Humans to help machines better serve humans
Medgate is a a globally operating Digital Health company that provides services to numerous health insurers in Switzerland.
Medgate realized that the future of their industry was Digital and decided to create an artificial intelligence powered symptom-checker able to guide patients to the right caregiver. Medgate began its digital journey with IBM in 2016 and selected Open Web Technology as a key strategy and technology partner since 2017. Read our dedicated case here.
The system is live since November 2019 and has already helped and triaged thousands of patients.
While the chatbot-like initiative is delivering impressive results, Medgate doctors are always pushing for better performances.
Indeed, patient safety, well-being, and satisfaction are at the core of their belief which is why Medgate decided to invest in a solution that allows them to inspect and improve the knowledge acquired by the algorithm.
In particular, one of the challenges faced by the machine program is to absorb the knowledge of the millions of historical medical cases recorded by Medgate over the years.
This task of grouping annotations together is difficult for a machine because semantically different medical concepts may mean the same thing in essence. For example, pain in the head and headache could be used to describe cephalalgia, but migraine might point to a slightly different concept.
Doctors were helping machines better absorb these millions of data points using processes based on Excel.
Enabling doctors to operate on the brain of a machine
To allow our client to refine how the triage system interprets data, we developed an intuitive symptom management portal able to import and export content from and to the model.
Doctors are able to list, filter, and compare all symptoms extracted by the machine learning powered system. The dozens of parameters available for each symptom can be fine-tuned. Medical concepts can be split or merged together.
This important work helps the algorithm better understand what belongs together and what doesn’t so that the similarities between the medical cases can be computed correctly and better performances can ultimately be achieved.
Translators are also able to work directly in the tool, or import/export content to standardized formats and work using their own methodology.
This technically allows Medgate to release their product into any language of the 3’995 written languages that exist on earth!
Open Web Technology is grateful to Medgate, IBM, and its team of consultants and technical experts for the trust, the great collaboration that helped make this project successful.
Tanks to our learnings we could also design and develop within a week the "Coronavirus assessment bot"
When applying machine learning in health care, a pure naive data-driven approach alone is too limited. Yet incorporating structured medical knowledge is complex and time-consuming. Proficient expert telemedical physicians are typically not also data scientists! The AIDA Ontology Management Tool enables our clinical experts to directly visualise and tune key aspects of the AIDA system learning pipeline. In order to achieve this, Open Web Technology had to first understand both the challenge at hand, and the way doctors think, and then conceive and build an effective solution. We are now beginning to work with the tool, and the first results are promising!
Anthony Dyson, CIO
- Deliver Blueprint of the solution. In particular, mockups and specifications to mitigate uncertainty and ambiguities.
- Collaborate with Medgate’s medical affairs and IBM to establish needs, technical possibilities and limitations.
- Implement and deliver project, increment after increment, following the Scrum Agile framework.
- Frequently take feedback from all stakeholders into account and adapt project based on newly discovered requirements and scope changes.
- Three servers, Test, Staging and Production to ensure proper development and deployment cycles.
Doctors can display and filter the thousands of symptoms available in the system.
Doctors can update dozens of attributes attached to every symptom. They are able to refine the behavior of the system using their medical expertise.
Organize medical concepts
Doctors are empowered to group and separate medical concepts that belong together or don’t.
Available symptoms (7’000 are available to patients)
Displayable and editable symptom attributes
Possible languages to translate the AI system into