Sigmoidal Success Story
Reduction in the cost of drug safety surveillance systems
more accurate ADR identification via Social Media
What was the business problem?
Although the safety of drugs is tested during clinical trials, many adverse drug reactions (ADR) may only be revealed under certain circumstances, for instance: long-term use, when used in conjunction with other drugs, or by people excluded from trials, such as children or pregnant women.
Nowadays consumers often report their experiences with ADR on social media instead of traditional channels, which makes drug safety surveillance systems less efficient.
What savings and profits does the client achieve?
Platform development in collaboration with Sigmoidal enabled Novartis to monitor the long-term side effects of drugs and opinions on social media. The system was equipped with the ability to automatically generate an executive summary, utilizing a Natural Language Processing solution.
As a result, more efficient identification of novel adverse drug reactions was available, reducing the overhead cost of the drug safety surveillance systems.
How did we accomplish it?
Thanks to our solution, drug safety surveillance systems can be easily improved by incorporating the knowledge extracted from social media into them.
Our system enables fast scanning of various data resources, such as Twitter, Facebook, or forums posts.
Then, using state-of-the-art Natural Language Processing techniques it filters out irrelevant information. The remaining, relevant data is processed and certain entities, such as drug name, ADR, pharmaceutical company name, person age, and gender, are extracted and analyzed.
The last step is automatic report generation. The executive summary containing all found ADR and insights on how they occurred is created and could be used in further drug development.
How did we boost the project with ?
Sigmoidal engaged Enterprise Machine Learning patterns. We rebalanced the representation of Design patterns resulting in exceptional predictive performance choosing the right performance metric and sampling method.
We used the Event Storming approach for the purposes of Process Modelling of the view-model rebuild challenge, allowing complex domains to be designed and understood.
Sigmoidal focused on Accuracy and A/B testing before release, resulting in easy migration of a machine learning model to production by separating inputs, features, and transformations.
Sigmoidal employed Senior staff training extending their capabilities and business applications of the technology, and its future potential, avoiding common pitfalls.