AI decision support is scaling-up fast — can the evidence keep up?

The global market for AI-enabled clinical decision support tools is projected to reach US $15 billion by 2033. Companies are integrating these AI tools into clinical workflows, aiming to enhance evidence synthesis and decision-making at the patient's bedside. However, concerns persist regarding the evaluation methods, supporting evidence, and governance of these systems at scale.

Major players in the US market include generalist consumer AI tools like ChatGPT and specialized CDS systems from firms such as Abridge, Atropos Health, and OpenEvidence. These specialized companies have differing perspectives on the future development of the field. Abridge focuses on ambient AI technology that generates clinical notes from clinician-patient conversations, while OpenEvidence markets an AI tool to physicians, claiming it's the most widely used medical AI software among US clinicians, and emphasizes distilling evidence from published literature.

Atropos Health, a Stanford University spinoff, offers a CDS system that enables clinicians to ask novel questions when traditional evidence is lacking. Much of the data powering these CDS systems comes directly from patient records, with electronic health records becoming widespread over the past decade, particularly with the advent of 'ambient scribes' that automatically transcribe clinician-patient consultations.

Questions remain about how clinicians will trust and effectively utilize these AI-powered decision support tools, as demonstrated by studies showing that physicians often override around 90% of drug-drug interaction alerts generated by CDS tools. Trust issues have been highlighted in a 2026 audit of AI scribe rollout among 30,000 public health doctors in Ontario, Canada, where all 20 approved vendors exhibited inaccuracies in system-generated notes.

A class action lawsuit filed in California also challenges the use of AI scribes without patient knowledge or explicit consent. Critics, such as Karandeep Singh, a nephrologist and chief health AI officer at UC San Diego Health, warn that understudied AI CDS systems are being adopted haphazardly, raising patient safety concerns.

Singh's editorial emphasizes that clinicians are currently using general-purpose consumer AI tools without institutional oversight. Although OpenEvidence and Abridge have partnerships with leading medical publishers like the New England Journal of Medicine and JAMA, Atropos Health, led by CEO and co-founder Brigham Hyde, argues for the need to address the lack of real-world patient data in many AI CDS systems.

To tackle this, Atropos is developing a novel medical library called Alexandria, containing retails of unpublished observational evidence on various topics, generated from de-identified patient data from Stanford Health Care.