The impact of an adaptive patient-centered web application on medication optimization in patients with HFrEF: Rationale and design of the SmartHF trial

Abstract

Introduction

Heart failure (HF) remains a significant public health burden, with frequent readmissions often driven by a lack of guideline-directed medical therapy (GDMT) optimization. While clinical decision support (CDS) can improve healthcare processes, patient-centered tools that empower individuals to participate in their own care are increasingly needed. We developed SmartHF, an adaptive web application that provides patients with tailored HFrEF GDMT recommendations based on their clinical data.

Methods

This 12-week, multicenter randomized controlled trial enrolled adults with HFrEF (LVEF </=40%). Participants were randomized 1:1 to either the SmartHF intervention or usual care. The application uses a computable algorithm to integrate electronic health record (EHR) data, including medications, laboratory results, and vitals, via patient portal authorization. The primary endpoint is the change in a GDMT medication optimization score from baseline to Week 12.

Results

Enrollment concluded in August 2025 with a total of 225 participants. The study population had a mean age of 61.9 years, with 36% identifying as female and 52% as Black or African American. At baseline, 84% of participants were prescribed an ACEi/ARB/ARNI, 90% were on beta-blockers, 67% were prescribed an SGLT2i, while only 53% were on a mineralocorticoid receptor antagonist (MRA). The mean baseline medication optimization score was 68.5. Final trial results are expected in Winter 2026.

Conclusions

SmartHF is an innovative, low-touch digital health intervention that leverages EHR integration to facilitate patient-provider discussions on medication optimization. By providing personalized, guideline-based recommendations, the tool aims to improve GDMT prescribing in patients with HFrEF.

Clinical Trials Registration

ClinicalTrials.gov ID NCT05858320

1. Introduction

Heart failure (HF) readmissions are a substantial concern and a significant public health burden. By 2030, over 8 million Americans will have HF, and HF will account for $69 billion in annual medical expenditures [1]. Much of this cost is related to hospitalizations for worsening HF, which also exerts a significant toll on patients. Hospital readmissions occur frequently, as more than one-fifth of patients with HF are readmitted within 30 days of initial discharge [2]. These recurrent hospitalizations not only increase the mortality risk [3] but also worsen health-related quality of life (HRQOL) [4], contribute to functional decline, and increase the risk of long-term institutionalization [5]. Hospitals also suffer the consequences; HF is the primary driver for federal 30-day readmission penalties [6], of which now exceed $500 billion annually. Reducing HF readmissions is thus an essential goal across the healthcare spectrum, from patients to payers. One of the most common clinical problems leading to HF readmission is suboptimal guideline-directed medical therapy (GDMT).

Clinical trials have established that guideline-directed medical therapy (GDMT), including beta-blockers [7], [8], [9], ACE inhibitors (ACEi) [10], [11], angiotensin receptor blockers (ARB) [12], angiotensin receptor neprilysin inhibitors (ARNI) [13], mineralocorticoid receptor antagonists (MRA) [14], [15], and sodium-glucose cotransporter-2 inhibitors (SGLT2i) [16], [17], reduces mortality and hospitalizations and improves quality of life among patients with HF with reduced ejection fraction (HFrEF). Multiple contemporary studies have shown that GDMT, in chronic HF with a reduced ejection fraction (HFrEF), is suboptimal, and even fewer patients are prescribed target doses [18], [19], [20], [21]. The IMPROVE-HF prospective cohort study was designed to characterize the management of patients diagnosed with HFrEF [18]. In all eligible patients with no contraindications, angiotensin-converting enzyme inhibitors (ACEi)/angiotensin receptor blockers (ARB) were only used in 79.6% of patients, beta-blockers were only used in 86% of patients, and mineralocorticoid receptor antagonist (MRA) was only used in 36.1% of patients. Researchers have directly matched patients with clinical trials, demonstrating low use of evidence-based therapy in HFrEF [19]. More recent data from the CHAMP-HF registry [20] and the Veterans Health Administration [21] have also demonstrated the need for GDMT optimization in patients with HFrEF.

