Authors (including presenting author) :
Chan TCH(1), Mak CM(1), Woo PPS(2), Song FE(1), Chan FCH(2), Chan GPY(2), Pang TLF(2), Au BSC(2), Chong YK(3), Lam CW(4)
Affiliation :
(1) Chemical Pathology Laboratory, Department of Pathology, Hong Kong Children’s Hospital, (2)Statistics and Data Science Department, Hospital Authority (3)Chemical Pathology Laboratory, Department of Pathology, Princess Margaret Hospital, (4)Chemical Pathology Laboratory, Department of Pathology, Queen Mary Hospital
Introduction :
Patient registries are crucial for rare disease management. However, manual registry construction is labor-intensive and often not user-friendly. Our goal is to establish a first local artificial intelligence (AI)-assisted patient identification tool for data mining in electronic patient database, and to explore its application in building the Uncommon Disease Database (UDD) for 45 inborn errors of metabolism conditions.
Objectives :
1. Development and validation of AI tool for accurate patient identification in electronic patient databases of Hospital Authority
2. Building patient registry for rare disease – the Uncommon Disease Database (UDD)
Methodology :
A list of candidate patient is first selected by structural data in laboratory information system, which involves IEM biochemical and genetic tests performed during a ten-year period from 2010 to 2019, followed by feature extraction with unstructured data in electronic clinical notes including discharge summaries, out-patient consultation notes, laboratory, and radiology reports from Clinical Management System. The AI-assisted patient identification tool was developed with a rule-based natural language processing algorithm. Patients identified by the algorithm are curated by Chemical Pathologists who were specialized in inborn errors of metabolism for accuracy check. Secondary ascertainment with another independent known cohort of IEM patients was also conducted.
Result & Outcome :
Out of 46,419 patients with IEM-related tests, the algorithm identified 100 as “IEM-related.” After pathologists’ validation, 96 cases were confirmed as true IEM, with 1 uncertain case and 3 false positives. Secondary ascertainment yielded a sensitivity of 92.3% compared to the independent known IEM cohort. A model of UDD was established for 45 inborn errors of metabolism conditions.
Our AI approach provides a novel and efficient method to identify IEM patients, facilitating the creation of centralized rare disease patient registry. The approach can be extrapolated to the establishment of other patient registry with slight upgrade of the current algorithm, enhancing healthcare management for rare diseases.
