To the public, IBM trumpets its Watson supercomputer as the next big thing in medicine, a new kind of machine that melds human expertise with digital speed to give patients personalized treatment advice.
Meanwhile, in the halls of Congress, company executives have been delivering a blunter message: We will revolutionize patient care, so please get out of the way.
Like any new technology, Watson poses unknown risks; for example, what if its advice is wrong and harms a patient? But IBM argues that its machine doesn’t need to be regulated because it’s different from other medical devices. It’s not like a pacemaker or a CT scanner, so the company shouldn’t have to prove to the government that it’s safe and effective.
Now, as federal regulators prepare to weigh in on that issue, a STAT examination shows the lengths to which IBM has gone to shield its prized machine from government scrutiny.
The company’s fingerprints are all over legislation passed last year that exempted several types of health software from FDA jurisdiction. A former IBM executive helped draft the blueprint for the law. In the months before its filing, IBM hosted an event with the eventual bill sponsor to introduce Watson to influential members of Congress. And the company then deployed a team of lobbyists to press its position that Watson should be legislatively walled off from regulation.
IBM got most of what it wanted, though not the ironclad protection it sought from FDA oversight. In coming months, the agency will issue new guidelines clarifying which software products will be exempt under the new law, known as the 21st Century Cures Act, and which will still require some level of review.
The implications are huge for IBM, and it’s not standing idly by: Eight of its employees were registered to lobby on this issue during the first half of this year. The company also is pressing its positions through a newly formed “AI Caucus,” a group of lawmakers whose stated goal is to ensure that innovations in artificial intelligence benefit Americans.
Watson and similar “clinical decision support” technologies help guide physicians in their diagnosis and treatment of diseases. This new generation of machines promises to harness the power of big data to improve the care of patients, but it also blurs medicine’s traditional line between doctor and machine, posing new potential risks to patients.
Some doctors and consumer groups have argued that artificial intelligence systems such as Watson are exactly the kind of technology the FDA should be examining more carefully.
“Until we understand the basis for the AI and how the algorithms work, there really needs to be that third-party check and transparency,” said Dr. Reshma Ramachandran, co-chair of the FDA task force at the National Physicians Alliance. “We want to make sure there’s no harm done at the end of the road.”
The alliance was one of a handful of groups that raised concerns about the software exemptions included in the 21st Century Cures Act. Its lobbying was dwarfed by the efforts of IBM and other deep-pocketed companies and interest groups that spent large sums to sway lawmakers.
Between 2013 and June 2017, IBM spent $26.4 million lobbying Congress, the White House, and federal agencies, according to the company’s disclosure forms. It is unclear how much of that money was spent pushing for proposals to reduce regulatory barriers facing health software. IBM said in a statement that its advocacy on these topics “represented a fraction of one percent of our lobbying spending from 2013 to 2017.”
But the company declined to say exactly how much it spent or answer questions about whom it lobbied during that time period. It said it “was one of many organizations, including patient and physician groups, that supported a common sense regulatory distinction between low-risk software and inherently higher risk technologies.”
“Until we understand the basis for the AI and how the algorithms work, there really needs to be that third-party check and transparency.”
Dr. Reshma Ramachandran, co-chair of the National Physicians Alliance's FDA task force
IBM had one of the largest teams of lobbyists working the issue over the past four years, with 14 of its own employees registered to lobby on the software proposals, along with three outside firms. Their advocacy coincided with a massive marketing campaign featuring celebrities such as Bob Dylan and Serena Williams, in which the company repeatedly dangled promises that Watson would transform health care.
A recent STAT investigation found that one of Watson’s flagship products, Watson for Oncology, is falling far short of that goal. The system claims to recommend the best treatments for individual patients, but Watson is struggling to master the various forms of cancer and has only been adopted by a few dozen hospitals worldwide. Doctors and executives at some hospitals using the system don’t fully understand how it works, while others complained its recommendations are not easily generalizable to patients of different incomes and cultures.
IBM is also selling products that help match patients to clinical trials and provide cancer treatment recommendations based on genomic data, through a partnership with Quest Diagnostics.
Whether the Food and Drug Administration will regulate these products should become clear shortly. The agency has said it intends to issue guidance documents spelling out how it will implement the exemptions in the 21st Century Cures Act in the first quarter of 2018. The agency’s commissioner, Scott Gottlieb, has signaled a desire to streamline regulation of digital health technologies. He has suggested the FDA may use a certification process to allow companies to commercialize products without pre-market approval, or through a very limited review process.
Watson poses unique challenges for the FDA, which until now has used a clear dividing line to exercise its powers: It does not regulate the judgment of doctors, only the safety and effectiveness of medical devices and drugs they prescribe.
But systems that rely on artificial intelligence, by their nature, are meant to combine man and machine, presenting questions that test the relevance of the FDA’s framework. Dr. Meghan Dierks, a professor at Harvard Medical School and director of clinical systems analysis at Beth Israel Deaconess Medical Center in Boston, said the implications of this new union must be considered carefully.
For example, IBM executives say Watson for Oncology is designed to “democratize” medical knowledge and help guide doctors with less specialized expertise in treating cancer patients. However, in advising those doctors, Watson cannot fully explain the rationale for its decisions. It can cite medical literature, but it cannot explain why it selected a particular treatment for a particular patient.
So in those situations, where the dependency on the machine is higher, and the underlying rationale for the decision less clear, more oversight of the system might be warranted, she said.
