This week, Color has become the first virtual clinic in the world to receive ASCO Certification. This milestone is making me reflect on the AI revolution in cancer. We created Color’s Virtual Cancer Clinic to reinvent how cancer is managed - in a technology-first, AI-driven world. So, what does this revolution look like?
If you ask any CEO leading the AI revolution what AI will do for humanity, "solve cancer" always tops the list. The image they conjure is familiar: new cancer drugs, discovered by AI models, that finally crack the disease. This image reminds me of the genetics wave a few years ago - when we all expected a “precision medicine revolution.” The silver bullet dream never actually delivered, but a revolution happened nonetheless.
AI will grow our drug arsenal against cancer, but silver bullet drugs will not be the primary way AI reduces the mortality and cost of cancer. The AI revolution in cancer looks more like a barrage of lead bullets. The cancer revolution, which is already underway, consists of a set of upheavals in how cancer is screened, diagnosed, treated, and lived with.
The reason has little to do with AI itself. Today's cancer experience is not dominated by a failure of science, but by a failure to deliver the science we already have. Most patients don't know their cancer risk, even though the data to estimate it exists. Most cancers aren't diagnosed as early as current tests allow. At a time when treatment complexity is exploding, only a fraction of patients have access to oncologists and other specialists with the right expertise to effectively manage their treatment. Most patients lose weeks or months waiting between each step, and many lose their life savings before treatment is over.
Our belief is that while scientific miracles are being pulled off every day, the AI revolution in cancer will come through the compounding of upheavals in care delivery. These are making it possible to build an AI-first, nationally distributed cancer management health system. Here are its components:
Even in a world where miracle cures exist, catching cancer early remains a primary goal. Late-stage treatment - even when curative - exacts a brutal toll through surgery, aggressive therapy, cost and lost time. Intercepting cancer before it takes hold will always be the better path. To put this in perspective, today, Stage 1 breast cancer has a 1% 5-year mortality rate, while by Stage 4, mortality goes up to 68%. You can find a similar stage-based escalation across most cancers. An early stage cancer that is localized is a fundamentally different disease than a late stage cancer that is spreading across the body. It is unlikely that any treatment breakthrough will outshine the dramatic benefit of catching cancers early for the foreseeable future. This is why early detection remains a cornerstone of our battle against cancer.
AI is reshaping early detection on two fronts:
The standard guidance that women begin getting mammograms at age 40 implies that breast cells sit benignly until a 40th birthday flips a switch. They don't. Some women are at meaningful risk a decade earlier; others won't develop breast cancer until much later, if at all. Family history, genetics, BMI, race, reproductive history, and dozens of other factors already let us estimate individual risk, but the system rarely uses them. This isn't a science problem. It's a delivery problem. AI lets us run nuanced models for every person and every cancer, and tailor the timing and intensity of screening accordingly.
Most of the data we collect goes unused. Our diagnostic tools (MRIs, blood tests and even patient reported information) produce a mountain of data that today is never used. AI-assisted mammogram interpretation already outperforms radiologists in published studies. As we feed models richer inputs - pathology images, multi-omic profiles, longitudinal records - sensitivity will keep climbing.
A key transition that is occurring already is that we’re moving from single-test positive/negative interpretation that was “dumbed down” to provide consistency in clinical use to more of a probabilistic mindset. We come from a world where every test is seen as diagnostic, where we act as if the outcome of testing is binary (“does this test detect cancer - yes or no?”). This discards mountains of information and makes healthcare data-phobic (any inconclusive data is seen as a liability). With AI, testing can embrace the probabilistic reality of biology.
The science to dramatically improve early detection is already in hand. The revolution is in running the logic at scale, in a way no clinician can individually. For example, the WISDOM Study, run by UCSF, and powered by Color, has shown a ~30% reduction in late-stage breast cancer diagnoses by connecting risk models to screening. This is a small sampling of what's already possible at population scale.
When you have cancer, you are in a race against time. The tumor is growing and evolving, and every week between a suspicious finding and the right treatment is a week the cancer gets to grow. The mental toll of waiting is its own injury, but the clinical cost can’t be overstated: lost time worsens outcomes. Multiple studies (1, 2, 3) find roughly 1-3% increase in mortality per week of delay or gap in treatment.
