Patients talk about their illnesses, treatment and side effects in various online forums, blogs and social networks. However, until now, this broad wealth of experience has not been available to medical practitioners and pharmaceutical researchers. But at the same time, these communication fl ows have also served as fertile ground for misinformation. Computer linguists from the University of Stuttgart's Institute of Natural Language Processing (IMS) are working on ways to automatically search networks for content of this kind and to structure it.
“Watch out!”, the Twitter user warns all those following her updates on her microblog network: “I might be grumbling about pain more often in the next few weeks”. She goes on to say that she is switching from one painkiller to another. For Dr. Roman Klinger, Senior Lecturer at the IMS, tweets like this are extremely interesting: he is developing ways to automatically search information exchanged in social networks for biomedical knowledge and to link it to what is known from medical research, whereby he and his team have to deal with two completely different challenges.
A combination of expertise and emotion
First, there is medical expertise, which, says Klinger, is usually hidden away in the form of data that is not easy to access. Researchers, medical practitioners and laypeople nearly always have to plough through vast amounts of literature to collate all relevant information pertaining to a given illness. “If I, for example, want to know which proteins or genes are known to play a role in the emergence of some type of cancer, then there are databases for that”, Klinger explains. “However, the most recent findings are only ever available in scientific publications.
Next, Klinger wants to link this data to subjective information relating to people’s personal experiences. In the past few years, researchers have been trying to fi gure out how to determine the type, cause and objective of emotions in texts. So Klinger started to ask himself: “What information pertaining to biomedical knowledge is there on social media”? Sufferers generally talk about their illness and medication in emotional terms. “Emotions can be implicitly formulated in a number of very different ways”, Klinger explains, citing the aforementioned Twitter user, who suffers from neuralgia, as an example. Just four days after her fi rst tweet, she wrote: “The pain is much better. But the insomnia is getting worse every night”. Even this short period is interesting for Klinger from an analytical perspective: “It demonstrates that she's expecting rapid results”. The fact that the drug did work quickly, he goes on, in turn enables an analysis of the emotions involved. “On the face of it, her statement is an endorsement of the new medication as well as a deprecation of the old one. But then she talks about a side effect – insomnia. That casts doubt on the new drug”.
What was previously hidden is now in plain sight
However, before all this it was important to clarify who sends out tweets or posts blogs about diseases. One of Klinger’s master students looked into this. “We developed a process with which we can Followers of a special kind Computer linguists analyze what patients are reporting on the Internet automatically identify the category to which a given author belongs. Is it a doctor, a patient, a relative or perhaps and industry expert”? Building on this, Klinger now wants to come up with a set of rules for analyzing the actual contents. Of interest in this context are the disease profi le and resulting medical circumstances, the drug taken and any side effects that may have been experienced as well as the patient’s feelings about them. “We want to be able to recognize all these things in social media and, in about two years’ time, to be able to link the statements to scientific texts”, Klinger states. “Then, not only could you find the scientific findings relating to a given illness in the databases, but also information provided by those affected”.
The twitter feeds from the above-mentioned neuralgia suffers demonstrate why that would be useful: she was advised to take yet another drug to counteract the insomnia. “When you analyze conversations like this on a large scale”, Klinger says, “you discover which drugs are being combined and why”. To date, he continues, practically no research has been carried out in this area, and certainly not involving texts written in German. The relevance is obvious In this way, side effects could be identified about which doctors and pharmaceutics had known nothing before. In more drastic cases, the research being done at the IMS could even save lives, as Klinger demonstrates with another example. In 2014, the current President of the USA, Donald Trump sent out a tweet about a healthy child that allegedly became autistic after being inoculated and that he knew of many such cases. “Statements like this are very common. We can make it our task to ensure that, whenever anyone tweets anything like this, we can supply the facts”. For, it has been scientifi cally proven that there can be no link between inoculations and autism. “Fact checking has been around for a while – but not for pharmaceutical drugs and illnesses. And, there is no program available that automatically collates false information of this kind with what has been published in the scientific literature”. Not yet.