From Automation to Action in Heart Failure: Digital Solutions, Pragmatic Evidence, and the Integrative Role of Implementation Science

Heart failure with reduced ejection fraction (HFrEF) remains a major public health challenge, with persistent gaps in guideline-directed medical therapy (GDMT). Improved guideline adherence could translate into years of added survival, underscoring the urgent need for innovative solutions that are both used and effective – a challenge that implementation science (IS) was developed to tackle.

In the Artificial Intelligence Mobile Health Trial of a Digital Platform to Optimize Guideline-Directed Heart Failure Therapy Using Wearable Sensors (AIM-POWER) trial DeVore et al. assess a digital platform to integrate patient home vital signs and laboratory data to improve clinician initiation and titration of HFrEF medications.¹ The trial demonstrated significantly greater improvement in the HF optimal therapy score in the intervention group (mean change 1.72 ± 1.75 SD vs 0.44 ± 1.18 SD, P<0.001), with more patients reaching quadruple therapy and target doses more rapidly. However, few medication optimizations were made once the digital intervention was withdrawn, suggesting that sustained benefit may require ongoing support.

Designing Next Generation Digital Solutions

As digital solutions advance in sophistication and increasingly incorporate artificial intelligence (AI), heightened awareness is needed to preserve autonomy and balance agency between technology, clinicians, and patients. Notably, responsible and replicable use of AI methods requires transparency in reporting how AI is used. Despite implications from the trial’s name, the specific innovations of how AI was included in the AIM-POWER trial are unclear.¹ DeVore et al. appropriately acknowledge in their earlier protocol paper the potential for future digital solutions to evolve into closed-loop systems that operate without direct clinician involvement.² However, such a future state would likely be appropriate only in low-risk, clear cut scenarios and would still necessitate some level of human oversight. Oversight is necessary to address possible unintended consequences arising from data drift and other well-recognized limitations of AI and automation, which remains constrained by the quality and availability of the underlying data used to generate its patterns.³

As the number of novel digital solutions increases, rigorous evaluations of their effects and costs, ideally compared to alternatives, are needed to guide selection. Cost and cost-effectiveness will be an important driver in decisions to adopt, especially for digital solutions that serve similar functions and lead to the same outcome. Assessments of costs should consider the perspective of whether the patient or health system bears the burden and personnel time needed for any technical integration, training or delivery. In AIM-POWER, the analysis didn’t address the presence of increased patient engagement or more frequent provider monitoring in the intervention group. Thus, it is unclear what the cost of implementing the platform is or whether the effect was due to the intervention itself, the by-product of increased patient engagement, or more frequent clinician monitoring. With the growth of digital solutions that enhance patient and clinician engagement, it is important to evaluate how costs and value compare to alternative approaches.

Pragmatic Study Designs for Actionable Knowledge

Pragmatic study designs are needed to maximize generation of relevant and generalizable knowledge.^4,5 This includes maximizing participation and representativeness (e.g., including participants similar to those you wish to generalize to) during recruitment and informed consent.⁶ Participant ‘opting in’ consent procedures may bias results favorably toward interventions requiring learning or extra effort, as such strategies may self-select patients who are more engaged or able to devote time and resources. Recruitment sampling should aim for representativeness across relevant characteristics, roles and perspectives (e.g., years of practice of clinicians, age, patient comorbidities) aiming for a generalizable population.⁷

When clinical equipoise is assumed, complete waiver of consent is increasingly accepted given the process of consent can reduce participation and representativeness, including Hawthorne effects -when individuals are aware they are being observed and alter their behavior- which can diminish rigor and generalizability.⁶ In AIM-POWER, patient but not clinician consent was required. Using the Expanded CONSORT approach to report and compare the characteristics of study participants to those of the broader sampled population can help clarify how various recruitment and consent approaches affect study outcomes Representativeness of study participants under-pins scalability of interventions and findings,⁸ which is vital for closing global HFrEF prescribing gaps. Additionally, for digital interventions to be sustainable and scalable, it is important to use technical build aspects that minimize resources for ongoing maintenance and that are ideally portable or feasible within the available resources of other settings.

When designing a pragmatic study for a digital intervention, it is important to address considerations that are specific to the digital nature of intervention. Some specific considerations include learnability, usability, trust and issues related to the digital divide. The digital divide refers to the gap between those that have access to digital technologies and the internet and those who do not, which can result in disparities in information, resources, and opportunities. Well established concepts in human computer interaction and user centered design are key to developing a minimally viable usable intervention.⁹ These concepts and methodologies assist in creating interventions that are “good enough” – recognizing that perfect is unrealistic when the technology is advancing so rapidly and iteration will be needed – while ensuring they are appropriately intuitive for the situation and promote trust. Although digital interventions offer opportunities to reduce barriers to healthcare access, restrictive recruitment eligibility and accessibility strategies may inadvertently exacerbate the digital divide. Designs must address technology access and digital literacy for all segments of the population, rather than focusing solely on the average demographic.

