Routine care data analyzed using explainable artificial intelligence identify best therapeutic sequence of 2nd line biologic treatments after a 1st line with Anti-TNF or Vedolizumab in inflammatory bowel disease
Inflammatory bowel disease (IBD) includes Crohn’s disease and ulcerative colitis, two conditions characterized by chronic inflammation of the gastrointestinal tract, which can damage the bowel and cause serious disabilities.
Treatments only have a suspensive effect, and the therapeutic sequences are currently based on reimbursement criteria or the prescriber’s expertise, in a “trial-and-error” approach rather than on validated scientific evidence.
Routine clinical care data from 1,190 IBD patients and 1,269 variables collected by a private tertiary care center dedicated to IBD patients, using the customized Instamed® platform, includes comprehensive information on the patient’s treatment history, clinical variables, and disease activity scores, in particular the Harvey-Bradshaw Index (HBI), Partial Mayo Score (PMS) and Simple Clinical Colitis Activity Index (SCCAI).
KEM® (Knowledge Extraction and Management) is an explainable Artificial Intelligence (xAI) platform using Formal Concept Analysis (FCA) that systematically extracts and evaluates all relations between variables in a database, thus enabling the identification of patient subgroups with greater likelihood of response to treatment sequences.
This powerful “proof of concept” preliminary study shows a setup using routine clinical care data from an initial set of 1,190 IBD patients and the KEM® xAI platform to provide actionable hypotheses supporting therapeutic decisions for disease management.
These initial results will be further improved and validated as we accumulate real-world data on new patients and regularly complete the follow-up of patients already included.
The flexibility of our platform will enable us to include more therapeutic sequences and to add all the new treatments available in current clinical practice.
This data will lead to the building of a therapeutic decision support tool which will gain robustness with time and accumulated data.
This will allow us to offer continuous, up-to-date support for difficult medical decisions, enabling patient management to be guided by results observed in real life and leading de facto to digital precision medicine.