Could AI catch what clinicians miss? Examining a compelling cancer case
Industry Buzz
We have built a vast, intricate, overburdened system that tries, heroically, to deliver care to people with infinite variability and demand using finite resources and time. The system is... incapable of giving every person the kind of thoughtful, individualized, adaptive care that they need.
—Steve Brown, CEO of CureWise, to Stat News
When he turned 60, Steve Brown knew something wasn't right: He'd lost weight and had no appetite. His abdominal pain left him with a sense that something serious was going on, so he made an appointment with his doctor.[]
After receiving full body scans, a colonoscopy, an endoscopy, and cardiac function tests, there were few answers.[]
His gastroenterologist removed some polyps and told him he may have mild gastritis. His cardiologist suggested he may be stressed or depressed. A few weeks later, Brown found himself in severe pain, so he went to the ER.[]
The doctors in the emergency department, Brown wrote in an essay for Stat News, saw his case with "fresh eyes." Within days, they discovered an aggressive form of blood cancer in his bone marrow related to multiple myeloma. While they caught it early, the cancer was already beginning to impact Brown's kidneys, gut, and heart.[]
As he adjusted to the news of his cancer diagnosis, Brown fed his medical records into an AI tool he had created himself—and it flagged a life-threatening condition that his many doctors had missed.
Missed diagnosis?
Brown developed a medical AI agent named "Haley," which utilized LLMs from OpenAI, Google, Anthropic, and xAI along with layers of medical context to explore his medical history. He fed Haley the same data that was shared with all of his doctors, including his chart history, labs, imaging results, and EHR notes.[]
Haley flagged a pattern: mild anemia, elevated, ferritin, low immunoglobulins. These are signs of immune dysfunction and bone marrow issues that were not addressed in Brown's previous doctor's visits.[]
Once Brown fed his cytogenetics report into his AI tool, he was able to analyze the significance of his genetic mutations and variants that might make standard cancer treatments less effective for him.[]
He later sought out oncologists and other specialists to provide feedback on what he'd learned through his algorithm to get multiple opinions on the treatment plan the AI had suggested to him. "The collaboration between my doctors and me, powered by AI, reshaped my treatment," Brown wrote—rather than the standard protocol, Brown said he's on daratumamab plus venetoclax, an off-label combination that has not had full clinical trials.[]
Brown created CureWise, an app version of his original AI tool, which he said has been shared with other cancer patients in a private beta.[]
"We have built a vast, intricate, overburdened system that tries, heroically, to deliver care to people with infinite variability and demand using finite resources and time. The system is overwhelmed," Brown wrote for Stat News. "It is structurally incapable of giving every person the kind of thoughtful, individualized, adaptive care that they need."[]
Personalized treatment
Personalized medicine is changing cancer care. Researchers at University of California San Diego School of Medicine have led the first clinical trial in the world to show that cancer drug treatments can be safely and effectively personalized based on the unique DNA of a patient's tumor.[]
"Every patient and every cancer is unique, and so should how we treat for them,” Jason Sicklick, MD, lead author of the trial and surgical oncologist at UC San Diego Health, said in a press release. Our findings demonstrate that precision oncology at the individual level is achievable. When every patient's treatment is guided by their tumor's distinctive DNA, we can treat cancer with better accuracy."[]
What this means for the clinic
Brown’s experience serves as an example of how AI in healthcare can help support more collaborative, patient-centered care—something physicians continue to struggle with among their patients.
Related: When patients complain about their health—but won’t take care of themselvesRather than replacing clinical expertise, tools like Haley can help patients better understand their data, surface relevant questions, and engage more meaningfully with specialists. In this case, AI functioned as a bridge—translating complex lab patterns and genomic findings into actionable insights that Brown then validated with his care team. This kind of bidirectional exchange may strengthen shared decision-making, particularly in complex diseases like multiple myeloma.
Takeaways for the clinic
Don’t brush off subtle lab abnormalities. Anemia, elevated ferritin, and low immunoglobulins can be early clues to bone marrow disease, especially when symptoms persist and the clinical picture doesn’t quite add up.[]
AI may help augment clinical pattern recognition. Tools that synthesize longitudinal data, including labs, notes, and imaging, may help flag missed connections and support earlier diagnosis, particularly in overburdened care settings.
Precision oncology is increasingly actionable at the individual level. Molecular profiling can guide tailored, multi-drug regimens, including off-label combinations in select cases, potentially improving outcomes compared to standard protocols.