Traditional quality improvement efforts have focused on in-person GDMT optimization through clinic visits and clinician-facing alerts. However, this approach slows medication titration. An urgent, unmet need remains for implementation strategies that support timely medication titration without placing prohibitive demands on clinical staff. Digital health interventions that use patient-facing technology to encourage self-monitoring and feedback offer a promising way to empower patients and foster sustained engagement [22].

Patient-centered clinical decision support (PCCDS) addresses quality and safety issues while attaining better outcomes. Clinical Decision Support (CDS) delivers the right information to the right people, in the right format, at the right time, through the right channels. Clinical practice would improve significantly if the four independent predictors of CDS were implemented [23]. These features are crucial to the success of a CDS. A systematic review of CDSs examined their effects on clinical outcomes, healthcare processes, workload and efficiency, patient satisfaction, costs, provider use, and implementation [24]. The review found that locally developed and commercial CDS products improved the healthcare process by ordering clinical studies and prescribing therapies. CDS facilitates the transfer of knowledge to clinical care, significantly increasing the likelihood of ordering the appropriate therapy [25], [26], [27], [28], [29], [30]. PCCDS supports individual patients and their caregivers in health decisions and actions by leveraging patient-specific information [31]. Empowering and engaging patients in healthcare decision-making has demonstrated positive clinical outcomes [32]. These outcomes include improved disease management, health status, and medication adherence. In cardiovascular disease, hypertension is a prime example of PCCDS. A systematic review and meta-analysis of interventions to improve blood pressure in hypertension found that patient-level interventions were associated with a significant reduction in blood pressure [33]. These types of interventions also allow for more robust shared decision-making.

We have developed an adaptive medication-optimization web application, SmartHF, that provides patients with tailored information on optimizing HFrEF GDMT. SmartHF integrates patient EHR data, using open-access patient portal authorization, into a validated computable algorithm. The computable algorithm analyzes patient medications, laboratory results, allergies/intolerances, race, and vital signs from the EHR to generate GDMT optimization recommendations and a GDMT medication quality score. SmartHF guides patients to share GDMT optimization recommendations with their provider through the patient portal messaging system before a clinic visit. SmartHF is based on the idea that informed, engaged patients can facilitate shared decision-making by initiating discussions with their clinicians about evidence-based therapies.

In this clinical trial, we aim first to determine the effects of an adaptive medication-optimization web application on guideline-directed prescribing of medications in HFrEF. Participants with HFrEF will be randomized to the SmartHF intervention or usual care. We will collect medication and dose changes from baseline to Week 12. Second, we plan to assess concordance between the medication optimization algorithm's recommendations and the GDMT medications prescribed. We will compare SmartHF algorithm recommendations at baseline with GDMT medications from baseline to Week 12. Lastly, we aim to identify patient and provider characteristics that influence the medication optimization web application's efficacy.

2. Methods

2.1. Study design

This multicenter randomized controlled trial (University of Michigan, Washington University, Trinity Health, Grady Health System, Henry Ford Hospital) will evaluate the effectiveness of an adaptive web application, SmartHF, in facilitating GDMT optimization. Our central hypothesis is that the SmartHF web application can improve GDMT prescribing in patients with HFrEF. This is a 12-week, prospective, randomized, controlled, multicenter clinical trial. Participants will be randomized to receive the web application intervention or control, using the Mersenne Twister pseudo-random number generator method [34], in a 1:1 ratio, up to 30 days before a clinic visit, and will be followed for 12 weeks after the clinic visit. This trial will assess the efficacy of SmartHF for prescribing GDMT in HFrEF patients compared with controls. Table 1 lists all participant inclusion and exclusion criteria.

Table 1.

Patient inclusion and exclusion criteria.