Yet wait times dominate the patient experience today. Why? Not because of any fundamental limit - but because of serialized steps that could be parallelized, waiting for insurance approval, disconnected human workflows and delays in access to scarce experts’ time. The use of AI tools to coordinate across data, expertise, and institutions collapses these gaps.
Here is a truly pedestrian and unforgivable example we address routinely at Color. Many people take stool-based tests to screen for Colon cancer. A positive result is not a diagnosis, but represents >100x increase in likelihood, so you have to get a colonoscopy ASAP. Today, about 40% of Americans never get a clinical followup after a positive stool test. And even for those who do, they often get scheduled for a followup six months later. What do you think the pre-cancerous polyp is doing during those six months? Just by integrating steps like these, and similar ones in treatment, we have already cut the time from screening to treatment in half.
Days matter in cancer. Through the AI cancer revolution, SLAs for cancer followups are measured in hours, not weeks or months.
As cancer science advances, the clinical complexity of the disease compounds. Even when good options exist, getting access to experts who know when and how to use them is a challenge at every stage: assessing risk; confirming the diagnosis; choosing among surgical, therapeutic, and radiation options; managing side effects; and monitoring for recurrence.
What we call “the standard of care” is not the standard care most people get. This is where AI makes state of the art care the standard everyone receives.
Dozens of studies have shown that multi-disciplinary reviews (often called tumor boards) improve survival rates meaningfully - by about 33%. That is a bigger impact than most blockbuster cancer drugs. Today, fewer than 10% of cancer patients receive a true multidisciplinary review - this number should be 100%. In the pre-AI world, that kind of access to expertise was rate-limited by access to individual physicians. But now, greater access is possible because AI does the heavy lifting of intersecting clinical guidelines with the specifics of an individual's record. It is quickly becoming unacceptable to start a patient on an untargeted therapy while a more effective option is available.
Another facet of the expertise gap is that roughly 8% of adult cancer patients in the US enroll in a clinical trial. For many cancers, the best available treatment is through a trial, and yet patients never learn it exists. The matching problem (intersecting guidelines and trial eligibility criteria against a patient's evolving molecular, pathology, and treatment history) is almost a parody of an AI-shaped hole. Proactive, continuous guideline evaluation and trial matching are among the most consequential changes AI-driven cancer care will bring. This is a logistics breakthrough that unlocks the research breakthroughs we've already made.
Through this work, we built tooling that enables Color’s Virtual Cancer Clinic to do a continuous expert review for 100% of cancer patients we care for. Our clinical experts still review every case, but these now rely on AI tooling that allows them to serve orders of magnitude more patients than they did previously.
A cancer diagnosis takes over a person's life. The mental, financial, and logistical burdens layer on top of the disease itself, and most patients and their families absorb them alone. The results are bleak: 40% of Americans diagnosed with cancer lose their entire life savings within two years. This is unacceptable.
Every cancer patient deserves a layer of always-on services that handles the things they shouldn't have to: insurance fights, bill negotiation, side-effect management, mental health support, logistical coordination. Patients have already paid premiums for the care they need. They shouldn't have to fight to receive it.
This is one of the most concrete near-term applications of AI in healthcare, and one Color is already deploying across the patient journey. Every cancer patient will be surrounded by services that are always-on and able to alleviate the layers of burdens that are put on their shoulders.
Whenever a powerful new technology arrives, we instinctively imagine the silver bullet it will produce.
The early days of genetics promised "precision medicine" - and to those who lived through that hype cycle, the original vision never quite materialized. But look at what came out of it: risk stratification, blood-based cancer testing and targeted therapeutics. Every modern tool that has materially improved cancer care is a descendant of that wave. The silver bullet failed, but its vision fueled a revolution nonetheless.
The AI revolution - in addition to accelerating science - scales, amplifies and improves human-bound services. Today, we fail to deliver the value that scientific discoveries have already unlocked. The revolution doesn’t happen because of the arsenal alone, but through its effective use. We won't beat cancer with one dramatic reveal. We'll beat it by closing, one by one, the dozens of gaps where our system routinely falls short today. This is the cancer AI revolution.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。