In the middle of Omaha, Nebraska, just down the hill from Warren Buffett’s house, is a medical complex. A large swathe of it is still under construction, some of it funded by money that has trickled in from Buffett’s family, including a cancer center that bears the name of his first cousin, Fred—who died of cancer—and Fred’s widow, Pamela. Though not fully complete, the $323-million center officially opened in late May, with former US vice president Joe Biden—who’s also striving to push forward cancer research after the death of his son Beau from brain cancer—dedicating the new complex at a ribbon-cutting ceremony.
Quartz visited about a week earlier. Painters were still touching up spots in the halls, and workers were hanging artwork throughout the building. There were corrals of desk chairs in plastic covers huddling on every floor. Research labs were bare husks of carrels and plugs; offices still needed to be outfitted with furniture. There was a fair amount of work to be done before the Fred and Pamela Buffett Cancer Center’s first patients were set to arrive June 5.
But when I met with Ken Cowan, the center’s director, he was beaming with excitement, and sure everything would open on time. (It did.) I was in Nebraska to, theoretically, get a glimpse of the future of cancer treatment.
There are lots of things that make the Fred and Pamela Buffett Cancer Center unique. For one, the center is designed to force serendipity upon the staff: researchers and clinicians will share break rooms and other facilities, and there’s a communal cafeteria, where Cowan, taking a page out of Silicon Valley’s playbook, hopes staff will mix, talk about treatments and new technologies, and be inspired to serve their patients or research in novel ways. The cafeteria is also in the lobby, and open to patients as well as staff; the idea is for the health-care workers to realize “the urgency of their work,” Cowan says. The facility’s mission, he says, is to offer “unlimited hope.”
Walking through the main atrium, waiting areas, and random hallways, I saw more pieces of art than medical devices. Cowan says there will be rotating galleries throughout the hospital, changing seasonally. There’s also an entire indoor-and-outdoor gallery of Dale Chihuly glasswork donated by the artist himself, as well as a meditation room custom-built by the artist that is open to patients and staff at all hours of the day. The conical room, draped in brilliant-white glass sculptures opening up to a small skylight, is as serene a place I’ve come across in any museum installation, let alone a hospital. Cowan says he hired an art director to see if they can use the facility’s installations to learn more about the palliative effects of art therapy.
The hospital also features other innovations like mood lighting and bamboo décor in radiology rooms, meant to put patients’ minds at ease as they’re run through an MRI machine. The facility is set up so patients have their blood drawn—or undergo any other procedure to gather vital information—on the ground floor, right after they walk in. The blood samples are then sent up in a pneumatic tube to be analyzed. The results are waiting for your doctor by the time you get up to their office. “It’s one-stop shopping,” Cowan says.
But the most important innovations in the hospital aren’t as obvious. Many reside in the small future offices of researchers yet to move into the new building, and in the server farm the hospital shares with the University of Nebraska. Upgrades to cloud services solutions don’t sound sexy, and they won’t get ribbon-cutting ceremonies or be decorated with the works of world-renowned artists. Instead, they’re in windowless rooms populated with IT support staff and logbooks.
But these seemingly boring improvements, combined with new technologies from IBM’s Watson artificial intelligence division, could be what really propels this hospital into becoming one of the first of a new generation of world-class cancer facilities.
Once upon a time, in the late 1990s, cancer was broken down generally by what organ it was affecting. You had a tumor in your lung; you were treated, as best as possible, for “lung cancer.”
But in 2003, after more than 13 years and $3 billion, a group of scientists working from 20 research institutions cracked the human genome, mapping out the DNA code to the human body with complete accuracy for the first time. Since then, researchers and doctors have begun to use the human genome to more specifically define types of cancer and to devise disease treatments based on a person’s DNA instead of simply where the cancer is in the body. For example, if a drug has proven successful at treating kidney tumors caused by a certain gene mutation, and a doctor discovers a tumor on someone’s lung caused by the same sort of genetic mutation—well, then, maybe that drug would work for that type of lung cancer, too.
