Intelligent system for diagnosing lysosomal storage diseases: analysis of verification results.

Accurate diagnosis of patients with genetically determined lysosomal storage diseases is a global challenge. Rare population prevalence, phenotypic polymorphism, and vague clinical presentations hinder early identification of these disorders. Clinical decision support systems can assist physicians during the pre-laboratory diagnostic stage. Purpose of the study is to verify the developed intelligent (expert) system for supporting medical decisions in the pre-laboratory diagnosis of lysosomal storage diseases. Materials and methods. Verification of the GenDiES expert system was conducted using retrospective de-identified electronic health records from 3 medical institutions across 3 regions of the Russian Federation. System success was defined as inclusion of the confirmed (true) diagnosis within the top 5 hypotheses (preliminary diagnoses) generated by the system based on patient phenotypic manifestations for subsequent laboratory confirmation. Results. Verification utilized de-identified data from 38 patients with lysosomal storage diseases: 33 mucopolysaccharidoses, 4 gangliosidoses, and 1 mucolipidosis. System accuracy reached 90% (95% CI [75%; 97%]). Analysis of 4 cases lacking the correct diagnosis in the top‑5 hypotheses revealed correct diagnoses ranked 6ᵗʰ (two cases) and 7ᵗʰ (two cases) due to atypical clinical presentations in these patients. Comparison with validation (electronic health records from an institution) demonstrates consistent performance across data sources. Findings. The GenDiES system for pre-laboratory differential diagnosis of lysosomal storage diseases achieves high accuracy (90%) with consistent results. It is suitable for use in specialized medical institutions during hypothesis generation prior to laboratory confirmation.