Presentation (CTI)

OBJECTIVES: While International Classification of Diseases (ICD) diagnostic codes facilitate universal classification of diagnosis, symptoms, and procedures, they lack precision for conditions primarily identified through clinical presentation such as non-cystic fibrosis bronchiectasis (NCFB). This abstract demonstrates a data-driven approach using clinical keyword mining within real-world data (RWD) asset library to to bypass limitations in ICD diagnostic codes and isolate clinically recognised NCFB patient populations in China.
METHODS: This study leveraged the integrated JKT RWD asset library combining the Hospital Potential Insight (HPI) database (n=17,800 Tier 2-3 hospitals) and an Real-World Evidence (RWE) clinical database (n=2,200+ Tier 3 hospitals) in China. A four-step clinical assessment was conducted: (1) mapping hospitals treating bronchiectasis through HPI; (2) clustering analysis to select a representative sample for medical record abstraction through RWE; (3) keyword-driven phenotyping to identify adult patients based on specific clinical presentations and high-resolution CT (HRCT) confirmation; and (4) longitudinal tracking of symptoms and infection to assess severity. Findings were extrapolated across 50-100 top-tier hospitals using predictive neural network models.
RESULTS: The methodology established a systematic attrition flow that addresses the imprecision of ICD coding. Keyword-based analysis demonstrated that the initial ICD-identified patient pool was overinclusive, with only a subset of these patients meeting the gold-standard criteria for adult HRCT-confirmed NCFB. Furthermore, the hierarchical flow further differentiated disease severity through identification of distinct sub-phenotypes characterized by frequent exacerbations and positive bacterial (e.g., Pseudomonas aeruginosa) colonization. This approach demonstrated the asset library’s capability to resolve mis-categorisation of patient in standard electronic health records.
CONCLUSIONS: Reliance of ICD codes may misrepresent the true burden of conditions lacking formal ICD diagnostic codes, such as NCFB. Applying keyword-driven phenotyping methodology within integrated RWD assets enables accurate identification of target populations. This methodology provides a scalable framework to support robust outcomes research and evaluation of disease burden of similar therapeutic areas.