Inclusion Criteria
1. Age 18 years old and older at the time of screening 2. Have a diagnosis of heart failure with a left ventricular ejection fraction (LVEF) </=40% 3. Seeing a general medicine provider or general cardiology provider for HFrEF 4. Have access to the internet and their health system's patient portal 5. Must meet at least three of the following: systolic blood pressure ≥ 110mmg, heart rate ≥ 70, potassium ≤5.5, serum creatinine ≤2.5, eGFR ≥30 6. At least two eligible heart failure therapies are not yet initiated or are below 50% of the target dose 7. Fluent in spoken and written English
Exclusion Criteria
1. Actively being treated for cancer, except for non-melanoma skin cancer 2. Have an implanted ventricular assist device 3. Currently being treated with chronic inotropic therapy 4. Seeing an advanced HF specialist 5. Currently on dialysis 6. Had a previous solid organ transplant 7. Are in end-stage HF 8. Are currently pregnant or intend to become pregnant during the study period 9. Have Severe Aortic Valve Stenosis or Severe Mitral Stenosis

2.2. Intervention

The intervention is an adaptive medication-optimization web application called SmartHF. Participants grant the app access to their medical records, and the algorithm is designed to process inputs from medications (including name and daily dose), New York Heart Association (NYHA) classification, systolic blood pressure, heart rate, serum creatinine, potassium levels, allergies/intolerances, and race for each patient. The algorithm outputs medication optimization recommendations. The algorithm uses cut points for systolic blood pressure, heart rate, serum creatinine, and potassium to provide clinically appropriate recommendations. Fig. 1 shows a screenshot of the SmartHF application.

Fig. 1.

SmartHF application screenshot.

SmartHF medication recommendations were provided to participants in the intervention arm at baseline. The recommendations are based on the 2022 ACC/AHA HF guidelines. The SmartHF web app displayed the recommendations and emailed them to intervention group participants. We sent reminder emails to participants in the intervention arm 1 week before, 3 days before, and on the day of their clinic appointment. Participants in the intervention arm also received a reminder text message on the day of their clinic appointment, prompting them to share the medication recommendations with their physician. The control group was shown all SmartHF web application screens except the results screen with medication recommendations. The control group did not receive the medication recommendations or email/text message reminders. The SmartHF app connects to all participants' EHRs on 3 days, 6 weeks, and 12 weeks after the clinic appointment without requiring participant input. Data collected at these time points were for clinical trial endpoint assessment and were not shared with participants or providers.

2.3. Study endpoints

Our primary objective is to assess the impact of an adaptive SmartHF web application on GDMT optimization in HFrEF patients by measuring the change in the GDMT medication quality score generated by the computable algorithm from baseline to Week 12. This computable algorithm provides a medication optimization score (MOS). This score is continuous, ranging from 0 to 100%, with higher values indicating greater optimization of evidence-based HFrEF therapy. The SmartHF computable algorithm generates the score by integrating patient-specific clinical characteristics, medication doses, laboratory values, vital signs, and eligibility for each GDMT class. We previously validated this computable algorithm using clinical trial and real-world data, and it has been used in prior literature [35], [36], [37]. Second, we aim to assess concordance between the medication optimization algorithm's recommendations and the medications prescribed by examining the degree of agreement between the algorithm's recommendations and the actual GDMT prescribed. Third, we aim to assess the intervention's impact on HRQOL, as measured by the Kansas City Cardiomyopathy Questionnaire (KCCQ-12), from baseline to Week 12, as well as on HF admissions and mortality. Lastly, we aim to identify patient characteristics (including race, comorbidities, and medications) and provider characteristics (specialty, race, and ethnicity) that moderate the effectiveness of the SmartHF web application. Table 2 outlines a schedule of activities for the outcomes assessed in the clinical trial.

Table 2.

Schedule of activities.

