The Healthcare Battle
Pharmaceutical companies discover the possibilities of artificial intelligence. But Internet giants such as Google and Amazon have long been competing with them in the healthcare market. Insights into a competition for a resource that has become very valuable.
The “Sense Bridge” of the pharmaceutical company Novartis in Basel is a large room with tables and chairs arranged in a semicircle around six large screens. It looks just like air traffic control. Colored dots, bars, curves and maps light up on the screens. They give an overview of the majority of the 440 clinical trials that Novartis is currently conducting worldwide, in which active ingredients are tested on patients. The company uses pattern recognition and machine learning (artificial intelligence /AI) to evaluate the data obtained.
For example, it can draw on the results of a total of two million patient years of clinical trials. This is the total time period in which all study participants were analyzed: This includes scans, videos, chemical data or data on one and a half million molecular compounds. The Sense Bridge brings together previously separate information and gives it new meaning.
It takes an average of ten to twelve years for a potential active ingredient to go through all the clinical studies and reach the market. This can cost a pharmaceutical company up to two billion dollars. Approximately only one in ten investigated substances receives approval and makes it to market. The Sense Bridge is a kind of early warning system, and the technology pays off. “We can see the current status of each study at a glance in traffic light colors, and the AI forecasts its further progress,” says Michael Bartlett, Head of Clinical Strategy. “Red means a high risk that the schedule will not be met. Green means everything is going well, yellow is somewhere in between. We can zoom deep into the data for each study until we find the causes of the problems”. In addition, the AI suggests, for example, a suitable patient group for a planned study that might be particularly responsive to a particular drug, as well as countries where comparable studies have worked well.
For a long time, the hype about AI had been a thing of the past for the pharmaceutical industry, but since projects like Sense Bridge have a direct impact on research success, study duration and sales, at least some companies are turning their attention to digital data. Novartis wants to become a data group, Hoffmann-La Roche is pursuing a similar strategy. The industry has become nervous since IT companies such as Google, Amazon, Apple, Microsoft and IBM have set their sights on the healthcare market. These companies have made AI algorithms market-ready and have the necessary technology. However, these companies still lack health data and medical expertise, but it is only a matter of time before they have both. Then they want to put the pharmaceutical companies in the second row in the healthcare sector.
The market is attractive. In 2017, healthcare spending as a percentage of gross domestic product was 17.9 percent in the USA, the frontrunner, and 11.5 percent in Germany. But in the future, the money will be earned less and less with drugs and radiological equipment and more with software and data. AI supports, for example, diagnoses and therapies. AI-based patient apps document medication intake, symptoms and side effects or give tips on how to deal with diseases. Health apps such as WebMD, Medisafe or Ada Health are on the rise.
Pharmaceutical companies are now just as aggressive on this topic as the IT competitora. Hoffmann-La Roche, for example, has developed a digital tumor board that combines data from various specialist areas for each patient, which doctors need when deciding on a therapy. Oncologists will be able to use it to gain an overview within minutes of what previously took them an hour. Nevertheless, the industry does not want to talk about competition with IT companies. “We see above all opportunities for new applications, not competitive pressure,” says Achim Plückebaum, head of the future project “Data42” at Novartis. But he admits: “Of course, we are talking about data sciences here, and IT companies have this expertise. But we have the data and the competence in molecular research. With this advantage we can establish new approaches in medicine”.
These new approaches have one thing in common: they require a lot of data. Whether it’s a diagnostic aid, drug research, study management or patient apps – AI algorithms are based on logic and statistics. In diagnostics, they can only distinguish between sick and healthy people if they have as much information about both as possible. The algorithms search for a pattern that makes the difference between sick and healthy. The developers of AI often do not know what this pattern is, but AI can optimize itself in the field of machine learning. Researchers often just enter data, check the results and determine when they are meaningful.
“Running the algorithms over data is the easiest thing to do,” says Regina Barzilay, a professor at the Massachusetts Institute of Technology and a pioneer in the use of machine learning in breast cancer prognosis. “The main obstacle in our research is indeed the lack of availability of qualitative data in large quantities”.
The major breakthroughs in AI first occurred in image recognition, because from 2009 onwards with the Imagenet a public data set was created. It currently contains more than 14 million images of everyday objects that volunteers have described with metadata such as “cup” or “pencil”.
With this data, research has made immense leaps forward within a few years. With the help of Deep Learning, the AI taught itself to improve. It learned to recognize patterns in countless sample images and thus to identify cups and pencils in previously unknown images.
Such data sets are not available in medicine. If they were available in large quantities, for example on different stages of tumor diseases, the tumor of a new patient could be better classified with the help of AI. “These data sets exist, but only a handful of researchers have access,” says Barzilay, “and even if these people are brilliant, there are too few. If the data sets were opened up to everyone, we would achieve an immense amount. The pharmaceutical companies are among the guardians of the data, which is an exclusive resource for them.”
Some research institutions are already building up large databases. The German Center for Neurodegenerative Diseases (DZNE) in Bonn, for example, is conducting a study in which up to 30,000 middle-aged people from the Rhineland region are medically screened for a total of around eight hours – every three to four years, over several decades. The aim is to obtain a comprehensive collection of health data from a cross-section of the population. Some of the participants will probably later develop a so-called neurodegenerative disease such as Alzheimer’s, multiple sclerosis or Parkinson’s disease. With the help of AI, the researchers can find differences between the sick and the healthy as early as possible in the lives of the study participants.
