Study Inclusion and Exclusion Criteria.
Further definition and the rationale for selected inclusion and exclusion criteria are as follows:
Summary of GUIDED-HF Interventions.
Our primary end point was a composite global rank outcome based on a prespecified hierarchical ranking system using the following end points within 90 days from the initial ED visit: (1) CV death; (2) a hospital admission, an ED revisit for AHF, or a clinic visit for AHF during which intravenous (IV) diuretics were administered; and (3) change in KCCQ score. Patients were assigned lower ranks (ie, a smaller number) if they had worse outcomes. Specifically, patients who died within 90 days of ED discharge were included in the first hierarchy rank based on time to death, where patients who died the earliest ranked the lowest and patients who died the latest ranked the highest.48,49 If the patient did not have an outcome and fulfill a rank, we continued assigning ranks in the second hierarchy, which included patients alive at 90 days after ED discharge but with any HF-related events specified within 90 days. Patients were assigned a rank based on time to the earliest occurrence of an AHF-related AE. The remaining patients free of AHF-related events over 90 days were included in the third hierarchy. Following the highest rank in the second hierarchy, they were ranked based on 90-day changes in KCCQ score, where the patient with the smallest change was ranked the lowest and the patient with the largest change was ranked the highest. Patients lost to follow-up within 90 days were censored in the first round. Patients with missing 90-day change in KCCQ score and not ranked in the first 2 rounds (ie, those without any HF-related AEs at 90 days) were censored at the highest rank in the second round.
Our secondary end points included our primary outcome at 30 days and our HF components of the primary outcome (CV death, ED revisit and hospital admission), patient satisfaction, HF knowledge, and KCCQ score changes over 30 and 90 days. Our safety end point was analyzed separately from our primary outcome and included all-cause ED visits, hospital admissions, and death within 90 days. This end point was selected to ensure that our intervention did not result in unintended non-AHF events because of patients' increased focus on their HF care:
The primary objective of the Clinical Events Committee (CEC) was to manage the process of coordinating the independent review and adjudication of suspected end points. Primary and specified secondary end point events to be adjudicated required sites to complete specific event case report forms (CRFs) and provide source documents. Two emergency physicians (Jennifer Martindale and Brian Hiestand) and 1 cardiologist (JoAnn Lindenfeld) made up the CEC and reviewed the CRFs. Dr Lindenfeld served as director of the CEC. Systems were developed so that the adjudicated data were available, as specified and required by trial protocols, to the project statisticians and data and safety monitoring board (DSMB) for interim analyses. Care was taken to blind the reviewers to any information that could identify the patient or reveal the randomized allocation.
Study personnel collected data during ED evaluation and treatment. At ED discharge, patients entered the follow-up period. The data collected are indicated in Table 3. We specifically designed the data collection to be efficient (<40 minutes of patient interview time) and specific to study-relevant HF measures and outcomes because scalability and implementation after the study is complete will depend on allowing flexibility in the ED personnel (registered nurses [RNs], case managers, paramedics, social workers) who could perform data collection and the willingness of patients to participate. The PI worked closely with the coordinators and data managers to create, manage, and review all electronic data capture forms. Aim 1 evaluated the impact of our tailored intervention in the overall cohort and in reducing disparities in time to HF events, CV mortality, and KCCQ score in vulnerable patients with AHF discharged from the ED. We approached all patients with AHF discharged from the ED for inclusion in the study. However, to ensure that we could evaluate aim 1 (90-day global rank between study arms in the overall cohort), we also captured whether patients had vulnerable characteristics, as defined previously, and their address or zip code of residence. To determine the primary and secondary outcomes, assessors who were blinded to intervention arm called patients at 30 and 90 days after discharge from the ED.
Data Collection Time Points, Role in Analysis, and Time Required With Patient.