Procedure	Baseline (0–30 days before clinic visit)	Clinic Visit (Day 0)	Day 3 (+5 days)	Week 6 (+/− 7 days)	Week 12 (+/− 14 days)
Informed Consent	SS
Inclusion/exclusion	SS
Demographics	SS
Medical history	SS
Randomization	SS
Web application	Participant		CCC	CCC	CCC
All Aspects of Health Literacy Scale	Participant
Social Determinants of Health Survey	Participant
KCCQ-12	Participant				Participant
EHR review	SS		SS	SS	SS
• Clinical events (death, hospitalization, other AEs)
• Medications
Follow-up phone call					SS/Participant
Final Status					SS

Key: EHR = electronic health record, SS = study site, CCC = clinical coordinating center.

2.4. Sample size and statistical analyses

We will compare changes in the GDMT medication optimization score (MOS) over time using repeated-measures mixed models to account for inter- and intra-subject variation. Based on preliminary data from the GUIDE-IT trial, we anticipate a clinically meaningful difference in the medication optimization score of 7.5% to 12.5%. For power, we estimated a 10% change in the absolute score. When the sample size is 83 in the intervention and 83 in the control group (N = 166), a two-sided test for the time-averaged difference between two means in a repeated measures design with a 0.05 significance level will have 80% power to detect a difference in means of 10% in a design with four repeated measurements (baseline, Day 3, Week 6 and Week 12) when the standard deviation is 28% and the between-level correlation is 0.5. All participants who attend their clinic visit will be included in the intent-to-treat (ITT) population. We will exclude participants who miss scheduled clinic appointments from the analysis. ITT analyses will be included in all treatment comparisons of primary and secondary outcomes. All randomized participants who received an intervention will be included in safety analyses. The primary outcome will be the change in GDMT MOS generated by the SmartHF algorithm, with MOS scores calculated at baseline, Day 3, Week 6, and Week 12. The quality score ranges from 0% (least optimized) to 100% (most optimized HFrEF medication regimen). The medication optimization score incorporates the four HFrEF GDMT classes: ACE inhibitor/ARB/ARNI, beta-blocker, MRA, and SGLT2 inhibitor therapy. To calculate this score, we will collect all HF medication changes from baseline to Week 12.

For repeated-measures outcomes, we will use generalized estimating equations (GEEs) to evaluate the association between treatment group and each outcome over time. GEE extends general linear models for analyzing longitudinal data in which repeated observations for each subject are expected to be correlated. It provides consistent estimates of population-averaged effects by specifying a working correlation structure to model the relationship between repeated measures within subjects. We will select the working correlation structure based on the Quasi-Likelihood Information Criterion and use the identity link for continuous outcomes. Longitudinal analyses will use all available repeated observations. We will not impute missing outcome data; analyses will assume data are missing at random. Secondary endpoints are exploratory, and we will not apply multiplicity adjustment. Findings from these analyses will therefore be interpreted cautiously.

The secondary endpoint is the level of agreement between the SmartHF MOS algorithm recommendations (Yes/No) and the actual GDMT prescribed. For each participant, we will determine whether the algorithm made a recommendation and whether that recommendation changed the provider's prescribed GDMT medications over the 12-week period. Cohen's kappa will assess agreement between the SmartHF MOS algorithm's recommendations and the prescribed GDMT. Cohen's kappa is the difference between the rate of agreement that is observed and the rate of agreement expected by chance alone. Cohen's kappa ranges from 0 to 1. A kappa of less than or equal to 0.20 is poor, 0.21 to 4.0 is fair, 0.41 to 0.60 is moderate, 0.61 to 0.80 is good, and 0.81 to 1.00 is a very good level of agreement. We will calculate a confidence interval for kappa. The associated confidence interval provides a more reliable assessment than the kappa value alone.