“If how Watson got from input to output is not obvious to the end user, it’s a harder case to be made that the practitioner is independently making the choice and simply using this as an information tool,” Dierks said.
She served on a federal panel that advised the FDA on how it should regulate medical software and mobile health apps. The panel, known as the FDASIA work group, called for a flexible approach to these technologies, advising the FDA in April 2014 to clarify its stance on what types of products should be regulated.
IBM, which also served on the panel, disagreed with that position. The company supported the work group’s approach of assigning risk levels to various categories of software, and exempting low-risk decision support tools from review. But it wanted more clarity on the definition of low risk, and it wanted it to be codified in legislation.
“We believe that Congress should update the statutory framework to provide clarity that ‘health management’ or clinical decision support software should not be regulated as a medical device,” Tim Sheehy, an IBM lobbyist, wrote to the FDA in July 2014.
By then, the company was already pushing a bill known as the Software Act.
The bill, filed by U.S. Rep. Marsha Blackburn, a Tennessee Republican, was modeled on a legislative framework created by a nonprofit Washington think tank, the Bipartisan Policy Center, which has received funding from IBM and an array of other technology and health companies, including Intel, McKesson, and Johnson & Johnson.
The drafting of the blueprint was led by Janet Marchibroda, a former chief health care officer at IBM. Marchibroda, who worked for the company in 2009, said IBM was an active player in the effort to develop the framework, but so were many other companies and stakeholders. She added that the group did not directly address questions posed by Watson and other systems that use artificial intelligence.
“At the time — no one was really working in that area to be honest,” Marchibroda said. “I mean Watson was beginning, but … I think our principles still stand.”
The FDA office in charge of device regulation opposed the Software Act, saying it could limit its discretion to balance patient safety against the potential benefits of new software. But IBM was among the earliest and most ardent backers of the bill.
Zachary Lemnios, vice president of research strategy at the company, told the House Energy and Commerce Committee in November 2013 that Watson could “impact patient treatment and save lives” and that such software is “low risk” and need not be regulated. He said the Software Act would “ensure that U.S. consumers have access to the most innovative health care IT without being burdened by regulatory requirements that will not improve the safety of the products.”
Blackburn echoed those talking points two years later, when she addressed IBM executives during the company’s annual congressional fly-in, which brings more than 100 of the company’s leaders to Washington to lobby lawmakers. “Data is not a device, and it is not a drug,” she said. “It must be treated differently.”
Blackburn had also hosted a Capitol Hill reception with IBM in May 2013 to introduce Watson to members of Congress. She did not answer questions from STAT about whether she had additional contacts with IBM. In an emailed statement, she said the bill “was an effort to bring much-needed technology to those who would benefit most from it.”
The 21st Century Cures Act emerged in the spring of 2015, a massive bill that included billions of dollars in funding for biomedical research and provisions to accelerate the approval of promising new medicines. It also included the provisions of the Software Act and a similar bill, the Medtech Act, filed in the Senate.
Just a few months earlier, the FDA, in a guidance document on mobile medical applications, had suggested that it might finally begin to regulate some medical advisory tools like Watson.
But now, 21st Century Cures appeared to be taking most such products out of the FDA’s jurisdiction. The bill exempted from regulation health software “intended for use to analyze information to provide patient-specific recommended options to consider in the prevention, diagnosis, treatment, cure, or mitigation of a particular disease or condition.”
The legislation included an important caveat. It stated that the exemptions for advisory systems would apply only if the software enables a “health care professional to independently review the basis for such recommendations.”
Several lawyers who represent software makers said that language will be a major area of contention as the FDA decides how to implement the law, especially for artificial-intelligence-based products such as Watson, because it’s not obvious what “independently review the basis” means.
In 2015 and 2016, the voices in support of the bill far outnumbered the opponents. Among those raising questions were the National Consumers League and the National Physicians Alliance.
Ramachandran, of the physicians alliance, said she met with several lawmakers to convey the organization’s concerns about the legislation, including the bill’s sponsors, Reps. Fred Upton, a Michigan Republican, and Diana DeGette, a Colorado Democrat. On each visit, she noticed that she was far from the only one seeking to weigh in.
“For every one visit I could do per month, there was someone daily in their office from the industry side,” she said. “Their presence was quite apparent.”
More than 1,300 lobbyists pushed for passage of the the bill. On Nov. 29, 2016, the day before the House vote, IBM issued a press release to emphasize its strong support.
“IBM innovations are addressing the world’s most pressing health challenges through the use of data and cognitive insights,” IBM lobbyist Christopher Padilla said, alluding to Watson. “To realize the full potential of technology in health care, America needs a more modern regulatory framework. This is what the 21st Century Cures Act delivers.”
The bill passed overwhelmingly, and it was approved in the Senate a week later. President Obama signed it into law on Dec. 13.
Less than two months later, IBM announced its first deal to sell Watson for Oncology to a U.S. hospital, Jupiter Medical Center in Florida. In a statement, IBM portrayed the timing of the announcement as coincidental.
Meanwhile, some who worked on the bill are closely monitoring how the FDA applies it to existing technologies. Kristen O’Neill, a health policy advisor for U.S. Rep. Raymond “Gene” Green, a Texas Democrat who co-sponsored the Software Act, said IBM was one of the “really interested parties” lawmakers consulted during the effort to pass the bill.
But she said the bill was not necessarily intended to exempt technologies like Watson, especially since the supercomputer is used to recommend treatments for seriously ill patients.
“Someone needs to be making sure that’s working as it should be, safely and effectively,” O’Neill said.