Watson-oncologist: not everything is so rosy
How Dr. Watson Couldn't Beat Cancer
Semyon Kvasha, Copper news
A few years ago, IBM began selling Watson, a cognitive system consisting of 90 servers with eight–core processors and a total amount of RAM of more than 15 terabytes. Watson is promoted in almost any field where you need to work with large amounts of data, computing speed and complex algorithms. On the robot's website there are sections "health", "education", "finance" and so on.
But almost immediately, the company declared Watson a doctor and set him the task of helping doctors around the world to make oncological diagnoses and prescribe treatment.
To do this, computers trained for two years, fed them 605 thousand medical documents, 25 thousand medical histories, processed the giant archive of the Memorial Sloan Kettering Cancer Center research Center in New York. As a result, the robot prescribes the correct (that is, corresponding to modern recommendations) treatment in 90% of cases.
The Stat publication conducted a large investigation of the current state of affairs of the doctor and found that everything is a little less rosy than in the IBM and Memorial Sloan Kettering press releases.
How does Watson work and how much does it cost?
In essence, it is a huge, very smart and very fast-running database. The nurse gives the doctor the patient's medical history (theoretically, he can read them himself, in practice, abbreviations, errors and medical handwriting require controlling this process and typing in a significant part of the data manually), he compares it with his huge library of scientific articles and clinical recommendations and after clicking on the "Ask Watson" button gives his clinical recommendations based on the best modern clinical practices, the most up-to-date research and other achievements of evidence-based medicine. For this, hospitals pay from $200 to $1,000 per patient. Sometimes you also have to pay extra for consultations and connecting the system to the hospital's medical history database, this also costs some money.
Where do they put the Watsons?
In the USA, this system is installed in two hospitals. There are more of them in India and Southeast Asia. There is one such computer in Mongolia. They are also found in Europe, however, there they are treated without much optimism, according to STAT.
In America, only two computers have been fully adopted, and 50 supercomputers are employed in the oncology program around the world
What is the complaint against Watson
The main problem is that Dr. Watson has never passed clinical trials. There is not a single independent study of the robot. All the articles that talk about its effectiveness are written by the system's clients, and an engineer from IBM is necessarily listed as a co-author. This is not good for modern evidence-based medicine. However, engineers from IBM say that some things do not need clinical trials. "Nobody's going to do a double-blind randomized parachute test, right?".
A good result for Watson is a high level of coincidence of the therapy prescribed by him with that prescribed by a doctor in an advanced hospital. This level of coincidence reaches 96% in various medical institutions (but not in Denmark; in Danish hospitals, doctors conducted their research, found out that their level of coincidence is only 33% and did not install a supercomputer). But it bothers doctors: why do we need a supercomputer that will prescribe the same treatment to the patient that they would have carried out themselves, or one that they have already considered and rejected?
Doctors outside the United States also pay attention to the fact that Watson's recommendations do not take into account the peculiarities of national health and medical practice. In South Korea, the robot often prescribes treatment that is not covered by the national insurance system, in Taiwan, it is customary for the patient to prescribe smaller doses of drugs, preventing side effects.
Where do Dr. Watson's recommendations come from?
As STAT correspondents found out, ordinary doctors think that Watson is a collection of the best clinical practices and research from around the world. In fact, all the data (including medical articles from journals) and all the supercomputer algorithms are driven in by employees of Memorial Sloan Kettering Cancer Center. And these are two problems at once: firstly, not all scientists and doctors agree with the approaches of this research center and not all consider it an absolute authority in oncology.
Secondly, the amount of data that Watson operates on is not really very large, and it would be much more if they were really uploaded to the system by all hospitals connected to the neural network. "Let's say you have 10,000 lung cancer patients. It's actually not very much. If there were more of them, you could see patterns, groups of patients who respond or do not respond to therapy in a certain way, who have certain toxic reactions. This would make it possible to create a more personalized and accurate medicine. But we can't do that if we don't have a way to collect this data," says Dr. Lynda Chin, who was in charge of installing and training Watson at MDAnderson Hospital in Texas before she quit, and this organization stopped cooperating with IBM.
What's good about Watson
Ideally, the system works in countries where there are no oncologists at all. In Mongolia, for example, there is a UBSONGDO hospital in Ulaanbaatar, where 100% of appointments go through a robot (and appointments are made, if possible). Otherwise, patients would be treated by general practitioners without oncological experience, and so the most advanced American clinical experience is available to patients.
In hospitals that have Watson, the pressure of old experienced doctors on young people is eased – because any experience and any authority can be checked very quickly by finding a suitable article with the right conclusions: Watson is really looking for literature very quickly and needs data from this literature. Dr. S.P. Somashekhar says that for the absolute majority of cases, the recommendations of the program coincided with the recommendations prepared by the oncological council of the hospital. But there are twenty doctors sitting on the council, who usually sort out each case for a week and then discuss it for an hour. Watson saves the hospital the work of 20 doctors during the week, and this is incredibly valuable.
The Doctor is not alone
IBM's robot was the first such supercomputer to enter medicine. Now there is also Google DeepMind, which is used in London hospitals for the diagnosis of eye diseases and for reference help in the treatment of certain oncological diseases. Amazon is launching a digital medical laboratory to work with big data – electronic medical records and is going to create an electronic assistant for a doctor, that is, it immediately sets modest goals and is going to play exactly on the field in which Watson is being used right now.
But this, like surgical assistant robots like DaVinci, will still be just another, albeit incredibly convenient, tool to help a doctor, and not an incredible IT revolution in healthcare.
Portal "Eternal youth" http://vechnayamolodost.ru
20.09.2017