Evaluation of acute FGF2 and PRDX3 for long COVID prediction across cohorts and proteomic platforms
Abstract
Blood proteins measured during acute COVID-19 may help predict Long COVID, but models may fail when moved between cohorts and proteomic platforms. We fitted a reference model using age and sex (C0) in Su/INCOV and a protein model that also included FGF2 and PRDX3 (C2). The pair was fixed before outcome access because both proteins had one-to-one assay mappings. Protein values were standardized within each cohort, and Su coefficients were applied to Zurich without outcome-based refitting. Lower Brier scores indicate better prediction. Su contained 204 participants with complete predictors, but only 125 had linked follow-up. The strict analysis included 122 participants (75 events). Its optimism-corrected C2-minus-C0 Brier difference was + 0.00342 (95% interval, − 0.00608 to + 0.02834). Zurich included 113 participants (40 events). The transferred model difference was + 0.01093 (− 0.01721 to + 0.03882), and strict-model transfer gave + 0.00902 (− 0.01809 to + 0.03696). Both models overpredicted Zurich risk. Reviewer-requested intercept updating also yielded an inconclusive difference. Thus, FGF2 and PRDX3 did not reliably improve prediction beyond age and sex. Because protein standardization depends on a target cohort, this pipeline is not ready to predict risk for a patient.
Related
- Self-supervised plasma proteomic representations for prospective disease prediction across varying protein availability
- Generalizability of proteomic risk prediction across biobanks reveals dependence on phenotype definitions
- Plasma Proteomic Profiles of Alzheimer's Disease and Neurodegeneration in African Cohorts
- Socioeconomic position and functional dependency trajectories across six longitudinal ageing cohorts
- Timing uncertainty in first reported chronic disease across international aging cohorts
BibTeX
@article{duan2026evaluation,
title = {{Evaluation of acute FGF2 and PRDX3 for long COVID prediction across cohorts and proteomic platforms}},
author = {Chuan-Xin Duan and Lang Yang},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-75618-6},
url = {https://www.nature.com/articles/s41598-026-75618-6},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Evaluation of acute FGF2 and PRDX3 for long COVID prediction across cohorts and proteomic platforms
AU - Chuan-Xin Duan
AU - Lang Yang
PY - 2026
DA - 2026/10/10
JO - Scientific Reports
DO - 10.1038/s41598-026-75618-6
UR - https://www.nature.com/articles/s41598-026-75618-6
AB - Blood proteins measured during acute COVID-19 may help predict Long COVID, but models may fail when moved between cohorts and proteomic platforms. We fitted a reference model using age and sex (C0) in Su/INCOV and a protein model that also included FGF2 and PRDX3 (C2). The pair was fixed before outcome access because both proteins had one-to-one assay mappings. Protein values were standardized within each cohort, and Su coefficients were applied to Zurich without outcome-based refitting. Lower Brier scores indicate better prediction. Su contained 204 participants with complete predictors, but only 125 had linked follow-up. The strict analysis included 122 participants (75 events). Its optimism-corrected C2-minus-C0 Brier difference was + 0.00342 (95% interval, − 0.00608 to + 0.02834). Zurich included 113 participants (40 events). The transferred model difference was + 0.01093 (− 0.01721 to + 0.03882), and strict-model transfer gave + 0.00902 (− 0.01809 to + 0.03696). Both models overpredicted Zurich risk. Reviewer-requested intercept updating also yielded an inconclusive difference. Thus, FGF2 and PRDX3 did not reliably improve prediction beyond age and sex. Because protein standardization depends on a target cohort, this pipeline is not ready to predict risk for a patient.
SN - 2045-2322
ER -