Treato.com, a social health website, was initially conceived after high-tech veteran and entrepreneur, Gideon Mantel, experienced a frustrating, real life event. Gideon’s daughter had torn her ACL playing basketball, and he went online to look for other patients’ experiences with this type of injury to learn how they coped. He couldn’t find any relevant posts. It took him three days of intensive searching to find a post describing an identical injury and the medical procedure that helped. That was a moment of epiphany for Gideon. He couldn’t possibly understand why it had taken days for a web-savvy professional like himself to find any valuable medical data online. Inspired to change this, he decided to develop a tool that would enable people to view other patients’ medical experiences.
You are a high-tech veteran, with many years in the industry (Commtouch). Aside from your personal epiphany, what attracted you to this project?
Over the last two decades, the internet has changed pretty much every aspect of our lives: the way we work, shop, commute and communicate. The medical field, on the other hand, has seen only minor changes – definitely not the paradigm shift we witnessed in almost any other industry. We still go to the doctor when something’s wrong, get a prescription, take it to the pharmacy and receive our medication. From the patient’s perspective, this process has basically remained unchanged over the last 20 years. In that sense, I tend to view healthcare as the “internet’s final frontier.” There’s an unmet need to leverage the wisdom of crowds via social media to empower patients – just as the Web has empowered consumers in almost any other field. The second answer to your question about why I went into healthcare has to do with the scope of this market. I can’t think of a more profound way to impact the lives of countless people. At some point in life, every person becomes a patient or a caregiver to someone close. What we bring to the table, so to speak, is therefore relevant to virtually every person.
Who is the intended end user of this site? Researchers, patients, health care professionals?
All of the above – and many more. In fact, any person or entity along the healthcare value chain can gain a higher level of understanding of the market from listening to patients – and that’s something that no one else is currently doing, at least not on such a scale. And we’re not talking only about healthcare professionals. Think of Wall Street analysts covering pharma companies- since we monitor patient discussions on new drugs still in clinical trials, we can provide valuable insights on the market potential of these drugs before they even hit the shelves. That said, our first target is the patients, who currently have the least amount of access to information, and therefore the site is directed at them.
There are no ads or sponsors on the site. What is your business model? Do you have premium users?
At present, we are aiming to introduce our service to as many patients as possible, and we will keep Treato.com 100% free of charge and of ads. Our business model relies on future partnerships with entities along the healthcare value chain that have a constant need to understand patients’ motivations. We will provide them with high-level analytics and tracking capabilities, based on analysis of millions of online discussions. We’re also considering releasing a premium version of our site in the future, enabling users to access additional information, such as drug ranking, for a modest fee.
I tried Treato and searched for information about the drugs I currently use. I got a large amount of related posts (for Novorapid I got 7,680 posts and for Novolog – the U.S. version I got 18,156). Even after refining my search I was left with hundreds of posts. How do you categorize the posts?
Each of the billions of posts we analyze is categorized by the main topics that are discussed in the context of the mentioned drug. These are typically side effects of that drug, as well as indications of switching between drugs. Almost any search of a drug brand sold in the U.S. would yield a list of the ten most discussed topics in regards to that drug, ranked by the number of discussions for each topic. This gives you an immediate, high level indication to what patients are saying about that drug. You can always click on each of the topics and delve into actual posts talking about it, in the context of that drug. The comparison feature allows you to quickly compare the leading topics between competing drugs and learn about the most complained side effects for each of them.
Could so much information be too much information? Do you expect patients to read hundreds of posts to learn about the medications they receive? How can they asses the value of these posts?
The power of our system is in the aggregation of billions of posts. Each individual post may yield little relevance and validity, but when so many of them are analyzed, the user can get the big picture of what patients are saying. Think of it as if you’re staring at a large impressionist painting from up close: at first glance you only see a smudge of color, but as you take a few steps back you understand what you’re seeing. So, to answer your question, there’s no need to read all the posts, or even a fraction of them. But having so many discussions analyzed definitely helps to provide accurate findings, based on experiences of millions of real life patients.
How big is the corpus of scraped comments and sites?
We scan, analyze and index over one billion posts from thousands of medical forums and blogs. We constantly add additional sites and new posts to keep our knowledge base up to date. Posts can be retrieved from as early as 2000 – even pages that were removed and taken offline can still be viewed on our system.