We will calculate six additional HFrEF medication scores for each participant. The HFrEF medication scores summarize the medications participants are taking at the corresponding clinical encounter, with primary focus on angiotensin-converting enzyme inhibitors (ACEIs), angiotensin receptor blockers (ARBs), angiotensin receptor-neprilysin inhibitors (ARNIs), evidence-based beta-blockers (BBs), mineralocorticoid receptor antagonists (MRAs), and sodium-glucose transport inhibitors (SGLT2i). These seven scores are the most recent Heart Failure Collaboratory (HFC) scoring method described by Fiuzat et al [38], the latest version of modified HFC (version 13) [39], Quadruple Therapy Optimization score (QTO) [40], Optimization Potential Score (OPS) [41], Kresoja 2024 score [42], and Modified KCMO [43].

2.5. Data Safety Monitoring Board

The Data Safety Monitoring Board (DSMB) was formed and meets every six months. The DSMB approved the initial protocol and informed consent. The study team prepared interim reports for the DSMB, including study progress, recruitment, retention, participant characteristics, protocol changes, protocol deviations, and safety data. At each meeting, the DSMB reviewed the data and recommended continuing the clinical trial.

3. Results

The SmartHF study was launched on July 21, 2023, with participant enrollment commencing on July 24, 2023. Enrollment was completed in August 2025, with 225 participants. Table 3 displays participant demographics. After three months of follow-up, results are expected in Winter 2026.

Table 3.

Baseline participant demographics.

Characteristic or Category	Overall N = 225
Age (years), mean +/− SD	61.9 +/− 13.11
Sex (at birth), n(%)
Female	82 (36%)
Male	143 (64%)
Ethnicity, n(%)
Not Hispanic or Latino	225 (100%)
Hispanic or Latino	0(0%)
Race, n(%)
White	101 (45%)
Black or African American	117 (52%)
Asian	5 (2%)
American Indian or Alaska Native	0 (0%)
Native Hawaiian or Other Pacific Islander	0(0%)
Multi-Race	1 (0%)
Unknown or Not Reported	1 (0%)
Medication Class, n(%)
ACE Inhibitors/ARB/ARNI	190 (84%)
ARNI	85 (38%)
Beta-blockers	202 (90%)
MRA	119 (53%)
SGLT2 Inhibitors	151 (67%)
Hydralazine	8 (4%)
Nitrates	31 (13%)
Loop Diuretic	159 (71%)
Thiazide	2 (1%)
ACE/ARB/ARNI % Target Dose	N = 188
Mean	29.8
SD	21.64
Beta-blocker % Target Dose	N = 198
Mean	25.6
SD	20.4
GDMT Medication Optimization Score, %
Mean	68.5
SD	25.67
Self-Reported NYHA Class, n(%)
Class I	63(28%)
Class II	104(46%)
Class III	49(22%)
Class IV	9(4%)
Heart Rate, BPM
Mean	77
SD	14.33
Systolic Blood Pressure, mmHg
Mean	119.1
SD	18.83
Diastolic Blood Pressure, mmHg
Mean	73.4
SD	12.3
Serum Creatinine, mg/dL
Mean	1.2
SD	0.39
Serum Potassium, mmol/L
Mean	4.3
SD	0.45
Serum Sodium, mmol/L
Mean	139.2
SD	3.27
HF Etiology, n(%)
Ischemic	86(38%)
Non-ischemic	127(56%)
Missing	12(5%)
Left Ventricular Ejection Fraction, %
Mean	29
SD	8.03
eGFR, mL/min/1.73 m2
Mean	67.6
SD	19.4
Number of HF Hospitalizations in the last 12 Months, n(%)
0	90(40%)
1	95(42%)
2	21(9%)
3 or more	19 (8%)

4. Discussion

PCCDS can engage patients in their care and facilitate discussions on medication optimization in HFrEF. The SmartHF application connects to the patient portal and, after authorization, provides patients with personalized guideline-based medication recommendations to discuss with their provider at the next clinical visit. The application uses Epic on FHIR, which enables applications to connect to the Epic EHR via patient portal authorization. The data is transmitted in the FHIR format and includes, but is not limited to, the U.S. Core Data for Interoperability data classes and elements.