The analysis in aim 2 focused on the impact of our interventions on our patient-centered outcomes. We collected baseline data regarding the KCCQ score; HF knowledge, as measured by the DHFKS; patient satisfaction; medical comorbidities; medications; hospitalizations and ED visits in the previous 6 months; New York Heart Association class; the 7-item Adherence to Refills and Medications Scale50; and health literacy (BHLS).51 The 12-item KCCQ52 Summary Score includes subdomains of physical limitations, symptom frequency, QOL, and social limitations. The DHFKS score was developed to assess patient comprehension of HF; this 15-item questionnaire investigates a patient's understanding of HF treatment and symptom recognition. A score >10 has been associated with high HF knowledge.53
The site PIs worked closely with the coordinators and data managers to create, manage, and review all electronic data capture forms. Research personnel prospectively collected information for direct data entry into a secure internet-based electronic CRF, which was securely uploaded to REDCap. The BDMC performed random quarterly data freezes with comprehensive data quality checks. Queries with values out of range, dates with invalid temporal orders, or missing data were reviewed by the CCC to identify methods to increase data capture. Thorough screening and patient understanding were crucial to minimizing dropout. When dropout occurred, we determined (1) the reason for dropout and by whom, and (2) which types of participation the dropout affected and whether key outcomes could still be collected.
Frequency of missingness was calculated by intervention, overall and by site, for all baseline patient characteristics, KCCQ score, DHFKS score, and PROMIS score at each follow-up visit. Further exploration of the reasons for and patterns of missingness was considered for variables seemingly not missing at random, based on partial missingness for a time point vs complete missingness. In regression analysis, missing covariate information was imputed using predictive mean matching under the assumption of missingness at random.54
All progress report data were summarized and reported to PCORI per the CONSORT 2010 guide (http://www.consort-statement.org/consort-2010). We evaluated internal validity by including information about patient enrollment within each site and treatment arm; losses and exclusions after randomization together with reasons; and baseline demographic and clinical characteristics by site and overall for each arm. External validity was evaluated by comparing patient characteristics, blinded outcomes, and KCCQ scores with historical HF data from our prior studies. We also reported on patient and stakeholder engagement milestones.
Minority was defined as African American and White Hispanic. Low health literacy was defined as BHLS score <9. Low SES was defined as Area Deprivation Index (ADI) national percentile ranking >85.55
The global ranking outcome was a censored ordinal outcome, with higher values indicating better outcomes. Therefore, the same analytical approach used for a survival-type outcome was adopted. We used the proportional hazards model for the global ranking outcome, which is a rank-based semiparametric approach. To evaluate the overall intervention effect (regardless of characteristics of vulnerability), we used the following proportional hazard model: , where λ0 (t) was the unspecified baseline hazard function; X was an indicator variable denoting the intervention assignment (X = 1 for GUIDED-HF; X = 0 for usual care); V denoted characteristics of vulnerability (minority group, continuous BHLS score, and ADI national rank); and Z was a k-dimensional vector of participant covariates at baseline, including age, sex, SBP at ED visit, prior EF (moderate/severe vs normal), and estimated glomerular filtration rate. The corresponding parameter of interest, β1, was the log hazard ratio (HR) between the intervention arm and structured usual care arm, adjusting for the aforementioned covariates. If the intervention was effective, the corresponding log HR is < 0 (β1 < 0). Missing covariate information was imputed with multiple imputation techniques, such as predictive mean matching.54 The proportional hazards assumption was checked.56
The impact of our intervention also depended on a participant's propensity for a disparity in outcome because of their associated characteristics of vulnerability. To explore these differences in intervention effects, we identified and adjusted for this effect. We specifically explored minority group, low health literacy, and low SES in this manner. In separate analyses, we tested whether the intervention was effective within vulnerable populations and each subset of vulnerable patients (eg, African American race, low health literacy, and low SES) using the same model as in aim 1.1. A negative value for the log HR, β1, indicated that the intervention was effective in improving the global ranking within the subset of patients.
