B105-17 A Cohort-Based Phenotypic Risk Score Quantifies Phenotypic Similarity Among Pulmonary Patients

Rationale Human disease exists along a continuum shaped by both rare and common genetic variants. Yet quantifying this phenotypic spectrum to identify genetic determinants of shared disease features remains difficult. The Phenotypic Risk Score (PheRS), originally tested in predefined Mendelian disorders, quantifies an individual’s phenotypic similarity to a disease using weighted disease-related Human Phenotype Ontology (HPO) terms. However, it can overlook patients with atypical or overlapping features, who align better with an adaptive reference group, rather than fit a single diagnosis. This is particularly important in pulmonary genetics, where heterogeneous disorders (e.g., bronchiectasis) of related molecular etiologies (i.e., mucociliary-clearance genes) often exhibit a broad phenotypic spectrum. To overcome this, we developed a cohort-based method that weighs each HPO term included in the PheRS by its frequency within affected individuals, using real-world data from patients with diverse pulmonary presentations. Methods We analyzed data from 26,017 individuals in the Children’s Rare Disease Cohorts, a research platform integrating clinical and genomic data, including 118 patients with unexplained bronchiectasis, referred for further genomic evaluation. HPO terms extracted from electronic medical records were used to compute bronchiectasis PheRS using an established diagnosis-based model (Orphanet Idiopathic Bronchiectasis diagnosis; ORPHA:60033) and our new cohort-based models, which included nine alternative weighting strategies. Model performance was assessed by (1) discriminatory ability using Wilcoxon rank-sum testing, and (2) ability to detect undiagnosed bronchiectasis cases. Results Compared to the diagnosis-based model, among alternative cohort-based models, model 2 exhibited the greatest discriminatory power to differentiate diagnosed bronchiectasis patients from others (Figure, Wilcoxon P < 0.00001). To assess the ability to identify undiagnosed bronchiectasis cases, we screened 25,899 patients without a diagnosis of bronchiectasis using PheRS values greater than the bronchiectasis cohort median as a cutoff (54.7 for the cohort-based model; 17.9 for the diagnosis-based model). Whereas the diagnosis-based model identified 88 patients, the cohort-based model 4 identified 239 patients with high PheRS, of whom there was an enrichment of immune-related disorders, suggesting a shared underlying genetic etiology. Notably, a patient with situs inversus, sinusitis, and chronic otitis media (PheRS 79.8), consistent with undiagnosed primary ciliary dyskinesia, was identified by the cohort-based model but not the diagnosis-based method. Conclusion We describe a novel phenotype similarity scoring method with superior performance in quantifying phenotypic similarity across broad pulmonary conditions. This cohort-based PheRS improves detection of clinically relevant pulmonary phenotypes and may accelerate gene discovery and precision evaluation for bronchiectasis and other heterogeneous lung diseases.