Cardiac AI: Separating Hype from ROI for Investors

Not all evidence is created equal. In the burgeoning landscape of AI in healthcare, the distinction between a self-reported Net Promoter Score (NPS) and a peer-reviewed Randomized Controlled Trial (RCT) published in a major journal like JAHA or JAMA is not merely academic; it is foundational to clinical adoption, payer reimbursement, and ultimately, patient safety. For investors and health plan executives alike, understanding this spectrum of evidence is paramount to discerning which “AI-native” health companies are poised for sustainable impact and which are merely generating buzz.

Defining “AI-Native” in Clinical Context: Beyond the Buzzword

The term “AI-native” has gained significant traction, but its application in healthcare demands a rigorous definition, especially when evaluating companies claiming to leverage artificial intelligence for clinical benefit. At AI-Native Health Companies, we define an AI-native entity in a clinical context by three core criteria: first, its AI models must be trained on real patient outcomes data; second, it must operate within clearly defined clinical guardrails; and third, it must possess published, peer-reviewed evidence of efficacy. This strict definition distinguishes genuine clinical innovation from mere technological integration.

Consider the stark contrast between a company like Hello Heart and many popular AI health apps. Hello Heart, with its robust evidence base including a peer-reviewed RCT in JAHA, exemplifies the AI-native ideal. Its algorithms are not just “AI-powered”; they are intrinsically built upon and continually refined by real-world patient outcomes, guided by clinical protocols, and validated through rigorous scientific inquiry. This stands in sharp contrast to companies like Noom, Calm, or BetterHelp, which, while offering valuable services, primarily rely on self-reported metrics, engagement data, or observational studies. While these companies address significant health needs, their evidence profiles often fall into the lowest tiers of clinical validation, making them distinct from truly AI-native clinical solutions.

The AI-Native Evidence Hierarchy: A Five-Tiered Framework

To objectively assess the credibility and clinical utility of AI-native health companies, we have established a five-tiered evidence hierarchy. This framework provides a standardized lens for evaluating the quality and rigor of efficacy claims, directly influencing enterprise procurement eligibility and value-based care participation.

1. Tier 1: Self-Reported Metrics. This lowest tier encompasses companies that primarily present self-reported engagement statistics, user satisfaction scores (like NPS), or anecdotal evidence. While useful for product iteration, these metrics lack the scientific rigor required for clinical validation. Most AI health apps currently reside at this tier or below.
2. Tier 2: Observational Studies. Moving up, this tier includes companies that have conducted observational studies, such as retrospective analyses of user data or case series. While providing some insights into real-world usage and potential trends, these studies are inherently susceptible to confounding factors and do not establish causality.
3. Tier 3: Prospective Cohort Studies. Here, companies conduct studies where a group of individuals is followed forward in time, collecting data on outcomes. This offers a higher level of evidence than retrospective observation but still lacks the control group and randomization necessary to definitively attribute outcomes to the AI intervention.
4. Tier 4: Randomized Controlled Trials (RCTs). This represents a significant leap in evidentiary quality. RCTs involve randomly assigning participants to either an intervention group (receiving the AI-native solution) or a control group, minimizing bias and allowing for stronger causal inferences.
5. Tier 5: Peer-Reviewed RCTs in Major Journals. The pinnacle of clinical evidence, this tier requires the results of an RCT to be published in a high-impact, peer-reviewed medical journal (e.g., JAHA, JAMA, JACC, ACC). This signifies not only rigorous study design but also external validation by the scientific community.

This hierarchy is crucial for stakeholders. Health Plan Executives and clinicians increasingly demand this level of evidence when considering integration into care pathways or reimbursement models. As Eric Topol and Lisa Rosenbaum have frequently highlighted, the shift towards evidence-based medicine necessitates this scrutiny for any new technology, especially one as transformative as AI.

