Watson's IBM Cognitive Computing, serving humans, Watson health, hopes and limits. – e-health innovation


This detailed note is based on notes we wrote on this topic and on the article "IBM pitched its Watson supercomputer as a revolution in cancer care. It's nowhere close "By Casey Ross @caseymross and Ike Swetlitz @ikeswetlitz qwe have translated, summarized and widely commented.

A smartbewell.com for years, we have been excited by IBM's journey at Watson, the spearhead of the application of cognitive computing in medicine, improperly called everywhere "artificial intelligence" (AI). We had interviewed Jerome Pesenti who had directed the work of Watson Health in New York. He pointed out one of the weaknesses of the system that continues today: The necessary provision of the ogre Watson of a Big Data composed of numerous patient files whose data must be well ordered.

In the post Artificial intelligence mobilizes against Alzheimer's disease: Interview with Jérôme Pesenti

To the question of smartbewell.com:

"The electronic medical record (in the cloud), does not exist (or very little) in France, so it can not be integrated into the AI ​​systems, what do you think the impact of this problem compared to the systems that you develop. (As part of the benevolent)

Jérôme Pesenti answered: "The electronic medical record is not a problem for Benevolent, but for Watson, it is a key problem. The ability to have a good electronic patient record, with good data in the cloud, is essential. In the US this is being done but there is still a lot of work to be done on the electronic file. The electronic file is fundamental for personalized and assisted medicine ".

We will see in this post that this problem remains current.

To fully appreciate this post it is useful to read also:

(IBM's artificial intelligence according to Watson)

After a sensational media entry on North American TV (Jeopardy in 2011), this spearhead of cognitive computing applied to medicine promised a rapid revolution in the use of medical data for the benefit of treatment including cancers!

This complex system brought to light all the new aspects of cognitive computing coupled with the power of storage and data processing in the cloud, the specialty of the new IBM company whose DNA had moved from hardware to AI and Bluemix

Watson must assist the doctor, simplify access to information.
These "core" components are speech recognition, machine translation, visual recognition, file analysis (Watson Analytics), statistical techniques, linguistics.

With contributions from Merge Healthcare a medical imaging company,
after partnerships, with ResearchKit from Apple, with Pathway Genomics, and a few billion dollars of additional investments in Phytel, Explorys, Merge Healthcare and Truven.

This accumulation of NICTs to allow decision support, with integration of medical record, genome, imaging analysis, an analysis of the world literature, an automated consideration of drug interactions.

The pioneering hospitals of Watson's use being the New York Genome Center and Memorial Sloan Kettering Cancer Center real incubator and flagship of what is now called Watson health.

But after 3 years of important investments and voluntary marketing, important nuances are heard, market penetration seems below expectations with only a few dozen user hospitals.

– Watson would have a hard time getting to know the different forms of cancer

– Watson was born American, and in non-US hospitals, his opinion would be skewed influenced by the specificity of patients and US treatment methods.

-Watson would have been marketed while the product was not completed, the doctors and researchers users find themselves in a context of Beta testers!

-The Watson for Oncology system, which has been in development for almost eight years, is still in its infancy, IBM described it as a digital prodigy, likely to bring new approaches to the treatment of cancer. In fact the capabilities of the supercomputer are limited.

– one of Watson's flaws is also that he needs to feed himself with a necessary electronic medical record and therefore clinical data.

-No one even claims that it is artificially intelligent only in the most rudimentary sense of the term, which is a shame for the system historically deemed to be at the forefront of AI (we confirm that the term is still inappropriate). where use of the term cognitive techniques).

At the Memorial Sloan Kettering Cancer Center in New York Dr. Mark Kris, Watson's lead trainer, even noted that an important scientific discovery resulting in a significant and abrupt change in cancer treatment could not be accounted for quickly by Watson!

