The Rapid Rise of AI Clinical Decision Support
Around the world, technology companies are racing to develop artificial intelligence (AI)-enabled tools for clinical decision support (CDS), targeting a global market forecast to reach US $15 billion by 2033. A new generation of AI tools powered by large language models (LLMs) is moving rapidly from development into clinical settings. Health systems and technology companies are integrating these tools into clinical workflows, promising faster evidence synthesis and more-responsive decision-making at the bedside. But as CDS systems enter into practice, fundamental questions remain about how they are evaluated, what evidence supports their use, and how they will be governed at scale.
The promise of AI in healthcare has long been touted, but recent advances in natural language processing and generative AI have accelerated the pace of adoption. Unlike earlier rule-based systems, modern CDS tools can read and interpret unstructured clinical notes, synthesize vast amounts of medical literature, and generate personalized recommendations in real time. This capability has captured the attention of clinicians and administrators alike, who see potential for improved diagnostic accuracy, reduced administrative burden, and more efficient care delivery.
Key Players and Divergent Strategies
Among the US leaders vying for market share are generalist consumer AI tools such as ChatGPT, as well as specialized CDS systems, including those from Abridge, Atropos Health, and OpenEvidence. These specialist CDS companies have diverging opinions on how the field will develop next.
Pittsburgh-based Abridge develops ambient AI technology that listens to clinician–patient conversations and automatically generates clinical notes and other workflow-support tools for health systems. By automating documentation, Abridge aims to reduce physician burnout and free up time for patient interaction. The company's approach focuses on integrating seamlessly into existing electronic health record (EHR) systems, providing real-time transcription and note generation that can be edited and approved by clinicians.
Boston-based OpenEvidence markets an AI tool directly to physicians that the company claims is the “most widely used medical AI” among US clinicians. The platform is focused on distilling the best available evidence from the published literature, providing clinicians with concise, evidence-based answers to clinical questions. OpenEvidence's CEO and founder, Daniel Nadler, envisions a future where AI automates much of the “standard of care” parts of evidence synthesis, allowing physicians to focus on complex cases and patient communication.
Atropos Health, a company spun out of Stanford University, promotes a CDS system designed to enable clinicians to ask novel questions in situations where gaps may exist in traditional, formally published evidence. By leveraging real-world data from patient records, Atropos aims to generate evidence on the fly, addressing clinical uncertainties that are not covered by existing studies. This approach represents a shift from evidence-based medicine to evidence-generating medicine, where AI can produce actionable insights from data in near-real-time.
The Data Foundation: Electronic Health Records and Ambient Scribes
The data that power CDS systems increasingly come directly from patient records. Over the past decade, electronic health records (EHR) that automatically scan patient data to track disease indications and issue alerts to clinicians have become ubiquitous in many health systems worldwide. However, it was not until about three years ago, with the mass adoption of ‘ambient scribes’ that automatically transcribe clinician–patient consultations into clinical records, that AI-powered CDS seemed poised for prime time.
Ambient scribes have revolutionized the way clinical data is captured. By automatically generating structured notes from conversations, they have reduced the documentation burden on clinicians and improved the completeness and accuracy of EHR data. This rich, real-time data stream provides the fuel for AI algorithms to learn and generate insights. As Nadler notes, the goal is to automate much of the “standard of care” parts of evidence synthesis, making the latest medical knowledge accessible at the point of care.
However, the reliance on EHR data raises concerns about data quality, bias, and interoperability. EHRs are notoriously fragmented, with varying formats and coding practices across institutions. AI models trained on such data may inherit biases that could lead to disparities in care. Moreover, the integration of AI tools into clinical workflows requires seamless data exchange between systems, which remains a technical and regulatory challenge.
The Evidence Gap: Evaluation and Governance Challenges
As CDS systems move into practice, a critical question emerges: can the evidence keep up? Traditional clinical trials are slow and expensive, often taking years to produce results. In contrast, AI tools are updated frequently, and their recommendations may change as new data become available. This creates a mismatch between the pace of innovation and the pace of evidence generation.
Regulatory bodies such as the US Food and Drug Administration (FDA) have begun to establish frameworks for AI-based medical devices, but many CDS tools are not subject to premarket review. The FDA has issued guidance on Clinical Decision Support Software, but the boundary between regulated and unregulated tools remains blurry. Some tools are considered “low risk” and are exempt from premarket notification, while others require more rigorous review. This regulatory uncertainty has led to calls for clearer guidelines and post-market surveillance mechanisms.
Furthermore, the evidence base for many AI CDS tools is thin. A recent systematic review found that few AI-based CDS systems have been evaluated in prospective clinical trials, and most studies suffer from methodological flaws such as small sample sizes, lack of control groups, and short follow-up periods. The lack of robust evidence makes it difficult for clinicians and health systems to make informed decisions about adopting these tools.
Governance is another major concern. Who is responsible when an AI recommendation leads to a misdiagnosis or adverse outcome? How should AI tools be monitored and updated to ensure they remain safe and effective? These questions are particularly pressing as AI tools become more autonomous and are integrated into critical care pathways.
The Path Forward: Balancing Innovation and Evidence
Despite these challenges, the potential benefits of AI CDS are too great to ignore. AI has the ability to synthesize vast amounts of data, identify patterns that may be missed by humans, and provide decision support in resource-limited settings. To realize these benefits, stakeholders must work together to develop robust evaluation frameworks, transparent reporting standards, and adaptive regulatory pathways.
One promising approach is the use of “living” systematic reviews and real-world evidence to complement traditional trials. By continuously monitoring AI tools in real-world settings, researchers can generate evidence on performance, safety, and unintended consequences. This approach requires collaboration between developers, clinicians, researchers, and regulators, as well as investment in data infrastructure and analytics.
Another key aspect is clinician engagement. For AI CDS to be accepted and effective, clinicians must trust the tools and understand their limitations. This requires education and training, as well as transparent communication about how AI recommendations are generated and what evidence supports them.
Finally, there is a need for international harmonization of standards and regulations. AI CDS tools are global products, and differences in regulatory requirements across countries can hinder innovation and access. Collaborative efforts, such as the International Medical Device Regulators Forum (IMDRF), are working towards common principles for AI-based medical devices, but more needs to be done.
Conclusion
AI-enabled clinical decision support is scaling up rapidly, driven by advances in large language models and the proliferation of electronic health data. The market is booming, and companies like Abridge, OpenEvidence, and Atropos Health are leading the way with innovative approaches. However, the evidence base for these tools is still developing, and significant challenges remain in evaluation, governance, and regulation.
As AI continues to transform healthcare, it is essential that we keep pace with rigorous evaluation and thoughtful oversight. Only by ensuring that AI CDS tools are safe, effective, and equitable can we fully realize their potential to improve patient care and outcomes.
This article is based on reporting by Nature Medicine. Read the original article.
Originally published on nature.com