Implementation Science: An Integrative Framework

IS is not a stand-alone solution, but a critical framework that can bring together digital, pragmatic, and traditional approaches.¹⁰ Integration of evidence from both pragmatic and traditional, tightly controlled randomized trials is essential for comprehensive progress to improve knowledge generation. At the heart of IS is the goal of thoroughly understanding and aligning with the context to maximize impact, sustainability, and scalability.¹⁰ The context includes the culture, perspectives, characteristics, needs, values and priorities of the care team, health setting, patient and home environment, as well as how these issues interact. For digital interventions, IS is used with user-centered design methods to comprehensively assess context (e.g., clinical workflows, clinician preferences) with specific attention to the unique considerations of digital solutions and their application.⁹ User-centered design methods are approaches that prioritize the needs, preferences, and experiences of end users throughout the development process to ensure solutions are intuitive, effective, and trustworthy. IS can also guide the use of behavioral economic principles to design solutions that make the right thing easy without restricting choices – a concept increasingly being leveraged for digital interventions.¹¹ IS also leverages multi and mixed methods to conduct pragmatic evaluations that prioritize both program delivery (process issues including, who is reached, implementation fidelity of the intervention, costs, and adoption of the intervention) and effectiveness (clinical) outcomes.¹⁰ Importantly, IS also guides the evaluation of contextual factors that can influence different outcomes. Understanding these issues is needed to identify areas for improvement and to assess scalability.¹²

The cardiology literature increasingly advocates for use of IS as noted in the AIM-POWER protocol paper, but the trial does not describe how IS was applied.² The study did not indicate use of an IS theory, model or framework, nor did it explore issues such as: why and when clinicians did not follow recommendations or trust the platform, deviations from the protocol (e.g., not uploading labs timely), differences in implementation or clinical outcomes across clinicians, patient perceptions of or adherence to the platform—all of which could provide valuable insights into real-world implementation. Despite these limitations, the authors provided significant insights into the value of the digital platform.

Implementation Science–Guided Recommendations for Sustainable, Equitable and Scalable Digital Heart Failure Interventions

Below (Table 1) we summarize four key recommendations to guide implementation, evaluation, and generalized use of sustainable and equitable digital interventions in heart failure care.

Table 1.

Key recommendations to guide implementation, evaluation, and use

Recommendation	Description
Start with a logic model and use an IS Theory, Model, or Framework (dissemination-implementation.org)	Define the intent of the intervention and study plan visually using a logic model Use a theory, model, or framework that addresses context and integrates user-centered design considerations
Design for Dissemination, Sustainability & Equity from the Outset	Design for dissemination, sustainability, and equity, with specific attention to the digital divide (e.g., create a guide documenting technical build details to promote replication in new and different settings).
Consider Multi-level Participation, Representativeness & Generalizability in Study Design Decisions	Evaluate the impact of chosen inclusion/exclusion criteria, recruitment strategies and informed consent procedures Compare your study population to the population you want to generalize at multiple levels (setting, physicians, patients)
Comprehensively & Transparently Report	Clearly describe interventions including integration of any AI methods Report both implementation and effectiveness outcomes, including any adaptations to the intervention from how it was intended to be used, implementation costs and contextual factors that would assist in understanding the findings. Use the expanded CONSORT to compare characteristics among those who participated at multiple levels (e.g., settings, clinicians, patients) and from the population they were sampled from.

Conclusion: Promise, Pragmatism, and Progress

Closing the gap in GDMT for heart failure requires technological innovation, rigorous study design, and the integrative power of pragmatic, contextually based IS. DeVore et al help move the needle,¹ but more is needed. As digital interventions evolve, ongoing research must address not only clinical effectiveness but also implementation outcomes such as reach and quality of intervention delivery and factors driving adoption, equity, impact and sustainability. Progress must be pragmatic—recognizing the need for broader, more inclusive study samples and designs attentive to the unique issues of digital interventions, including rigorous partner engagement. These approaches can optimize local relevance while planning for sustainability and relevance to new settings, preferably assessed with multiple methods of evaluation. For replication, there also needs to be transparent reporting describing the intervention, including any AI methods used, along with the findings and a description of the context. We discuss a number of IS strategies to improve the rigor and practical relevance of research on digital solutions, acknowledging the importance of efficiency and speed in both advancing efforts to address care gaps and staying current in the face of rapidly changing technologies. Only by combining these approaches can heart failure care be advanced in a sustainable, scalable, and equitable manner for all patients.