Plasma proteomic signatures preceding the first recorded bronchiectasis diagnosis: a prospective UK biobank study

Abstract

Bronchiectasis is increasingly recognised as a chronic airway disease, but is often diagnosed only after irreversible structural abnormalities have developed. To characterize plasma molecular patterns preceding the first recorded bronchiectasis diagnosis, we analysed baseline plasma proteomics in the UK Biobank Pharma Proteomics Project. Associations between 2,911 Olink Explore proteins and the first recorded inpatient ICD-10 J47 diagnosis after baseline were examined using multivariable Cox models, with temporal stratification using administrative censoring and a 5-year landmark design. An elastic-net Cox model was used to derive a 41-protein panel for subsequent descriptive analyses, while predictive performance was evaluated separately using repeated nested cross-validation with out-of-fold predictions. Among 52,916 participants eligible for prospective follow-up, 772 had a first recorded bronchiectasis diagnosis during follow-up. In the primary complete-case analysis (44,299 participants; 601 events), 891 proteins were associated with the first recorded bronchiectasis diagnosis at a false discovery rate < 0.05, with the strongest signals including WFDC2, LAMP3, MSLN, CD80, IFI30, CEACAM6, PLAUR, SPINT1 and IL2RA. Temporally stratified analyses identified 1,311 proteins associated with events within 5 years and 407 associated with events beyond 5 years, with stronger effects for near-term events. Diagnosis-aligned cross-participant profiling identified two oppositely directed patterns, with greater separation closer to the first recorded diagnosis. In repeated nested cross-validation, at 5 years, the AUC was 0.774 for the clinical model, 0.858 for the proteomic component of the combined model and 0.871 for the combined model. Polygenic risk explained only a limited subset of proteomic variation, whereas modifiable exposures showed structured associations with diagnosis-associated proteomic modules. Baseline plasma proteomic profiles were associated with a subsequently recorded inpatient bronchiectasis diagnosis, and fold-specific clinical-proteomic models showed higher internal discrimination than clinical models alone. Separately, the fixed 41-protein signature should be considered a candidate risk-stratification panel requiring external and prospective validation. These findings may provide insight into biological processes associated with a subsequently recorded diagnosis but do not establish biological disease onset or clinical utility.