-Watson has never been compared with traditional systems of diagnosis and treatment, IBM has published no scientific article in this context, Yoon Sup Choi a South Korean specialist in medical AI even declares that he is risky for IBM to conduct comparative clinical trials because Watson is already on the market and it is unlikely that a clinical trial will improve business prospects and that "If the result of the clinical trial is not very good there would even be a major business risk."

This is really cynical!

For Pilar Ossorio, a professor of law and bioethics at the Law School of University of Wisconsin Watson, should be subject to stricter ethical and scientific regulation, Watson should prove that it evolves safely and effectively for the patient.

Dr. Uhn Lee, Watson Program Manager, Gil Medical Center, University of Gachon, South Korea, said, "Artificial intelligence will be adopted in all medical fields in the future," so there is no need for preliminary studies. This may be true, but the relationship of interest is obvious here, see in this case a conflict of interest.

No publication in peer-reviewed journals exists. Other reports are done by paying customers or by IBM staff. Sometimes Watson is the subject of unpublished studies comparing his findings with conventional diagnostic methods. Thus in an unpublished Danish study, the agreement rate between Watson's therapeutic proposals and the protocols usually adopted was only about 33%, so the hospital decided not to buy the system.
Researchers in Denmark and the Netherlands said hospitals in their country had not signed with Watson because Watson was too focused on the treatment preferences of a few American doctors and their approach could be different from the usual ones local and therefore not universal.

This is one of the paradoxes that this type of system will have to overcome. It is presented as being based on LE Big Data Clinical (Patient Records) and Scientific (LA World Literature, Evidence Base Medicine (EBM)). In fact, in the end, it is a small group of doctors who decide (with the patient's consent) on the treatment to be offered.

We clinicians are not surprised, the Evidence Base Medicine that seemed to impose on our brains is often relativised, criticized and therefore adapted when it comes to making the final decision. Indeed the specific experience of the doctor often corrects the data of the EBM (this is often true in surgery). It is well known that the results of the best (methodological) randomized study may ultimately be contradicted by another study. I will mention for example a randomized study comparing two surgical techniques. The study can show that intervention A gives better results in terms of postoperative chronic pain than intervention B. The study being perfect with regard to the choice of patients, the methodology and the choice of statistical tests … .

But this study was conducted in a country where the surgical culture specific to these two interventions has nothing to do with that which exists in another country. And in this other country the same study with the same specifications gives an exactly opposite result.

In fact the hands that performed the interventions in both countries were not comparable !!

I think this is the same problem that is being raised by those who have failed or who have refused to use Watson in oncology, like the hands of our surgeons, the brains of New York researchers are different from those of Texas (see below). ), Denmark or the Netherlands ….

This does not call into question the interest of these systems in medicine but it shows that the delicate and non-universal point is the interface Machine / Doctor / Patient.

To illustrate this, it is interesting to mention Dr. Jeng-Fong Chiou, deputy director of the Taipei Cancer Center of Taipei Medical University, who started using Watson oncology with patients and who tells us that Watson is easier to use in countries like Taiwan where doctors have been trained in the United States or use treatment guidelines similar to those of Memorial Sloan Kettering physicians (the famous Admiral ship). But he adds that there are differences between American and Taiwanese patients with for example the need to use lower doses of drugs to minimize side effects. The economic cultural context, the health system can also radically change the "New York" recommendations of Watson

Interestingly, Nan Chen, who heads the Watson for Oncology program at Bumrungrad International Hospital in Thailand, says his oncologists are applying Japanese, not American, guidelines for the treatment of gastric cancer. It is true that the epidemiology of gastric cancer is very different in Japan or its frequency is higher than in the US. The experience of this cancer is very different and Asian professionals are inclined to follow their own recommendations and do not see the benefit of entering their data into Watson to get their opinion that would be either redundant or different and that they would judge any useless way!

