Keragon Founder Used Claude to Avoid Unnecessary Radiotherapy

Connor Christou describes himself as "lucky in my unluckiness." The 35-year-old founder of Keragon, a company that automates administrative operations for medical practices, discovered an 11-by-11-by-8 centimeter tumor behind his sternum while seeking treatment for blood clots. The diagnosis was a rare non-Hodgkin's lymphoma affecting roughly one in 420,000 people. What followed was an unusually data-intensive treatment journey, reported by TechCrunch, in which Claude played a recurring role.

Throughout his treatment, Christou fed "blood results, scan data, wearable output, journal entries" into the model. When his final PET scan came back ambiguous, Claude flagged a benign explanation: thymus rebound, a known phenomenon in patients under 40 recovering from this type of lymphoma. The model put roughly 90% probability on this reading. Three additional physician consultations confirmed no active disease was present, making radiotherapy unnecessary. Christou is direct about what the AI did and did not do: "It didn't replace the doctors...but it helped me ask the right questions."

Part of what makes this case notable is the underlying diagnostic landscape. According to Christou, end-of-treatment PET scans for his specific lymphoma type carry roughly a 60% false-positive rate, a number he frames pointedly: "It's 2026. Sixty percent." For rare diseases where, in Christou's words, access to comprehensive medical literature is "simply not the same as a Google search," that gap between clinical ambiguity and available information is exactly where AI can help patients prepare sharper conversations with their physicians.

A caution came from Mass General Brigham's clinical lead, who in the same piece described general-purpose chatbots as "frequently wrong" and noted they "have not been thoroughly evaluated" for personalized diagnoses. Christou's outcome also rested on unusual resources: 12 medical opinions gathered across his treatment, wearable health tracking, and hands-on use of frontier AI. Whether this pathway generalizes to patients without those advantages is a question the reporting does not answer.

The case for watching this space is straightforward. Rare-disease patients have always faced a literature problem: too much to read, too little time, and too few physicians who specialize in a condition affecting one in 420,000 people. If AI can reliably compress that into better-prepared clinical conversations, the value is real. What remains unsettled is whether it can be delivered safely and consistently at scale, rather than case by case.