Predictive analytics health research institute set up

Kira Radinsky and Morris Kahn Photos: E.Yizhar, PR

The Israeli institute has already developed technology for predicting colon cancer among Maccabi Health Fund patients.

Three leading parties in Israel have joined forces to set up a digital health research institute at a cost of $6 million: billionaire, philanthropist, and biomed investor Morris Kahn; predictive analytics pioneer Prof. Kira Radinsky; and Maccabi Health Services. Maccabi Institute for Research and Innovation head Prof. Varda Shalev is managing the institute, which has already developed technology for predicting colon cancer among Maccabi patients who never considered going for a scan for this purpose. The prediction is based on the routinely accumulated medical information about patients. The institute is now conducting studies in order to personally adapt drugs to patients with hypertension, and is slated to address other diseases later.

Kahn says, "Israel has an extremely special advantage in digital medical information. Our health funds have collected unusual information, both because patients tend to stay with the same health fund for many years, and because the health funds underwent the digital revolution at an early stage. Israel has excellent doctors and excellent information analysts. Thanks to this combination between information that exists nowhere else and the specialties needed to use it, we can create here a system of information and prediction accessible to every researcher in the world, who can conduct research quickly and easily. Israel will get international credit for this - for the benefit it will bring to mankind. The system we develop with Maccabi will first help Maccabi patients, but also far beyond that."

Radinsky: "Through ventures of this type, we're making our country a leading center in medicine."

Varda Shalev says, "At this very moment, our doctors are spending time with patients and improving our information. We were so alarmed about putting that delicate meeting between a doctor and his patient on the computer. We now realize that the computer brings so much added value to that meeting. Doctors can't do everything."

At a conference held when the research institute was launched, Radinsky explained the conceptual breakthrough that must take place in order to enable the algorithms operated on medical information to predict the future. "I began to deal with this question 12 years ago as a Microsoft researcher. By examining correlations in Internet searches, we succeeded in discovering side effects of drugs and formulating their sequence. For example, we discovered that if a person is taking a given drug, and then experiences a headache accompanied by nausea, he should go to the doctor to make sure that something worse won't happen."

Radinsky was later asked by Google to use search data to predict outbreaks of influenza epidemics. "The algorithm worked, until it stopped working. Why? Because one day flu made the news, and people started to taking an interest in it and to conduct searches even if they weren't sick. That's the problem with correlation-based research - you don't know how to distinguish between cause and effect."

In order to predict cause and effect, Radinsky says, it is customary now to consult specialists. If you ask a human doctor why there was an increase in Internet searches for influenza during a given period, he can tell you right away, "Because it was on the news." "Our challenge is to also find the cause and the effect in the information itself," Radinsky says. How can we do that? By adding the specialists' knowledge to the information that the system reads. For example, you can let the system "read" newspapers or articles in which the specialists' fundamental assumptions appear, and to construct hypotheses from within them, which can then be tested against the information. "That's how we send our system to medical school," Radinsky explains.

Exposing an outbreak of Ebola

One example of how the system uses people's information to make hypotheses and to add to the knowledge of human beings is the way Radinsky traced the reasons for an outbreak of Ebola. "From the data, the system realized that when a certain country is looking for gold and diamonds, it causes forests to be burned, and animals to move. What is the connection? Only by reading articles about the possible sources of the Ebola infection can the system come up with the hypothesis that cutting down the forests really causes the migration of bats from the forest to settled locations, and that cases of Ebola were caused by eating uncooked meat from a bat. We went back to the descriptions of the case and the doctors, and verified the hypothesis with them."

In another case, the system found a correlation: cholera appears several years after a drought. Does a lack of water cause cholera? An analysis of the semantic information reveals that cholera is transmitted in water. It appears that the shortage of water makes water become polluted more easily, or pushes the local residents to consume polluted water. What is the solution? To send clean water to countries in which there is a drought.

In a study by the intensive care department at the Rabin Medical Center (Beilinson Hospital), by analyzing thousands of correlations between variables, Radinsky's system and her team discovered to what degree nutritional quality predicts the survival chances of an intensive care patient.

Radinsky told "Globes, "My participation in the research institute is part of my pro bono activity, in which I am a visiting professor at the Technion Israel Institute of Technology." Radinsky's main job is as a data scientist at eBay and eBay Israel's chief scientist. "Maccabi has accumulated an enormous store of information from millions of meetings between patients and the system. We're starting with hypertension, one of the diseases whose cost to the health system is among the highest in Israel. There is a real dilemma about which drugs should be used to treat each patient. The selection is currently made by trial and error. We predict for each patient what the two or three best drugs for him are, and increase the probability of achieving good results by 150%. We're also working simultaneously with both the system and the doctors, with the doctor confirming each finding. In one case, we proposed a drug for a diabetes patient, even though it was forbidden for use by diabetics. We didn't have this information."

After the algorithm for predicting the best treatment for hypertension is developed, it will be easier to apply it to other diseases, such as urinary infections and diabetes. Radinsky notes that at eBay, one of the challenges, but also one of the advantages, is the fact that the sellers describe their products in free text. In the medical world also, doctors write in free text, and a profound connection therefore exists between her two activities.

"Treat my patient with compassion"

Senior doctors who attended the event were asked to give their expectations of the system. Unexpectedly, they did not mention concern about being flooded with information and lawsuits backed by large amounts of information (although there is no doubt that they are concerned about this). They noted that they expected the system to give them more time, which they could devote to patients. "When I don't have to act like a computer, I have a return of compassion for my patient and a return of my passion to be a doctor," said one doctor, and the other agreed with him. The doctors also said that this system would free them from dependence on information supplied by drug companies.

Shalev: "Some of our planned projects also include image analysis and an analysis of the connection between the image and the text. For example, when a patient comes to a doctor with a clinical complaint and an ultrasound image, the doctor does not record all his complaints and all the possible findings in the image; he records only what seems important to him at the time. A computer system can document everything, and then search for valuable connections in all of the information."

"Globes": How do you make sure that the doctors are not swamped by the information?

Shalev: "Today, the doctor is already being flooded. We give the doctors tools to manage the information, and we're planning to also give the patients tools, so that they are better prepared for a meeting with the doctor."

Maccabi director general Ran Saar adds, "In another decade, the machines will handle the easy medical cases. The doctors will be kept for more complex cases, to which they can devote more time."

Radinsky says that the doctor must be the final authority in order to avoid crude errors by the machine.

Shalev: "That's why a researcher can't just buy a database. He has to be close to the information. If you see that the algorithm doesn't work, you have to go back to the patient himself. It's an interactive process, from the data to the situation, and back to the data."

Are there already business concerns using the system?

Saar: "Regulation isn't ready for that yet. We believe that in a few more years, the world and Israel will be ready to take the next step."

Kahn: "I believe that it will happen even sooner."

Published by Globes [online], Israel Business News - www.globes-online.com - on January 19, 2017

© Copyright of Globes Publisher Itonut (1983) Ltd. 2017

5 Comments
View comments in rows
Update by email about comments talkback
POST
Comments
Kira Radinsky and Morris Kahn Photos: E.Yizhar, PR
Kira Radinsky and Morris Kahn Photos: E.Yizhar, PR
Twitter Facebook Linkedin RSS Newsletters גלובס MAD Conference 2017