“In the case of neurodegenerative diseases, research has long focused on the protein deposits in the brain,” says Joachim L. Schultze, genome researcher at the DZNE. “We now know that immune cells accelerate the disease, but we still do not know the causes of the diseases.” The more data available, the sooner they can be found. “We often hear complaints that we collected too much data,” he says. “But we need them, because they are a key to better therapies. As soon as the symptoms of Alzheimer’s appear, a large proportion of the nerve cells degenerate. That’s why treatment must start earlier, at the time the disease develops.
Since the researchers do not know what they are looking for, they are not only collecting blood and saliva samples and brain scans, but also lifestyle data. Because here, too, they have a major deficit. Nutrition, sporting activities, environmental influences or the daily stress level of a person – information about these is more likely to be available from IT companies.
It is not unreasonable to assume that a company such as Amazon would be quite happy if its customers did not immediately consult a doctor in case of discomfort, but instead consulted Amazon’s digital assistant Alexa. Ideally, she knows both the customer’s medical history and their lifestyle. She asks him questions and suggests a diagnosis. If it’s nothing serious, she can provide him with an over-the-counter medication via Amazon, possibly one that the company produces itself.
So far that is speculation. But Amazon is in the process of building up a team of experts who are familiar with Digital Health, the interdisciplinary combination of health and digital technology. The company has recruited Maulik Majmudar, a cardiologist and lecturer at Harvard. Also Martin Levine, former Iora Health, health service manager and provider of a digital patient platform. Also included is Taha Kass-Hout, who worked for the U.S. Food and Drug Administration as an information technology executive.
The driving force behind this is Amazon’s Vice President Babak Parviz. He holds a doctorate in electrical engineering and is an associate professor at the University of Washington. A few years ago, he developed a contact lens that records wearer health data, such as glucose levels. At Amazon, he sees the greatest potential in speech recognition, in Alexa. In 2017, for example, the company announced a competition with the pharmaceutical company Merck & Co. to initiate a so-called Alexa skill for diabetes patients.
A skill is similar to an app for smartphones, except that it works purely linguistically. Since more than 400 million people worldwide suffer from diabetes, such an application could bring the US company a lot of additional customers. The Alexa Diabetes Challenge was endowed with prize money of USD 125,000. One participant, the start-up Glooko, submitted a program that regularly asks patients what they have eaten and whether they have taken their medication. This would help patients get their disease under control – and Amazon’s AI would receive a lot of training data from the medical community.
But that’s not enough for a company that likes to offer everything. That’s why it is selling his first health products under its own brand “Basic Care”, including over-the-counter drugs such as Ibuprofen. Amazon also invested in Grail, a company that collects genetic information from cancer patients. It specializes in Next Generation Sequencing, a gene-analytical procedure that allows DNA molecules to be sequenced in parallel – which saves an enormous amount of time. Grail plans to evaluate hundreds of thousands of genome data. Amazon hopes that Grail will use Amazon Web Services for these large amounts of data. The U.S. National Cancer Institute already stores its 2.6 petabytes of genome data from cancer patients in the Google Cloud – at a price of around 19 million US dollars.
Alphabet, Google’s parent company, is just as busy. Its research team regularly publishes studies that demonstrate applications for deep learning in medicine. For example, the technology is designed to predict how long patients will stay in hospital and the likelihood that they will have to return after discharge.
Alphabet subsidiary Calico uses AI to research aging processes and age-related diseases. Deepmind, which also belongs to the holding company, has recorded initial successes in the prognosis of protein structures. The algorithms calculate which gene sequence produces which structure and create three-dimensional models from it. Once it works reliably, this would be an immense step forward and the basis for new drugs.
At first glance, the many approaches and the business models they make possible differ from one another, but the principle is always the same: data must be collected, purchased or reprocessed so that digital technology can recognize patterns in it. The Chinese Alibaba Group, also a pioneer in AI research, joined forces with L’Oréal in spring to launch a mobile test application for acne.
And this is the third drawback of the pharmaceutical industry. Li Ma, Vice President of Strategy and Investment at Alibaba Health Information Technology, says: “Many pharmaceutical companies are now also trying to involve patients. But it’s difficult because they don’t know exactly who their patients are”. They are taking drugs, but it is mainly the doctors who have been generating sales in the industry so far. This is changing with digital medicine – its revenue drivers are the patients or consumers who want to stay healthy. Amazon and Google have long been at home with them – and are accessing their data. The pharmaceutical companies, on the other hand, lack access.
“For them, there has never been the pressure to know the patients,” says Jan Ising, head of the Life Science business unit at Accenture. “The pharmaceutical industry now needs to build offerings, including through strategic partnerships within a new ecosystem. It has rested on its expertise for too long.”
The realization that pharmaceutical companies must become service-oriented companies has only grown in their minds in the past two years. Ising says: “CEOs are now realizing that doctors and patients have become digital. The patient leads a digital life, and companies must embrace this if they want to reach him or her, otherwise they will lose a big business”.
This text was published in brand eins.