To assess the effect of the TNC strategy on narrowing the disparities gap, we included the interaction term between the intervention and the binary indicator of vulnerable population (V*) in the model . The corresponding parameter of interest, β3, was the difference in log HR for the intervention arm between vulnerable and nonvulnerable populations. If the intervention arm was effective in reducing disparities (ie, if it reduced the difference in the time-to-event rate between the 2 populations), the HR would be less than unity (exp (β3) < 1) and the log HR would be < 0 (β3) < 0). If we found heterogeneity of treatment effect (HTE), we determined the vulnerable characteristic components associated with the highest (and lowest) improvement. The analytic strategy was similar to that in aim 1.1, where all indicator variables of vulnerable characteristic components were simultaneously included in the Cox model as well as their interactions with the intervention. Post hoc analysis was conducted to compare HRs of interaction terms. The component with the lowest (and highest) HR of the interaction term was associated with the greatest (and smallest) reduction in disparity from the intervention.
Our sensitivity analysis evaluated those who completed the 48-hour home visit and had scheduled outpatient follow-up.
The original planned sample size of 700 participants equally distributed across the 2 intervention arms and stratified by site and vulnerable and nonvulnerable populations would provide excellent precision to address aim 1 and aim 2 using a type I error rate of 5% in a 2-sided test. The power calculation was based on standard methods for a proportional hazards model.57 Based on our prior ED cohort studies and the existing literature,2-4,58 we expected, in the structured usual care arm, event rates of 26% and 62% and censoring rates of 2% and 4% at 30 and 90 days, respectively. For aim 1.1, evaluating the overall intervention effect, the proposed sample size at 90-day follow-up would offer 80% power to detect an HR of 0.76 = 1/1.32 for the intervention, which corresponds to an absolute event rate reduction of 10%.
For aim 1.2, the power consideration was based on the intervention effect within each subset of vulnerable patients. We assumed that 70% of the discharged ED patients would have a characteristic of vulnerability. The estimate of this proportion was felt to be conservative because our prior studies showed that the proportion of people from vulnerable populations discharged from the ED who were not insured, from a minority group, had low income, and had low literacy were 40%, 50%, 40% to 50%, and 40%, respectively.58-60 The proposed sample size (African American: 350 = 700 × 50%; low SES: 315 = 700 × 45%; low health literacy/uninsured: 280 = 700 × 40%) at 90-day follow-up would provide 80% power to detect an HR of 0.68 = 1/1.47, 0.66 = 1/1.51, and 0.65 = 1/1.54 for the intervention effect within each subgroup.
For aim 1.3, testing the intervention effects in reducing disparities, using the 90-day information, the proposed sample size would provide 80% power to detect an HR of 0.55 = 1/1.82 for the interaction term, which corresponds to a 1 − 55% = 45% reduction in disparity in relative risk between the vulnerable and nonvulnerable populations. Longer follow-up offers more precision because of the larger number of observed events. Note that the above sample size projections did not account for adjustments from covariates. Accordingly, it was reasonable to expect the projections to be conservative, because adjusting for important covariates tends to increase power.
The updated planned sample size of 540 participants equally distributed across the 2 study arms stratified by site and vulnerable and nonvulnerable populations would provide excellent precision to address aim 1 and aim 2 using a type I error rate of 5% in a 2-sided test. Based on our prior ED cohort studies and the existing literature,2-4,58 we expected, in the structured usual care arm, an event rate of 62% and a censoring rate of 4% at 90 days. Although the 30-day end point is more temporally related to the ED visit, the study team felt that the 90-day end point would provide a sufficient number of events for our primary outcome.
For aim 1.2 (testing intervention effects within vulnerable populations) and aim 1.3 (testing intervention effects in reducing disparities), we assumed that 75% of patients discharged from the ED would have a characteristic of vulnerability. The estimate of this proportion was conservative because our prior studies showed that the proportion of people from vulnerable populations discharged from the ED who were from a minority group, had low income, or had low health literacy were 50%, 40% to 50%, and 40%, respectively; our preliminary data suggest that >70% are from minority groups or have low health literacy.58-60 Subsequent determinations of low SES by geocoding will only increase this proportion. The power calculation was based on the original planned 90-day, HF-related AE survival outcome using standard methods for a proportional hazards model57 given a prespecified HR corresponding to a 15% and 20% relative rate reduction, respectively. The rate reduction of 15% is clinically meaningful relative to other interventions in both acute and chronic HF.
