Improving the Diagnosis of Inherited Kidney Diseases Through Domain-Specific Databases

The diagnostic accuracy of phenotype-matching tools for rare diseases depends not only on the algorithm but also on the quality of the disease definitions. Inherited kidney diseases (IKDs) pose a significant challenge for clinical diagnosis due to substantial phenotypic overlap and heterogeneity, but many patients remain undiagnosed despite growing access to genetic testing. This study assesses whether an expert-curated, IKD-specific database (CURATED) provides a closer phenotypic match to real patients than the Mondo Disease Ontology (MONDO), a broad, general-purpose resource.

Using a cohort of 340 real-world patients with genetically confirmed diagnoses across 8 IKD categories, we compared MONDO- and CURATED-based disease definitions across 3 independent, computational phenotype-matching tools (Phenomizer, Phrank, and Phen2Disease), which match Human Phenotype Ontology (HPO)-annotated patient phenotypes to disease profiles, using similarity scoring, diagnostic ranking, and mixed-effects models accounting for disease imbalance and interannotator variability.

With CURATED definitions, similarity scores rose by 18%-30% and Discounted Cumulative Gain (DCG) improved by 15%-37% relative to MONDO, reflecting a stronger tendency across all 3 tools to place the correct diagnosis at or near the top of the ranked list. These improvements were maintained after mixed-effects modeling.

The quality of the underlying disease database can be as consequential as the choice of diagnostic algorithm. In this cohort, domain-specific curation improved diagnostic prioritization, suggesting that investment in high-quality, nephrology-specific phenotype resources may advance IKD diagnosis as much as further algorithmic refinement.