This sort of DNA-based medicine is only possible due to the precipitous fall in the cost of genomic mapping over the last decade, from about $10 million per person in 2007 to less than $1,000 today. The US National Institutes of Health (NIH) argues that this price drop is ushering in an age of “personalized medicine,” where genomic information will soon allow doctors to treat diseases as they’re directly affecting a person, rather than continuing to paint with the broad strokes they rely on today.
That said, mapping, or sequencing, someone’s genome still creates a massive data management problem: Running a person’s DNA through a modern sequencer spits out a raw data file of roughly 200GB. Researchers then need to sift through what is essentially an endlessly long spreadsheet to see if there have been any mutations in the person’s genes. It’s the perfect task for a machine-learning system—like IBM’s Watson—that excels at pattern matching.
IBM launched its Watson commercial division in 2014, three years after the original machine-learning system defeated two very smart humans on Jeopardy!, and roughly nine years after a group of researchers at IBM’s facility in upstate New York first had the idea to build a computer that could beat people at the quiz show. What started as a PR stunt to show that IBM was still at the forefront of artificial intelligence research—that it hadn’t lost a step since its Deep Blue supercomputer beat grandmaster Garry Kasparov at chess in 1997—has grown into a multi-pronged business for IBM.
The Jeopardy!-winning technology was based on a relatively simple set of ideas: IBM realized that to win, it needed to build a computer that could understand the wordplay often inherent in the game’s questions, and recognize that the answers might come from a myriad of different information sources. IBM built a system that could grasp the rules and style of the quiz game, find relevant information, corroborate it across multiple sources, and provide answers to questions in a succinct manner. The result was Watson, which they named after the company’s founder, Thomas J. Watson—not after Sherlock Holmes’ sidekick, as people often believe.
Winning Jeopardy! was a massive publicity triumph for IBM. For a while, Watson was everywhere. And the researchers behind the technology believed it could be expanded into something that might be useful to IBM’s business, beyond hype.
The company, which has seen its revenue decline for 20 straight quarters, is in the midst of a massive readjustment of its core business, shifting away from mainframes to cloud services. Ginni Rometty, who has been attempting to orchestrate a turnaround at IBM since she became CEO in 2011, saw the potential in Watson. IBM began to advance Watson’s ability to find patterns, signals in the noise, and to understand what people are saying when they write or speak. Soon after winning Jeopardy, IBM announced it would start developing commercial products out of Watson, and hoped to have something within two years. In 2012, IBM partnered with Sloan Kettering, a cancer-treatment and -research center based in Manhattan, to create a “decision-support tool” to aid doctors in making real-time treatment decisions based on the most up-to-date cancer research.
Today the tool is used by doctors across the globe, and helped ensure the remission of the cancer afflicting T.J. Richard, a retired police officer in Florida. He was referred to the Jupiter Medical Center in Florida and its chief oncologist, Abraham Schwarzberg, in March 2017 after a scan found a mass in his lung. The hospital had recently started using Watson’s support tool, and after Schwarzberg’s team diagnosed Richard with lung cancer, they saw it as a perfect opportunity to test the AI.
“It knocked me in the dirt,” Richard says of his diagnosis. But Watson “took a lot off my mind,” Richard adds, “it gave me a sense of well-being, knowing it didn’t take long to get a handle on what needed to be done.”
Schwarzberg says Watson acts as an immediate second opinion for patients. “It’s an unbelievably powerful tool,” he says. When a patient comes into his facility, his doctors gather together to discuss the cancer in question, working as a group to determine what they believe to be the best treatment. This process is called a tumor board, and Schwarzberg’s team has brought Watson into their workflow. “It’s presented as if ‘Dr. Watson’ was in the room—like another voice in the room,” he says.
IBM started using Watson for genomics in late 2014. Over the course of 2015, IBM researchers worked with 20 institutions to understand and identify what oncologists were seeing when they analyzed and diagnosed a patient’s tumor. “By the end of the year, Watson could analyze in minutes what doctors did over the course of a week,” says Steve Harvey, IBM’s vice president in charge of Watson’s genomic work.