The SmartHF trial adds to the body of evidence on interventions to improve GDMT in HFrEF [44], [45], [46], [47], [48], [49]. Many interventions require providers to intensively review patient records and require significant healthcare resources to implement. SmartHF is similar to the EPIC-HF trial intervention [45], which is designed to engage patients in discussions with their providers to optimize GDMT in HFrEF. The key difference between SmartHF and the EPIC-HF tool is that SmartHF incorporates electronic health record data, enabling personalized recommendations for each patient.

This clinical trial has some limitations. Participants were enrolled if they responded to an email invitation, provided remote consent, and authorized access with their patient portal login. Consequently, the study population may represent patients with greater digital literacy and healthcare engagement than the broader HFrEF population. In addition, eligibility criteria requiring English proficiency, internet and patient portal access, and minimum physiologic thresholds for safe GDMT optimization may have resulted in a relatively stable outpatient population with opportunities for medication intensification. Therefore, the findings may not generalize directly to patients with advanced heart failure, limited digital access, limited English proficiency, or clinical characteristics that preclude safe GDMT optimization. Additionally, participants were required to have at least two eligible GDMT therapies that were either not initiated or prescribed at less than 50% of the target dose. Consequently, the study population was enriched for patients with substantial opportunities for medication optimization, a necessary condition to evaluate the intervention's intended effect. However, this eligibility criterion may have increased the likelihood of detecting improvements in GDMT prescribing compared with a less-selected HFrEF population that was already receiving more comprehensive GDMT. Accordingly, the magnitude of the intervention effect may not be directly generalizable to populations with higher baseline GDMT optimization. Although both groups completed this process, it may limit generalizability to those who were able to perform these tasks. Given the nature of the intervention, participants knew their group assignment. Participants in the intervention initially received the recommendations and were subsequently emailed them before their clinic appointment. Control participants did not receive any information after authorizing through their patient portal. Even though participants were unblinded, there are few options for blinding participants in an intervention such as SmartHF. Additionally, as part of our ITT population, we required that participants attend their clinic visit to be included in the analysis. This could introduce bias into the population studied.

In conclusion, SmartHF is an innovative, low-touch intervention that has the potential to improve GDMT optimization in patients with HFrEF.

CRediT authorship contribution statement

Michael P. Dorsch: Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Amy Krambrink: Writing – review & editing, Methodology, Formal analysis, Data curation. Sabah Ganai: Writing – review & editing, Writing – original draft, Validation, Project administration, Data curation. Kaitlyn M. Greer: Writing – review & editing, Writing – original draft, Project administration, Data curation. Mohamed S. Ali: Writing – review & editing, Writing – original draft, Project administration, Data curation. Rachel Stevens: Writing – review & editing, Project administration. Uzochi Nwoko: Writing – review & editing, Project administration. Modele O. Ogunniyi: Writing – review & editing, Investigation. Gregory Ewald: Writing – review & editing, Investigation. Khaled Abdul-Nour: Writing – review & editing, Investigation. Scott L. Hummel: Writing – review & editing, Methodology, Investigation, Funding acquisition, Conceptualization. Todd M. Koelling: Writing – review & editing, Methodology, Investigation, Funding acquisition, Conceptualization.

Informed consent

Written informed consent was obtained for anonymized patient information to be published in this article.

Ethics approval

Ethical approval for this study was obtained from the Advarra Central IRB (#Pro00067092).

Declaration of Generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the author(s) used Grammarly to help with spelling, grammar, and readability. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Funding

The research was funded by the Agency for Healthcare Research and Quality.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Michael P. Dorsch reports financial support was provided by Agency for Healthcare Research and Quality. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.