SURVEY – The start-up Kap Code offers manufacturers and health agencies a pharmacovigilance tool based on the monitoring of social networks. It is based on complex algorithms.
Imagine. Your wisdom teeth have been removed. In order to limit the pain, your doctor has prescribed codeine paracetamol for you. The pain subsides, but it is replaced by persistent nausea. Of a tweet angry, you are complaining about the medicine. The gesture is almost mechanical in the age of social networks.
But soon, that message could be picked up by an army of algorithms, decrypted and exploited for commercial purposes. It is already a reality. Several companies take advantage of these valuable testimonials. Among them, Kap Code, a French start-up chaired by Stéphane Schück.
Health authorities but also manufacturers are interested in this sector for a very simple reason. This monitoring refines the detection of side effects. And this faster than the classic system. But this service comes at a cost: analyzing your data will pay off. Sometimes even at two levels, because some forums only open their doors for a fee. An approach that should prove to be lucrative.
On the basis of this information, drug stakeholders can indeed trigger risk management plans, launch more in-depth investigations… in short, act more quickly when faced with an emerging risk. And perhaps, at best, reduce the number of hospitalizations. Because every year, 18,000 deaths and thousands of hospitalizations are linked to adverse drug reactions.
Algorithms to track down the negative
“We considered that it was a source that had to be explored,” explains Stéphane Schück. Many people share their experiences with drugs, whether positive or negative. It is rather the negative that interests the company and its tool, Detec’t. It relies on various forums and social networks to assess the adverse effects of drugs.
Stéphane Schück, President of Kap Code: ” We have developed extractors that retrieve messages. They are analyzed after isolating them, and algorithms identify the medical concepts. “
26 million messages, concerning 485 drugs, were screened by the algorithms of the Parisian start-up. But this impressive result is the fruit of a long development work, resulting from complex algorithms. They will seek, directly on social networks, the information that is specified.
“Usually, we start the approach with key words, specifies Patrick Ruch, group leader at the Swiss Institute of Bioinformatics.. For example, I specify that I want all the tweets that contain the term paracetamol. The choice of the blue bird’s social network is not made at random. It offers free access to information, provided you have the appropriate IT tool.
Successive filters
But the objective is not to recover all the data offered by the site. Successive filters are applied, which increasingly specify the information to be recovered. “ The volume is much too large to consider rough handling, ”suggests Patrick Ruch. Once the filters are juxtaposed, the mass of information is more reasonable. Manual verification is even possible.
But Kap Code took the analysis further, using self-learning algorithms. “Once a pair has been established between a drug and a medical concept, the system estimates whether the medical concept is an undesirable effect or not”, summarizes Stéphane Schück. It is somewhat the same philosophy as epidemic tracking software, which has been in use for a long time.
And the applications of these devices are manifold. Not only is it possible to detect known side effects, but it is also possible to identify new ones.
Detect unknown signs
For example, “the frequency of drug interactions is never reported in official reports, and this information is not reviewed in clinical trials,” says Patrick Ruch. Thanks to algorithms, this information can be identified and distinguished from a more classic side effect. In real life, such testimonies can strongly influence practices.
Stéphane Schück hopes for even more from his tool: to identify hitherto unknown symptoms and set up enhanced surveillance. In other words, make social networks an alert system that triggers pharmacovigilance surveys. But this use remains unlikely. “For having worked on the subject, I consider it a source of additional information, but not isolated”, tempers Joëlle Micallef, stationed at the PACA-Corsica pharmacovigilance center.
.
