“For health systems pursuing enterprise AI strategies, the most important lesson may be that trustworthy AI begins with trustworthy data”

Bill Siwicki, Managing Editor at Healthcare IT News

Modern cancer care is built on precision medicine, where treatment decisions depend on a complex combination of genomic markers, tumour characteristics, prior therapies and individual patient qualities. Clinical records can span years or even decades, with thousands of pages of pathology reports, imaging studies, treatment histories and physician notes.

For AI systems, this can create a significant challenge. Though recent advances like ambient AI documentation have successfully reduced administrative burden by generating clinical notes automatically, capturing a single consultation is only part of the problem. To be genuinely useful, AI must understand how new information fits within a patient’s entire clinical history.

However, much of healthcare’s most valuable data is buried deep within unstructured text, including physician narratives, radiology reports and pathology findings. Clinicians can interpret this information intuitively of course, but AI often struggles to distinguish between historical diagnoses, current conditions and treatment outcomes.

In oncology, these distinctions matter. Confusing a cancer diagnosis from 20 years ago with a more recent disease event could have monumental consequences on clinical decisions.

This challenge extends beyond direct patient care, pharmacovigilance teams increasingly use AI to identify adverse events, monitor drug safety and detect emerging safety signals. If these AI systems misinterpret clinical information or generate inaccurate outputs, organisations risk catching false signals or overlooking genuine ones.

The lesson for healthcare leaders is that data governance and AI governance are inseparable. Trustworthy AI depends on trustworthy data, robust validation processes and careful human oversight.