For aim 1.1 (the primary aim), evaluating the overall intervention effect, the power to detect the HRs corresponding to a 15% and 20% relative rate reduction was 63% and 87%, respectively. For aim 1.2, the power consideration is based on the intervention effect within the vulnerable population of patients. The HR corresponding to a relative rate reduction of 15% and 20% was 0.77 and 0.71, respectively. For aim 1.3, the power to detect an HR of 0.6 and 0.5 was 51% and 77%, respectively. Assuming relative rate reductions of 15% and 20% in the nonvulnerable population, these HRs correspond to the difference in absolute rate reduction between nonvulnerable and vulnerable populations of 26% and 33%, respectively. Longer follow-up offers greater precision because of the larger number of observed events. Note that the above sample size projections did not consider adjustments from covariates. Accordingly, we expected the projections to be conservative, because adjusting for important covariates tends to increase power.
KCCQ scores have been found to predict clinically meaningful changes in health among patients with HF. Changes in the KCCQ score of ≥5 points are considered clinically important.61-63 Further, the KCCQ has been studied in ED patients by members of our team.64 We checked spaghetti plots of KCCQ score by intervention to determine the appropriate statistical modeling. If the trajectory was not the same between 30-day and 90-day scores, we separately modeled 30-day and 90-day changes in KCCQ score while adjusting for baseline KCCQ score in addition to the prespecified baseline patient characteristics as in aim 1. If, however, the trajectory appeared linear, a linear mixed model was used to model KCCQ scores and examine the effect of the intervention arm on the reduction in disparities between vulnerable and nonvulnerable populations over time using the interaction between intervention and follow-up time. Random effects included random intercept for patients and random slope for follow-up time. Regardless of the choice of the modeling strategy, in the same spirit as the analysis plan for aim 1, we evaluated KCCQ score in the overall population first. For the intervention effect within the vulnerable population, we evaluated KCCQ score within the vulnerable population and each subset. Last, to assess the HTE, we included an interaction between the intervention effect and the vulnerable population. The regression model was adjusted for the same factors identified in the analysis of aim 1. A similar analytic strategy was considered for the DHFKS score and PROMIS score. A proportional odds model was used for patient satisfaction given that it is a 5-point Likert scale of ordinal outcomes. If our primary outcome was neutral and patients either improved or maintained the KCCQ score in the intervention arm while the control arm worsened, this would be an important patient-centered result. Therefore, to investigate this, we plotted KCCQ scores over time for each individual patient using a locally weighted regression-fitted line to represent the average trend for each treatment arm (spaghetti plots).
From October 28, 2015, to September 5, 2019, we approached 7148 patients; 525 of the 933 eligible patients (56%) consented to participate in the study. We randomly assigned 491 patients at 15 academic medical centers. The median number of patients enrolled per site was 15 (interquartile range, 8.5-28). The range of site enrollments was 3 to 136. Of these 491 patients, 245 were allocated to structured usual care, and 246 were allocated to our intervention arm. Twelve patients withdrew from the study (1 in the structured usual care arm and 11 in the intervention arm), which left 244 participants in the structured usual care arm and 235 in the intervention arm who were available for follow-up (Figure 4). The overall median age was 63 years, 63% of participants were African American, 36% were female, and 40% had a normal prior EF. Comorbidities, such as diabetes, hypertension, and chronic kidney disease, were prevalent and well balanced between the 2 arms (Table 4). There was a slight imbalance in prior EF between the 2 arms. Intervention arm–specific procedures included completion of an in-person home visit by 89% of patients, and coaching calls in the first 30 days were completed in 80% of cases. In-person (78%) or telehealth home (11%) visits were completed for 89% of participants in the intervention arm. Coaching calls in the first 30 days were completed for 94% of participants, and 89% received coaching calls in the 30- to 90-day window.
CONSORT Diagram for GUIDED-HF Patient Enrollment.