What is the current semantic analysis framework? That is, are sources classed based on lexical analysis alone (i.e., the presence of “Lantus” and “Levemir” in a post implies it is about switching from one insulin to another), or is more complicated structural analysis done (i.e., “I was using Lantus but am now using Levemir,” is classed differently than “I am using Lantus but used to use Levemir”)? Alternatively, are labels applied using some sort of machine learning technique? Supervised or unsupervised?
Our semantic framework is based on a proprietary Natural Language Process (NLP) algorithm that was developed here at First Life Research. This technology is able to deconstruct complicated sentence structure to extract its meaning, so that a certain post would be classified as a switch from drug A to drug B, while other posts containing the same words in different order would indicate the opposite switch. This is quite a challenge, because there are many ways to say essentially the same thing – and that’s just in proper English. At the same time, we’ve calibrated our system to extract the correct meaning of slang phrases, so that posts including the term “awfully well” would indicate a positive reaction, for instance.
Reactions to medication often vary according to phenotypic or genetic differences in patients. Is there any way at this point to try to correlate those with the statements made about the medication? Do you anticipate building a system that will allow users to begin to proactively supply such information, as the genomic analysis site 23andme.com does?
Our focus has always been on understanding the big picture of what millions of patients are saying – as opposed to looking into a single individual patient record. By leveraging wisdom of the crowds we help empower countless individual patients, making them aware of side effects of every drug and its alternatives, enabling them to make better healthcare decisions.
A number of companies are working in the reverse direction– type in your symptoms, see what the problem might be (for example, WebMD and medgle.com ). Why did you decide to go from medication to side effect, rather than side effect to medication? Do you see companies that go the other direction as competitors or potential collaborators?
Keep in mind that all the sites that offer symptom analysis are based on official medical sources – essentially, medical textbooks. We’re taking an entirely different approach: listening and understanding what patients are saying about the drugs they’re already taking, with special emphasis on the side effects of each of these drugs, as well as complications due to interactions between a few drugs and diseases. Add a drug comparison tool that highlights the leading side effect of each drug in a click of a button, based on the aggregation of millions of patient discussions, and you get a powerful decision making tool that no “self-help” medical site can offer today.
Clinical trials over the years have accumulated massive amounts of patient-reaction data that allegedly sits on paper in filing drawers at the FDA. Several analytics companies have in the past met with success making Freedom of Information Act (FOIA) requests for similar types of documents from other governmental agencies. That’s no small task, clearly, but it does build a big proverbial moat around your company. Is anything being done to pursue this sort of historical trial data?
We take a different route. Instead of going through a legal struggle to obtain documents from clinical trials, we turn to the place where patients are already sharing their experiences from all drugs, including those currently being tested in clinical trials. Keep in mind that the data collected in clinical trials is miniscule compared to all the information shared online by millions of patients about medications and their side effects. Our approach is that as long as the social media data is thoroughly collected and carefully analyzed – we can unveil key issues on virtually any drug and make these insights accessible to the public.
Do you have any case studies of how Treato’s data has helped someone so far?
Sure do, and from first-hand experiences. Patients who were granted access to our system before it was officially launched to the public have used it to look up their medications and see what other patients are saying about them. In numerous cases, the first people to test our systems have stumbled upon surprising insights on medical complications they have been suffering from. For example, people taking cholesterol reducing pills learned that the pain they have been experiencing in their neck and shoulders can be attributed to this medication. In some cases, this connection is not a known side effect – although prevalent in hundreds of patients, as can be seen on Treato.com. The same goes for a blood pressure lowering drug, that turns out to be attributed to a persistent cough, and ADHD medication that leads to “cold fingers” sensation.
Additional evidence comes from a seasoned family practitioner that happens to be one of our company’s advisory board members. A few years back, he was examining an 11-year-old asthma patient treated with Singulair. Her parents were frantic to see him, after the girl started expressing suicidal thoughts. When the doctor conducted a quick check on Treato.com, he discovered a few hundred incidents of Singular patients complaining about suicidal thoughts, or family members of these patients highlighting this disturbing occurrence.
In another case, a patient taking anti-inflammatory medication complained to our doctor about insomnia and major sleep disorders he was suffering from. In all the medical literature the doctor examined there was no link between sleep disorders and the type of anti-inflammatory drug this patient was taking. As in the previous case, typing the name of this drug in Treato.com yielded results from patients around the world who were complaining about the same side effects after taking this medication.
These two cases demonstrate the potential of transforming anecdotal evidence provided by patients into real medical insights. Forum discussions may not meet any scientific criteria; however, repetitive complaints from many patients about an unexpected side effect can definitely provide initial data that will draw physicians’ attention.
Karmel Allison contributed to this interview.
