Systematic identification of rare disease patients in electronic health records enables evaluation of clinical outcomes.

Identifying rare disease (RD) patients in electronic health records (EHRs) is difficult, as most of the over 10,000 RDs are not adequately captured by standard coding systems. To address this, we developed a semi-automated workflow to map RDs to SNOMED-CT and ICD-10 codes, enabling improved RD identification across EHR systems. The optimized workflow yielded 88.4% true RD codes in a subset of 1,715 manually curated diseases. Using this workflow and starting with 12,003 GARD IDs mapped to ORPHANET, we obtained 12,081 SNOMED-CT and 357 ICD-10 codes representing 6,342 RDs, organized into 30 ORPHANET linearization classes. We applied these codes to the National COVID Cohort Collaborative (N3C) dataset of over 21 million patients. Among these patients, 8.46 million were identified as COVID-19 positive, of which 4.8 million were used in analyses. Among these, 316,836 (6.55%) had a preexisting RD. Logistic regression, adjusted for age and BMI, revealed that most RD classes were significantly associated with increased odds of severe COVID-19 outcomes. Notably high odds of mortality were observed for rare cardiac (OR = 4.07) and otorhinolaryngologic diseases (OR = 4.00). Hospitalization risk was also elevated across all RD classes, with the highest odds seen in otorhinolaryngologic (OR = 4.31) and endocrine diseases (OR = 3.38). This approach enables scalable RD patient identification in EHRs and highlights the need for tailored healthcare strategies to improve outcomes in RD populations.