Over 14 million people worldwide are diagnosed with, and 8 million die from, cancer each year, according to Cancer Research UK. Rometty has called Watson Health her “moonshot” and IBM has said it chose for Watson to tackle cancer, as opposed to any other disease, specifically because it is so varied and personal, and so difficult to treat. We know how to manage other global killers like heart disease, but curing, or even uniformly treating, cancer has proven difficult for decades. Not only would demonstrating that Watson can guide cancer treatment like nothing else on the market be a strong sales pitch to hospitals, it might also stoke the imagination of business leaders in other industries about what difficult problems Watson could solve for them.
In 2016, working in partnership with the University of North Carolina’s health-care center, IBM had Watson read medical scans that had been sent to the center’s doctors—it was able to identify the types of tumors with complete accuracy. This proved the system was ready to scale and be used by other hospitals, Harvey says. It triggered IBM to partner with Quest Diagnostics, a lab-services company that analyzes samples from medical centers across the US, Mexico, the UK, Brazil, and elsewhere, to bring Watson’s genomic tumor analysis capabilities to any doctor, regardless of where they practice.
IBM is not the only company using artificial intelligence or pattern-matching to tackle cancer—there’s Clearview Diagnostics, a New Jersey-based startup using AI to better diagnose and treat breast cancer, and Perthera Labs, a Virginia-based startup aiming to provide precision medicine through genomic sequencing, much like IBM. But in both cases, the companies have not fostered the level of partnerships that IBM has; they also do not have IBM’s access to a breadth of information from dozens of medical institutions, nor its history of data management or cloud services. That doesn’t necessarily mean IBM will be the company that brings AI to the mainstream of cancer treatment, but the company’s size and the sheer amount of knowledge it’s already ingested could give it a leg up.
In Omaha, researchers at Cowan’s facility have been putting Watson to the test. A team led by Babu Guda, a professor at the University of Nebraska Medical Center and the director of bioinformatics at Cowan’s cancer center, has sequenced tumor samples from 650 patients that have come through the center in the past. That’s done in the cancer center’s own sequencing labs, run by James Eudy, in the basement of one of its existing buildings. The unassuming lab holds a group of machines that look like a cross between 1990s-era desktop computers, and laserjet printers. They’re actually DNA-sequencing machines produced by Illumina, a California company that specializes in technology for genomic research. Lab technicians slice up tumor samples, and insert them onto a tray that slides out of the machines, rather like how CD drives used to slide out from desktop computers. In about two days, each tumor will be completely sequenced, and then the researchers upload them to Watson.
Guda’s Watson project—this is one of a few centers testing Watson in this way—resides on a few blades of a server rack in an old brick administrative building around the corner from the new facility. Once a tumor is sequenced and uploaded to the server, it can be run through the technology developed by Harvey’s team. Right now, Nebraska is just using it as a proof-of-concept, but if successful, it could upend how the hospital treats its patients in its massive catchment area. Cowan says his hospital’s reach for patients extends to Minot, North Dakota, over 600 miles to the north; there’s just no other facility in this vast piece of the country doing the level of research happening here.
With a sequenced tumor, Watson can rapidly recommend a raft of services that, in the past, would typically have taken doctors ages to surface. Watson has been trained on medical data from a range of medical journals and documentation of the real-world experience from dozens of research facilities (such as de-identified patient records and case studies), and so can provide a set of treatment options based on the patient’s specific type of cancer and the latest scientific thinking.
This is essentially what a doctor would usually do, but with the sheer amount of research being published, it’s becoming untenable for even the most studious doctors to keep up and still see their patients. It’s always been difficult for doctors to stay up to date with all the latest research in their fields—before the internet, it would’ve been difficult to even know about work that wasn’t reported in the journals you subscribed to—but it’s become even more challenging in recent years. The number of articles published per year is rising exponentially, up from an average of about 275,000 articles a year in the mid-1980s to 2.8 million by the early 2000s, according to the US Journal of the Medical Library Association. It would take a team of doctors, working around the clock at all times, to read and digest all those articles—but Watson can find surface new research in minutes.
