POS1551-HPR MEDICAL EXPERT KNOWLEDGE MEETS AI: HOW EXPERT KNOWLEDGE CAN IMPROVE SYMPTOM ASSESSMENT APPS - A NEW APPROACH IN RARE DISEASES

Abstract

BackgroundBy definition, rare diseases occur in fewer than 5 in 10,000 people. However, in Germany alone, there are approximately 4 million people affected (1). Due to their rarity, rare diseases are often overlooked by general practitioners with limited knowledge about them. There are often only a handful of specialized experts for each condition. Yet, those experts are frequently not evenly distributed across the healthcare system, and often, patients lack access. Therefore, the time to diagnosis is often long and poses many challenges. Artificial intelligence (AI) approaches, such as those used in symptom assessment apps can potentially help to detect disease and thus, shorten the time to diagnosis (2).The general approach underlying the AI of symptom assessment apps is to gather medical knowledge from data (e.g. electronic health record or literature searches). However, for rare diseases, there is only limited research available. In this study we used a new approach: to abstract medical expert knowledge by conducting guided interviews and transforming them into clinical vignettes.ObjectivesThere are two objectives. First, we aimed to integrate expert knowledge on the lysosomal storage diseases (LSDs) Fabry, Gaucher and Pompe into the Ada symptom assessment application and use this expert knowledge to optimize Ada’s LSD condition models. LSDs are of particular importance as they represent import differential diagnoses to rheumatic diseases.Second, we will conduct questionnaires with patients and LSD experts, comparing the optimized to the previous condition models’ performance. We will investigate whether the novel approach of guided interviews, in combination with literature research, results in a better performance than literature research alone.MethodsOur novel approach is to curate expert medical knowledge from guided interviews with medical experts. The interviews aim to gather knowledge on the symptom constellations with which patients typically present to their physicians. This knowledge is subsequently used to create prototype clinical cases and transform them into structured case vignettes. The rare disease structured case vignettes can be readily transcribed into Ada’s knowledge base.ResultsWe conducted guided interviews with clinical experts from the Medizinische Hochschule Hannover (MHH) to create clinical vignettes for the LSDs Fabry, Gaucher and Pompe disease. We conducted interviews with four medical experts and created a total of 11 clinical vignettes: five vignettes for Fabry disease, four for Gaucher disease and two for Pompe disease. Figure 1 demonstrates the vignette creation process. Then, in combination with systematic literature searches, the vignettes were used to update Ada’s existing condition models for Fabry and Gaucher disease and to add Pompe disease.Figure 1.The novel medical model creation approach uses guided expert interviews to create prototypic clinical vignettes, which can be readily transcribed in a medical modeling language to create individual disease modelsConclusionParticipants will complete two assessments: an Ada version with the old knowledge base and an Ada version with the updated models. The study plans to enroll 15 LSD patients - five per condition - and nine LSD experts. For the conditions Fabry and Gaucher disease, we will ask the participants to rate both Ada versions. For Pompe disease, we will ask participants to rate the latest Ada version with the updated knowledge base. This novel approach has various clinical implications, including potentially shortening the ‘time to diagnosis’ for rare diseases, thus giving patients faster access to the treatments they need.