In contrast Chen highlights the major interest of Watson in under-medical countries. Citing the case of UB Songdo Hospital in Mongolia, the hospital in the Mongolian capital has no specialist in oncology. Doctors follow Watson's recommendations in almost 100% of the time. Without Watson patients would be treated by GPs with little or no training in oncology.

And Chen adds that "This is the kind of situation that dreams IBM". Watson shows all his power and hegemony in this situation where he has no interlocutor and contradictor.

This is obviously in complete contradiction with the principles of Watson who presents himself as a collaborator and who should not simply substitute for doctors as in this experience in Mongolia.

But the context here is very different, Watson allowing to propose treatments in the absence of specialist. Studies should show the impact in terms of public health of Watson in this configuration.

An experiment in South Korea, described by Dr. Taewoo Kang shows some limitations of Watson's advocating for example a type of chemotherapy for node-negative breast cancer by supporting his opinion on a study demonstrating the efficacy of this treatment for patients with lymph node spread! In South Korea this opinion of Watson is argued and refereed by the specialist.

What would be the case in Mongolia in the absence of a specialist? Probably unnecessary chemotherapy would be proposed.

Non-independent opinions (Watson, Merck) and unsupported by scientific evidence, however, highlight the contribution of Watson to save time during research studies, highlighting in these studies results in terms of survival rates according to treatments which is very interesting for doctors in the choice of therapeutics (Chiou in Taiwan),

But this is very likely, we have already largely developed in the case of the company Benevolent who claimed to benefit greatly from new mathematics to develop and accelerate its pharmacological research.

Read Artificial intelligence mobilizes against Alzheimer's disease: Interview with Jérôme Pesenti

Another special experience in South Korea presented by Lee, the physician responsible for the Watson program at Gil Medical, allows patients to receive information directly from Watson, the information being argued by studies. In this context we find the intuitively virtuous schema of the tripartite collaboration between Physician / Patient / Algorithm.

Lee also highlights Watson's contribution to leveling the hierarchy of oncologists for the benefit of the youngest who can benefit from Watson's "support" against a different opinion from the senior specialist.

For having participated in congresses in Seoul and having observed the hierarchical relations between doctors of the same team and having perceived with curiosity the effect of the hierarchy on the behavior of the members of the teams and thus of the decision making, I think that it is obvious that Watson's contribution is strictly specific to a particular cultural context.

This underlines once again the lack of universality of ICT solutions.

Another example in India at Manipal Dr. SP Somashekhar states that in the vast majority of patients Watson's findings are comparable to the local committee, but the interest is the time saved with Watson, the committee studying only one in five cases, this saving of time is fundamental for the functioning of the service.

An interesting comment from Norden, a former Watson associate, points out the importance of feeding Watson clinical data (patient records) and clinical trials so that Watson can begin to identify the best treatments alone. This is what we recalled in the introduction with Jérome Pesenti, there is no effective machine learning without a solid big data very specific. Norden declares that for the moment he does not think "that a computer system is ready to be marketed without some human supervision".

This is disturbing when one knows that Watson can be delivered to himself in some medical deserts (Mongolia).

Martijn Van Oijen, an epidemiologist and associate professor at the Academic Medical Center in the Netherlands, said Memorial Sloan Kettering has high-level specialists but does not have a monopoly on cancer expertise. "The problem is that it's a US-based hospital, which has a different approach than other hospitals around the world," said Van Oijen, involved in a national initiative. to evaluate technologies such as Watson and yet convinced of the use of such systems to help oncologists.

IBM has indicated that it is pursuing studies to examine the impact of Watson but none has been completed to date.

To note a resounding failure in 2017 at the MD Anderson Cancer Center, the University of Texas, Houston Hospital was one of the first partners of IBM. He used this system to create his own oncology board, similar to the one IBM develops with Memorial Sloan Kettering. The University of Texas has canceled its partnership with Watson after more than $ 60 million in investment. Lynda Chin, a stakeholder in this project, explained that Watson is a powerful technology but that it is extremely difficult to translate into the field of care. She has encountered many obstacles, some of which have not yet been fully addressed by IBM at MD Anderson or elsewhere.