The government estimates there are nearly 15 million people in the the US living with cancer, but according to the US National Institutes of Health, only roughly 3% of cancer patients ever make it into a clinical trial that could potentially benefit them. Not every cancer patient would benefit from participating in a trial, but every patient should be able to know that every avenue of treatment has been explored for them, whether that means looking at a trial, finding out about new treatment options, or considering new drugs that have just come onto the market. But because of the thousands of clinical trials for cancer running in the US at any given time, and the thousands of new medical papers generated each month, it’s a daunting task for any doctor to keep up. And so, what often happens is those patients who happen to live near centers of research, or have some other means or personal or professional access to research facilities, are more likely to hear about and get access to trials or research than those who live in the middle of nowhere.
Cowan told us that the third-largest “city” in the state is the football stadium at the University of Nebraska on game day. Roughly 90,000 people pour into a giant bowl in the town of Lincoln, decked in red and white, to watch football, flooding in from the disparately populated state. Roughly 1.9 million people live in all of Nebraska—about as many as live in Manhattan—with more than half residing in and around Omaha. According to Cowan, there are some counties in Nebraska that are so sparsely populated that have fewer than seven people per square mile. And according to the the US Census Bureau, there are a dozen counties with only a few hundred residents.
The level of expertise, institutional knowledge, and new treatment options available to the average Nebraskan diagnosed with cancer is nowhere near what’s accessible in research centers like New York, Boston, and Washington, DC. “It’s not fair that because of who they know or where they live that they have access to things I didn’t have growing up in southern Ohio,” Harvey says.
Watson also has data from the thousands of clinical trials running in the US at any given time.
And with a patient’s health records and specific genetic mutation, Watson can see if there are trials running anywhere in the US for new treatments for that subset of cancer, and help a doctor put together the applications required for the trial, both for the research facility, and the patient’s insurance company. Watson is helping ensure that information debt is no longer the reason patients—whether in Broken Bow, Nebraska, or New York’s Upper East Side—don’t make it onto potentially life-saving trials.
Most of Watson’s oncologic abilities are easy for just about any hospital to adopt; the process essentially boils down to integrating a new IT system. Watson’s most potentially ground-breaking skill, however, could require a major change in the way doctors approach treatment today: Genetic data parsed by Watson can generate radical treatment ideas.
In a strange twist of fate, leukemia oncologist Lukas Wartman was diagnosed with leukemia in 2011. After standard treatments failed, his colleagues at Washington University sequenced his genome. After running it through the school’s supercomputer, they discovered that one of his genes was acting erratically, causing his body to produce a protein that stimulated cancer growth. A new drug had recently been shown to cut down the body’s emission of this protein, but Wartman’s insurance had only covered the cost of it for use on kidney cancer. Wartman paid out of pocket, started taking the drug, and, as he told a group of journalists at a Watson conference in New York in 2015, his cancer has been in remission since. He’s currently advocating for more patients to be able to treat their cancer genomically.
“When you are dealing with cancer, it is always a race,” Wartman said at the time. “Unfortunately, translating cancer-sequencing results into potential treatment options often takes weeks with a team of experts to study just one patient’s tumor and provide results to guide treatment decisions. Watson appears to help dramatically reduce that timeline.”
Earlier this year, the US Food and Drug Administration (FDA) approved a drug called Keytruda to treat any solid tumors with a certain genetic biomarker. It was the first time a drug had been approved for a genetic mutation, rather than for use on a tumor in a specific organ. It’s a step towards a new paradigm where cancer will be targeted based on its—or the patient’s—genetic makeup, and tools like Watson will allow doctors to find that information.
When Wartman’s genome was sequenced just a few years ago, the process took up the capacity of a supercomputer. Guda’s system is running on a server farm, and it likely won’t be long before doctors anywhere can use the processing power found on business-grade cloud servers to analyze a tumor, instead of having to set up their own costly servers.