Descriptive Statistics for Patient Demographic and Clinical Characteristics by Study Arm.
A large proportion of patients in the overall cohort fulfilled our definition of vulnerable. Of the 479 patients available for 90-day follow-up, 77% (368) fulfilled at least 1 of our criteria for vulnerability. The median BHLS score was 13 (range, 0-15) for the overall cohort and 13 and 12 within the structured usual care and intervention arms, respectively. The proportion of patients with low health literacy scores, defined as <9 on the BHLS, was 14% in the structured usual care arm and 11% in the intervention arm. The median ADI (range, 0-100) was 83 for the overall cohort and within each arm. Based on a cutoff of 85 for low SES, 47% of our total cohort had low SES. Death from CV disease was low in the overall cohort (0.4% at 90 days). Within 90 days of the index ED visit, AHF hospital admissions occurred in 28% of participants, ED revisits in 30%, and clinic visits with an IV diuretic administered in 2%.
Follow-up at 90 days occurred in 98.5% of patients (98.3% in the intervention arm and 98.8% in the structured usual care arm). There was no difference in the 90-day global rank primary end point between the 2 arms (HR, 0.89; 95% CI, 0.73-1.10; P = .28). An absolute 4% risk reduction was seen in 90-day AHF events (CV death, ED revisit or hospitalization for AHF, unscheduled clinic visit for AHF), which did not reach statistical significance (Table 5). Compared with patients in the structured usual care arm, patients in the intervention arm had a 1% absolute reduction in CV death, a 1% reduction in hospital readmission, a 6% reduction in ED revisits, and a 2% increase in clinic visits with IV diuretics. The reduction in AHF events stayed consistent throughout the study (36% structured usual care vs 32% intervention arm; P = .29) (Figure 5). The adjusted model for 90-day CV death and HF events found a 22% relative reduction in events in the intervention arm compared with the structured usual care arm (HR, 0.78; 95% CI, 0.57-1.06; P = .11). The median KCCQ score at 90 days was 63 in the intervention arm and 59 in the structured usual care arm (P = .23). The proportional hazards assumption was not violated for any analysis.
Ninety-Day HF-Related Events for the Overall Population by Study Arm.
Kaplan-Meier Curve for CV Death and HF Adverse Events Over 90 Days, Stratified by Study Arm.
Follow-up at 30 days occurred in 100% of patients. There was a statistically significant difference between study arms in our unadjusted (P = .05) and adjusted analysis for the 30-day global rank end point. In our adjusted analysis, patients in the intervention arm were 20% less likely (HR, 0.80; 95% CI, 0.64-0.99; P = .04) to have a worse global ranking over 30 days than those in the structured usual care arm. Compared with patients in the structured usual care arm, those in the intervention arm had a 4% overall lower 30-day rate of CV death and AHF events (18% vs 14%; P = .19).
Within the subpopulation that met our definition of vulnerable, there were similar statistically nonsignificant reductions in ED revisits, hospital readmission, and CV death when compared with the overall cohort (Table 6). Our adjusted analysis suggests that patients in the intervention arm were less likely (HR, 0.94; 95% CI, 0.74-1.19; P = .61) to have a worse global ranking over 90 days than were patients in the structured usual care arm. Similarly, those in the intervention arm were 17% less likely (HR, 0.83; 95% CI, 0.65-1.06; P = .13) to have a worse global ranking at 30 days than were those in the structured usual care arm. When evaluating the intervention effect within each characteristic of vulnerability subset (minority group, low health literacy, low SES), we found that none of the subsets had a significant intervention effect (P > .28).
Ninety-Day HF-Related Events for Vulnerable and Nonvulnerable Populations by Study Arm.