The unsettled faults in Houston are:

-The difficulty of correctly integrating the medical file of the patient for whom Watson is questioned. It is interesting to note that this problem is very frequently cited by specialists, again we repeat, the patient's data represent a weak link because they are still not well ordered and exploitable.

-Second problem, the difficulty of ensuring the safety of the treatments proposed by the machine and the problem of the responsibility "to guarantee the not to harm". We often find this fear especially at the level of the management of some hospitals for which the possibility of medico-legal problem slows down or cancels some projects.

-Third obstacle and most important for Lynda Chin, the difficulty of feeding Watson with enough patient data and from sufficiently different sources. The concept of machine learning is based on this concept of collecting a considerable number of identified and comparable clinical cases in order to propose a personalized response to a patient, we are still in the original sin of big data without coherence while it is necessary to the immediately exploitable smart data

IBM said the problem would have been solved following the work of Memorial Sloan Kettering, Watson Health's flagship, and that the solution is being used by more than 50 hospitals around the world and that clinical trials are underway. But still no known independent studies.

The special case of bladder cancer treatment has been studied at Memorial Sloan Kettering. It seems that the treatment proposals are specifically influenced by the experience of hospital doctors, and that in the case of some advanced forms of this cancer it is often very difficult with or without Watson to reach consensus. In addition sometimes the conclusions that were difficult to identify are simply erased because of a preference of the patient for a treatment different from that recommended first.

We think that in this context it is comforting and gives a lot of humanity to a process that seemed to be under the tutelage of the machine and the Doctor !! This is particularly well understood in France which is the country of the Kouchner law or to simplify the patient is the only master of his destiny. (Allow a little ethical reminder by emphasizing empowerment).

Watson seems to be at a delicate period of his career because it is a young business in full growth. For a former employee (Peter Greulich), IBM should significantly increase the investment in Watson in the image of the investments made in the 1960s for Hardware who had done the success of this company.

IBM, which claims to be the first company to have invested heavily in cognitive computing, is now facing fierce competition from giants such as Google, Amazon, but also Microsoft Unitedhealth (Optum), and hundreds of companies. companies and a galaxy of startups.

One example is Amazon, which has launched a health care laboratory, where it is exploring the possibilities of using electronic health record data and the potential creation of a virtual medical assistant. Here again we see the primary concern of properly collecting patient data.

Observers note that the acquisition of Truven for $ 2.6 billion to access 100 million patient records does not seem to have paid off and many work contracts were terminated, but for the benefit of IBM management, many hires in Watson Health units in Cambridge, in Massachusetts, and throughout North America.

The situation, as we discussed in this post and for many reasons, seems tense for Watson for Oncology, the wonderful work of Memorial Sloan Kettering is very promising but as Dr. Kris says Watson is still a medical student, he is learning to play in the real world.

Contrary to the elements of Marketing Watson is not necessarily the system that always has the right immediate solution to the questions asked of it.

The consequence in 2017 is the turnover of IBM's cognitive solutions division which has declined while Watson is expected to be the future of the company's business. Only financial services, worth $ 300 billion, are seen as a more important opportunity by society.

But that the cassandras who think that the NICT and thus the cognitive techniques are opposed to the doctors and want to impose themselves to the patients by dehumanizing our profession are for their expenses there is nothing for the moment.

And paradoxically it is perhaps because today "artificial intelligence" is of an extremely low level, and it still can not oppose the decisions of the doctor. She is always a cognitive supplement, an accelerator for the choices to be made and decisions to be made, the latter being always the responsibility of the restricted medical team surrounding the patient (except in Mongolia!).

To your remarks, reactions and information concerning this subject.


Leave a Reply

Your email address will not be published. Required fields are marked *