It will still probably be some time before drugs are prescribed based solely on the results of genetic sequencing. The FDA requires clinical trials, and peer-reviewed research before it’ll approve a new drug, or approve an existing drug for a new indication. Although if studies like Guda’s can show that cancer is a genetic disease, perhaps it won’t be too long before more drugs like Keytruda are approved for molecular families of cancer. “Things move fast when society realizes [they] makes difference,” Cowan says.
In the near future, hospitals like Cowan’ will be able to operate in a hub-and-spoke model. If a cattle rancher in the middle of the South Dakota badlands comes into a regional medical clinic for tests, and a tumor is discovered, his data can be shared with a partner hospital like the University of Nebraska Medical Center. There, experts can identify clinical trials or new treatments available through research from renowned cancer institutions like the Mayo Clinic or Memorial Sloan Kettering—all without the rancher ever leaving the Mount Rushmore State.
Watson essentially uses its pattern-matching abilities to sift through every relevant piece of medical literature that has been fed into its database. That includes information from places like Sloan Kettering, the American Society of Clinical Oncology’s medical journals, hundreds of Elsevier textbooks, data from the European Bioinformatics Institute, the NIH’s clinical trials database, FDA drug labels, and dozens of other sources, to find the most appropriate treatments. Watson also takes into consideration a patient’s medical history and weighs the options accordingly.
Once the tumor board reaches a consensus, it presents the proposed treatment plan to the patient. Schwarzberg says they run the Watson program again with the patient, providing printouts of medical information or papers the patient might want, to show that Watson’s wealth of information has essentially backed up the doctors’ decision.
“I would’ve thought there would be a challenge,” Schwarzberg says of explaining Watson to his patients, “but because of the Jeopardy! issue, and the chess, it’s so ubiquitous—almost everybody knows a bit of the story that it’s at least on people’s minds, even if they don’t fully understand what it means.” And unlike doctors, who may have preconceived notions about what treatments are best, or a hospital that might have contractual commitments to fulfill, Watson just presents information. “It facilitates the patient’s comfort that the decision was made in an unbiased way,” Schwarzberg says.
Richard says he’s been telling his friends about Watson; at least for now, the treatment his doctors chose—and that Watson backed—seems to be working: “I’m doing pretty doggone good,” he says. One of the drugs he’s on is Keytruda.
Watson is being used like this at dozens of hospitals across the world, IBM says, including in Manipal, India, the Bumrungrad International Hospital in Thailand, and hospitals across China. “It doesn’t matter if you’re in Boston or Birmingham,” Harvey says. Or Bangkok.
Treating cancer with AI is not a panacea, and Watson is not curing cancer; no AI can right now, and it may be a very long time before it could, if ever. But through a series of common-sense innovations, IBM has set up a system where any cancer patient can get the same information, and the same potential opportunities, to treat their disease as every other cancer patient—and can get these diagnoses and treatment recommendations faster than ever before.
“AI tools are changing the role of the physician,” says Andrew Norden, the deputy chief health officer for Watson Health and a 10-year veteran of the Dana–Farber Cancer Institute in Boston. Of course, even the most advanced technology won’t be able to replicate the passion or empathy a human doctor could have for their patient. “Cancer is not only a data problem,” Norden adds.
But Watson and similar AI tools will hopefully put doctors in a position to spend more time with their patients. Sifting through an ocean of medical journals, filling out insurance forms, researching clinical trials that might fit their patients—all of these processes should be automated in the near future. That’s not to say doctors won’t continue to spend a chunk of their days learning about advances in their fields, but it should mean eliminating a lot of the busywork currently keeping them from meeting and caring for patients. “It’s going to facilitate a physician being a physician,” Schwarzberg says.
When we were walking through the genomics lab in Nebraska, Eudy told us that the industry has jumped ahead much more quickly than expected. “I didn’t see this coming,” motioning to the Illumina sequencing machines, that cost roughly $900,000—far less than the $3 billion price tag of the original sequencer. Now, in a matter of days, a patient with a new cancer diagnosis can go from having little hope to understanding their complete genetic makeup, and discovering an obscure drug or a new trial that could help them beat it. It’s a revolution in the way we deal with the second-most common cause of death in the US.
“The future’s here, man,” Eudy said.