Dart Throwing and the Algorithm – The Health Care Blog

By GEORGE BEAUREGARD

How AI Could Have Personalized My Cancer Journey in 2005

I don’t think I’m in the minority of baby boomer doctors when it comes to my curiosity and ambivalence about the progressive application of AI in medicine. But this curiosity is not only prospective, it is also retrospective. In 2005, I became an outlier who perhaps needed something other than standard care for an illness.

In the fall of 2005, I saw for the first time a single drop of blood fall into the toilet water while I was urinating in my bathroom. After hitting the water, the pink pearl sank slowly, twisting and contorting, dissipating like a puff of smoke. The evidence was fleeting – gone in seconds. If I were a spectator rather than a source, I might have admired his visual art. There was no associated pain.

Only one thought crossed my mind: did I just pee blood? I thought maybe I had imagined it.

I was 49 years old and did not have what were considered risk factors for kidney or bladder cancer: smoking, obesity, advanced age, high blood pressure, or exposure to cadmium, trichlorethylene, or herbicides. But I was adopted and had no knowledge of my family history. Do I have a dark genealogy? Perhaps what was significant, however, was that both of my adoptive parents had developed different types of urogenital cancer. This led me to speculate that environmental factors related to the materials of our house and/or the land it sat on or around it may have played a role.

I tried to dismiss all worry, but the adage “Painless hematuria is cancer until proven otherwise” crossed my mind like a chyron.

The episodes continued and worsened, leading to an ultrasound whose report said: “…soft tissue density is seen at the base of the bladder to the right. Although this may represent a thrombus, I cannot rule out a primary mucosal injury. The lesion measures approximately 4 x 5 cm in diameter.

I consulted a urologist colleague, who performed a cystoscopy. His comment on what he saw: “As you know, you have a mass in your bladder. I have a very good view of it. It looks quite angry, so I suspect it is not benign. I tried to remove as much of it as I could. It would have been quite risky to scratch any deeper and risk puncturing your bladder. I know I didn’t understand everything.” A TURBT soon followed. Pathology showed a high-grade urothelial carcinoma widely invading the lamina propria and muscularis propria. There was multifocal lymphovascular invasion, so I probably had a more advanced subgroup than localized SEER stage.

At that time, the five-year relative survival rate for stage II muscle-invasive bladder cancer was about 45 percent.

In the vast majority of cases, bladder cancer is an age-related malignancy. So there I was, at 49 years old, suffering from cancer whose median age of incidence – septuagenarians – was much higher than mine. A WTF moment.

The one that made me think about how much time I had left.

So I had cancer, but in a way I felt cautiously optimistic. I had access to Boston-based academic centers and specialist colleagues willing to see me quickly, as well as good insurance.

But getting the diagnosis was just the beginning. I saw three expert urologists, each of whom recommended radical cystectomy, small bowel resection, and construction of an orthotopic ileal neobladder. Convergence. A certainty for me.

By the mid-2000s, approximately five hundred thousand new research publications were indexed on PubMed. Back then, oncologists typically began their research on a complex case with the NCCN/ASCO guidelines (synthesized evidence), checked supporting RCTs (gold standard), meta-analyses, and possibly checked ClinicalTrials.gov for new or ongoing studies before making a treatment recommendation.

I also saw three expert medical oncologists from different renowned academic medical centers. A memorable comment from one was: “The wolf is already out of the cage,” meaning that the likelihood of microscopic disease extending beyond the bladder was high.

Each of them recommended what was known and available at the time: a different chemotherapy regimen, “one size fits all seventies,” in terms of the types and numbers of agents used (doublet, triplet, quartet) and the timing of their administration relative to surgery (neoadjuvant, adjuvant, or half-and-half). Conflicting opinions. Divergence. Uncertainty for me.

Lacking solid evidence on which diet conferred a longer survival benefit, I was left with the equivalent of what felt like a throw of darts. I wondered if my choice would leave me underwater but could eventually surface, instead of drowning. My decision-making process ended up being driven primarily by intuition. I told myself: make a choice and don’t look back.

In 2005, the benefit of adding trastuzamab (Herceptin) in the treatment of HER-2 positive breast cancer had already been established. The oncologist I chose had a conversation with a colleague at the University of Michigan, a researcher specializing in HER2 and bladder cancer. FISH data from my cancer cells demonstrated a subclone of HER2 amplified cells; the percentage was uncertain, but low. After a discussion about the benefit-risk of adding Herceptin to my regimen, I agreed. For me, this decision was not to satisfy an academic curiosity, but for a survival advantage.

So here I am and, for the most part, a grateful (and I think lucky) survivor for 20 years.

But the way things have changed in oncology since then is as cancer care gradually shifts from the old generic nuclear bomb approach to stealth bombers.

In the black bag of today’s oncologists are new and improved tools at their disposal. Improvements in NGS, ctDNA and cfDNA testing, CAR-T cell therapy, qPCR and RT-PCR, spatial transcriptomics, epigenetic profiling technologies, mass spectrometry-based proteomics, epigenetic profiling technologies, etc. The advanced frontier of medicine.

While it is nice to have many more sophisticated tools, if the diagnostician or repairer is unsure which one will work best for a single person’s unique mix of cancer characteristics, he or she returns to scouring the medical literature, remembering what worked (or sort of worked) in other “similar” patients, pattern recognition, guidelines, and intuition.

In the pursuit of precision medicine, a powerful ally – AI – is moving from sidecar to main engine, driven by large language models capable of gathering, absorbing and collating previously unimaginable quantities of different clinically meaningful data points, and synthesizing them, predicting and steering treatment options away from unseen and unforeseen dead ends and rabbit holes, and tailoring treatment recommendations to an individual patient. And make course corrections as necessary along the way. Interpreting medical numbers. At warp speed.

Fine scalpels, not blunt instruments, guided by iterative learning and adaptation.

While I’m grateful to still be here, I wonder what a data-driven personalization platform would have recommended for my abnormal N-of-1 situation at the time.

I will never know, but my optimism and hope for more gains in the future effectiveness of cancer care tailored to individuals is growing. Although it will never be perfect, it will likely mean better patient outcomes.

One important thing remains: going upstream to detect significant cancer earlier at lower stages. Hope lives there too.

George Beauregard, DO is an internal medicine physician whose experience includes over 20 years of clinical practice as well as leading organizational strategic and clinical initiatives.. It comes from its substack