Each year, over 3 million people with cancer in England receive care through the NHS, generating millions of clinical documents, including things like letters between clinicians, multidisciplinary team meeting reports, and test results, written in medical language. This information is valuable for not only treatment, but also research because they contain rich detail about diagnoses, genetic biomarkers, treatments, and outcomes. However, computers can’t read them in the way they’re written. Instead, researchers rely on coded records that computers can process, but these codes only capture basic information and miss much of the detail. Going through millions of clinical notes by hand to extract the useful information would be an enormous task, and sharing sensitive patient documents directly with researchers raises serious privacy concerns.
OncoLlama comprises an NHS-developed AI tool that can securely extract clinical information from text documents, such as clinical letters. It currently does this with over ninety-eight point five per cent accuracy across twenty types of cancer in adults and helps include relevant data that would otherwise be left out of research. OncoLlama is designed to make cancer data more complete and usable, so clinicians and researches can use that data to improve treatments, reduce inequalities, and help plan better cancer services for the future, all whilst keeping patients’ clinical notes private and within NHS systems. This project focuses on the validation steps required to progress the tool’s validation, and once made available to NHS clinicians and services, patients could benefit from better informed and streamlined communications and accessibility to records that comprise more effective tools for clinical planning purposes.