The KCCQ was completed at baseline and 30 days in 350 patients and at baseline and 90 days in 345 patients. It was completed in 78% of the intervention arm and 68% of the structured usual care arm at 30 days and in 80% of the intervention arm and 65% in the structured usual care arm at 90 days. The median 30-day score was 64 points in the intervention arm and 53 points in the control arm (P = .002). Median 30-day changes from baseline KCCQ score were 9.5 points in the intervention arm and 5.7 points in the structured usual care arm (P = .048) (Appendix Table A1; Figure 6). The mean 90-day score was 63 points in the intervention arm and 59 points in the control arm (P = .23). The median 90-day change in KCCQ score was 10.9 points in the intervention arm and 9.4 points in the structured usual care arm (P = .75) (Appendix Table A1; Figure 7). Based on the spaghetti plots (Figure 8), changes in KCCQ score were not linear over 90 days in the intervention arm, with KCCQ score increasing significantly within 30 days and leveling out from 30 to 90 days. In comparison, KCCQ score appeared to increase linearly over 90 days in the structured usual care arm. Therefore, we modeled the change in KCCQ score at the 2 time points separately using a linear regression model, with 30-day and 90-day changes in KCCQ score as the outcomes, while adjusting for baseline KCCQ score. Based on the adjusted analysis, 30-day changes in KCCQ score for patients in the intervention arm were 5.49 points higher (95% CI, 1.25-9.72; P = .01) than in patients in the structured usual care arm. There was no statistically significant difference in 90-day changes in KCCQ score (β = 2.7; P = .25) between arms. In the vulnerable population, we also saw borderline statistically significant differences between intervention and structured usual care arm KCCQ scores at 30 days but not at 90 days.
Box Plots for KCCQ Score at Baseline, 30 Days, and 30-Day Changes, by Study Arm.
Box Plots for KCCQ Score at Baseline, 90 Days, and 90-Day Changes, by Study Arm.
Spaghetti Plots for KCCQ Score Over 90 Days With LOWESS Curves, Stratified by Study Arm.
Research suggests that patients with a DHFKS score >10 (range, 0-15) have a high level of HF knowledge.53,66,67 The median HF knowledge scores at baseline were 11 and 10 points in the intervention and structured usual care arms, respectively. In the intervention arm, these values were unchanged at 30 and 90 days, while in the control arm they were 9 points at both 30 and 90 days (unadjusted between-arm comparison: P = .082 for 30-day changes, P = .006 for 90-day changes) (Appendix Table A1). We used linear regression with changes in DHFKS score as the outcome while adjusting for the baseline DHFKS score. The adjusted difference in DHFKS score between the intervention arm and the structured usual care arm was 1.22 points (95% CI, 0.34-2.09; P = .007) for 30-day changes and 1.94 points (95% CI, 1.02-2.87; P < .001) for 90-day changes; this implies that the intervention reduces the decline in DHFKS score at both 30 days and 90 days compared with structured usual care.
In the subset of vulnerable patients (Appendix Table A2), the median DHFKS scores were 11 points in the intervention arm and 10 points in the structured usual care arm at baseline. They were essentially unchanged (10 points and 11 points) at 30 and 90 days in the intervention arm, and decreased to 9 points and 8 points, respectively, in the structured usual care arm (unadjusted between-arm comparison: P = .12 for 30-day changes, P = .009 for 90-day changes). The adjusted difference in DHFKS scores between the intervention arm and structured usual care arm was 1.19 points (95% CI, 0.17-2.20; P = .02) for 30-day changes and 1.19 points (95% CI, 0.18-2.21; P = .02) for 90-day changes. Similarly, significant differences were observed among vulnerable populations in minority groups and in those with low SES. In the minority subset, the adjusted difference in DHFKS scores between the intervention arm and structured usual care arm was 1.21 points (95% CI, 0.12-2.30; P = .03) for 30-day changes and 1.59 points (95% CI, 0.46-2.73; P = .006) for 90-day changes. In the low-SES subset, the adjusted difference in DHFKS scores between the intervention arm and structured usual care arm was 1.67 points (95% CI, 0.43-2.90; P = .009) for 30-day changes and 1.76 points (95% CI, 0.44-3.08; P = .009) for 90-day changes. As in the overall cohort, this finding implies that the intervention reduces the decline in DHFKS score for the vulnerable patient subsets in minority groups and those with low SES at both 30 days and 90 days compared with the structured usual care arm.