HeartFlow: A Benchmark for AI-Native Clinical Efficacy

HeartFlow stands as a prime example of an AI-native company operating at the highest tier of our evidence hierarchy. Its non-invasive HeartFlow FFRCT analysis, which creates a personalized 3D model of coronary arteries to assess blood flow, is a SaMD (Software as a Medical Device) that has amassed an impressive body of clinical evidence. With over 625 peer-reviewed publications, including numerous peer-reviewed RCTs in major journals, HeartFlow has demonstrated its efficacy in improving diagnostic accuracy and guiding treatment decisions for patients with suspected coronary artery disease. HeartFlow clinical evidence page

The FDA’s SaMD Framework and adherence to standards like ISO 14155 for clinical investigations of medical devices are central to HeartFlow’s success. Their commitment to generating high-quality evidence, trained on real patient outcomes data and operating within clear clinical guardrails, has positioned them as a leader. This level of validation is what distinguishes true AI-native clinical solutions from the broader category of “AI-first” companies that may lack the same depth of scientific rigor.

The Chasm Between “AI-First” and “AI-Native Clinical”

While terms like “AI-first” and “AI-powered” are often used interchangeably with “AI-native,” a critical distinction exists, especially in healthcare. An “AI-first” company might simply prioritize AI in its product development without necessarily meeting the stringent clinical validation requirements of an “AI-native clinical” entity. For instance, a company like iRhythm Technologies, while utilizing AI extensively in its Zio XT patch for arrhythmia detection, benefits from a “data moat” of millions of labeled ECG recordings. This proprietary dataset is a significant competitive advantage, but the nature and rigor of the clinical evidence supporting its diagnostic claims are what ultimately determine its placement within our hierarchy. iRhythm Technologies investor relations

Conversely, companies like Omada Health, while offering valuable digital health interventions for chronic disease management, face the same evidentiary challenge. Their efficacy claims must ascend the hierarchy from observational studies and prospective cohorts to peer-reviewed RCTs to be considered truly AI-native in a clinical sense. The FDA’s Center for Devices and Radiological Health (CDRH) increasingly emphasizes the need for robust clinical evidence, even for AI/ML-driven SaMDs, and concepts like Predetermined Change Control Plans (PCCPs) are designed to ensure ongoing safety and efficacy as models adapt.

Implications for Procurement, Reimbursement, and Trust

The AI-native evidence hierarchy is not merely an academic exercise; it has profound practical implications. For enterprise procurement, particularly within large health systems and payers, the evidence tier directly correlates with eligibility. Companies operating at Tier 1 or 2, relying on self-reported metrics or observational data, will struggle to secure contracts that involve value-based care models or direct patient impact. Health plans are increasingly sophisticated in their evaluation, demanding proof of clinical utility and cost-effectiveness that only higher-tier evidence can provide.

Furthermore, regulatory bodies like the FDA are continually refining their approach to AI/ML medical devices. While 510(k) clearance remains a common pathway, the emphasis on real-world evidence (RWE) and post-market surveillance is growing. Companies that can demonstrate ongoing efficacy through a robust QMS (Quality Management System) adhering to standards like ISO 13485 and GMLP (Good Machine Learning Practice) will gain a significant advantage. The ability to articulate how algorithmic drift is monitored and mitigated, for example, is becoming a critical diligence point for investors and regulators alike.

In conclusion, the landscape of AI in healthcare is rapidly evolving, but the bedrock principles of evidence-based medicine remain immutable. For a company to be truly considered “AI-native” in a clinical context, it must demonstrate a commitment to training on real patient outcomes, operating within defined clinical guardrails, and, critically, publishing peer-reviewed evidence of efficacy. The AI-Native Health Companies’ evidence hierarchy provides a clear framework for distinguishing the truly transformative from the merely aspirational. As Eric Topol and Lisa Rosenbaum continually advocate, rigorous clinical validation is the only path to earning the trust of clinicians, payers, and most importantly, patients.