In an era when healthcare AI companies are racing to ingest every available piece of patient data, one prominent physician-entrepreneur is taking the opposite approach. Graham Walker, co-founder of the widely-used clinical calculator platform MDCalc and a former emergency room physician, argues that sometimes giving AI systems less information produces better clinical decision support.

Walker's new company, CrossCover, represents a deliberate counterpoint to the prevailing wisdom in healthcare technology. While major platforms expand their data collection and feature sets, CrossCover focuses on prognosis-specific AI that intentionally limits its information intake—a strategy Walker believes can reduce alert fatigue and maintain clinical utility.

The Industry's Data Appetite

The contrast between CrossCover's philosophy and current industry trends is stark. Epic, the dominant electronic health record system, now displays clinical recommendations directly within its platform. Abridge, originally focused on ambient medical scribing, has begun requesting patient insurance information. OpenEvidence, which started as a clinical question-answering tool, now requests patient records and visit data to deliver personalized recommendations.

This expansion reflects a common assumption in healthcare AI: more comprehensive data integration leads to better clinical insights. Companies are building increasingly expansive platforms that attempt to analyze all available patient information simultaneously.

The Case for Minimalism

Walker's thesis challenges this paradigm directly. According to his approach, limiting the information fed into AI clinical decision support tools can actually improve their performance by reducing noise and preventing the alert fatigue that plagues many healthcare systems. Rather than attempting to process every data point in a patient's record, CrossCover's prognosis tool focuses on specific, relevant information.

This minimalist strategy positions CrossCover against well-funded competitors pursuing platform consolidation. The philosophical divide raises fundamental questions about clinical AI design: does comprehensive data integration improve clinical decision-making, or does it create cognitive burden and generate false alerts that undermine physician trust?

Implications for Clinical Practice

The debate has real consequences for how AI tools function at the point of care. Overly complex systems that analyze vast amounts of data may produce more alerts, potentially desensitizing clinicians to warnings or slowing down workflows. Conversely, limited-data systems carry the risk of missing critical information that could affect diagnostic accuracy or treatment recommendations.

Walker's background lends credibility to his contrarian position. MDCalc became a trusted resource for clinicians precisely because it provided focused, evidence-based clinical calculators rather than attempting to be an all-encompassing platform. Whether that same philosophy can succeed in the more competitive and capital-intensive world of AI-powered clinical decision support remains to be seen.

As healthcare organizations evaluate which AI tools to adopt, they now face a choice between two competing visions: platforms that promise comprehensive analysis of all available data, or specialized tools that deliberately constrain their scope to maintain clarity and reduce alert fatigue.

Sources

  1. Why this former ER doc thinks giving AI less information is sometimes better — STAT News