Patient satisfaction was measured at the time of ED discharge using a 5-point Likert scale to determine each patient's satisfaction with the ED discharge process. This step was specifically incorporated because of patient feedback during our study design regarding variable previous ED experiences. The response “strongly agree” about being satisfied with the ED discharge process was endorsed by 56% of patients in the intervention arm and 48% of patients in the structured usual care arm. No statistically significant differences in satisfaction with ED visit were observed in the adjusted analysis for the overall cohort, vulnerable population, minority population, or population with low health literacy (P > .11). In the population with low SES, patients in the intervention arm were more likely to endorse a higher level of satisfaction than were patients in the structured usual care arm (OR, 2.31; 95% CI, 1.29-4.15; P = .005). No HTE was observed.
We started to collect baseline NIH PROMIS anxiety and depression scores, but as indicated in the “Protocol Changes” section, we amended this to collect only 30- and 90-day data. Linear regression analysis was used, with 30- and 90-day PROMIS scores as the outcome. No statistically significant differences between study arms were observed for the overall cohort, the vulnerable population, or its individual vulnerable subsets. No HTE was observed.
In our modified intention-to-treat analysis, we compared those in the intervention arm who were compliant with the protocol, defined as (1) completing the home visit and (2) scheduling a postdischarge appointment 7 days after discharge from the ED. Those in the intervention arm had a rate of 34% CV death and AHF events within 90 days, while those in the structured usual care arm had a rate of 36% CV death and HF events at the same time point. All results were consistent with our primary analysis.
Our safety outcomes were all-cause death and all-cause ED revisit and hospital admission. There were no unexpected increases in these events in the intervention arm: they occurred in 58% of participants in both arms.
We successfully completed the first randomized study of an intervention to improve outcomes in patients with AHF discharged from the ED and collected 30-day and 90-day event rates in 100% and 98.5% of cases, respectively. Our primary outcome, the 90-day global rank, was not significantly different between our intervention and control arms. Prior evaluations in similar cohorts were limited by retrospective record review methodology or relied on administrative claims data.4 We systematically screened, enrolled, and randomly assigned a diverse cohort of patients across 15 medical centers and followed patients over 90 days to identify important patient-related outcomes. Further, our TNC model could be delivered by several types of ED personnel, facilitating broad dissemination. The role of the TNC at our study sites was fulfilled by RNs, study assistants, and paramedics. All TNCs were trained under 1 common protocol for the home visits and coaching.
Although our intervention did not affect the primary study outcome, our study has 4 important secondary findings. First, at 30 days, we saw a similar difference in secondary outcomes, including differences in AHF events (4%), between treatment arms. We observed a clinically and statistically significant improvement in HF health status based on KCCQ score in the intervention arm. Combined with the 4% difference in AHF events, this finding resulted in a statistically significant (20%) difference between our 2 arms in both our unadjusted and adjusted secondary 30-day global rank end point of cardiac events. As an additional finding related to AHF events consistent with predictors of AEs from prior studies, SBP and baseline KCCQ score were also associated with 30- and 90-day events.68-70
Second, we found similar differences between study arms in our subset of vulnerable patients, who often use the ED as a safety net for health care; based on our prior experience, we projected that 75% of our patients would have characteristics of vulnerability. Indeed, 77% (368) of patients enrolled fulfilled our criteria for vulnerability. However, as with the entire study population, the intervention did not affect the primary study outcome among vulnerable patients. There was a statistically nonsignificant increase of 6% improvement in the global rank in patients in the intervention arm over 90 days compared with patients in the structured usual care arm.
Third, we report an in-depth evaluation of the impact of our intervention on another secondary outcome, HF health status, as measured by the KCCQ. This outcome includes the relative change and recovery of health status by study arm at both 30 and 90 days. Although this was reported by our group in a prior study, the majority of the patients in that study were admitted to the hospital.64 The findings from this study are the first in patients with AHF discharged from the ED. Our 30-day changes in KCCQ score were significantly higher (ie, better HF health status) in the intervention arm (median, 9.5) than in the structured usual care arm (median, 5.7), with an adjusted mean difference of 5.49 points. These results suggest statistically and clinically important differences in KCCQ score at 30 days.61,71 However, most of the HF health status benefit from our intervention occurred in the first 30 days, and there was no difference by study arm at 90 days.
Fourth, we report for the first time the impact of a self-care intervention strategy on HF knowledge over 30 and 90 days. Patients in both the intervention arm and the structured usual care arm had HF knowledge scores ≥10 at baseline, consistent with a high level of HF knowledge.53,66,67 Our intervention resulted in retention of higher HF knowledge, whereas knowledge decreased in those in the structured usual care arm over the subsequent 30 and 90 days. Our linear model suggested a significant difference in our adjusted analyses in HF knowledge between arms at both 30 and 90 days. We had similar findings in our subset of vulnerable patients. There were significant unadjusted and adjusted differences in the 30-day and 90-day changes in DHFKS scores between the intervention arm and the structured usual care arm. Similar significant differences were observed among vulnerable populations in minority groups and in those with low SES.
Our study had several imitations. First, while we had projected event rates of 62% in the structured usual care arm based on prior research,2-4,58 we experienced much lower event rates (36%) in this arm, limiting our power to detect significant differences. Although our overall event rate was lower than expected, we anticipated that our intervention would have a 15% to 20% relative rate reduction; we found a relative rate reduction of 10%. While our effect difference between the 2 arms did not reach statistical significance, our Kaplan-Meier curves suggest early separation of treatment arms. Second, our overall accrual rate was slower than expected, resulting in an extension of study duration. Although some sites met or exceeded anticipated enrollment, several sites did not meet enrollment expectations. Our study was designed to randomize at the patient level, stratified by site. Another option would have been to conduct a cluster randomized trial, but we would have had to ensure relatively balanced enrollment across multiple centers. Originally, we planned to conduct the study at 4 sites, making this design problematic as well. Cluster randomized trials often require a minimum of 8 to 10 sites, and it would have been difficult to conduct a cluster crossover trial with an open-label intervention.72 For this reason, we opted for the more rigorous patient-level, stratified randomization. However, this type of randomization introduces possible contamination of the structured usual care arm with self-care approaches from the intervention arm. We did monitor the structured usual care arm to ensure there were (1) no home visits performed and (2) no coaching calls conducted; however, other contamination could have occurred related to the initial ED encounter with study staff. There was also 27% missingness in the KCCQ scores, which could have affected the results. The missingness appears to be random and affected both arms equally, but we cannot completely exclude ascertainment bias as an explanation for our results. The success of our self-care intervention could be dependent on patient and caregiver comprehension of the intervention and their associated goals. We used teach-back to minimize this impact. Finally, patients who did not have a caregiver to assist with the program might have been at a disadvantage relative to those who had a team of caregivers to assist with implementation of the self-care program.
We successfully implemented a strategy of self-care and had high intervention fidelity and low loss to follow-up in patients with AHF discharged from the ED. Our intervention strategy was successfully delivered by a variety of health care providers, suggesting this strategy would be easily implemented by personnel already available in most EDs. The intervention did not have a statistically significant effect on our primary global rank outcome at 90 days. However, it did result in a significantly better global rank outcome at 30 days, a secondary outcome. Importantly, it caused significant and important clinical differences in patient-centered outcomes, such as improved HF health status (KCCQ score) and HF knowledge at 30 and 90 days.
We would like to acknowledge the hard work and dedication of all the study coordinators and PIs at our sites, the patients who participated in GUIDED-HF, and our CEC and DSMB.
Research reported in this report was funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (AD-1409-21656). Further information available at: https://www.pcori.org/research-results/2015/examining-whether-self-care-program-reduces-healthcare-use-and-improves-health
Collins SP, Storrow A, Liu D, et al. (2021). Examining Whether a Self-Care Program Reduces Healthcare Use and Improves Health among Patients with Acute Heart Failure—The Guided HF Study. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/04.2021.AD.140921656
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
