Plasma proteomic profiles predict individual future health risk.
Level 3 - non-randomized controlled study
Prospective cohort prognostic model development and internal validation
PubMed 38016990 · doi:10.1038/s41467-023-43575-7
What was done
Researchers trained a neural network to create disease- and mortality-specific proteomic risk scores (ProRS) using 1,461 Olink plasma proteins measured in 52,006 UK Biobank participants. The models were evaluated and internally validated across 45 common health conditions (including cancers, psychiatric, neurological, circulatory, and respiratory diseases) and all-cause mortality, and compared against established clinical indicators.
What was found
The proteomic risk scores stratified risk across all 45 conditions and mortality. For 10 endpoints, including cancer, dementia, and death, ProRS achieved C-indexes exceeding 0.80. ProRS performed better than or equivalent to standard clinical indicators for nearly all endpoints. Combining ProRS with clinical predictors yielded only minor predictive improvements compared to ProRS alone. Specific proteins, such as GDF15, showed broad discriminative value across multiple disease categories.
Why it matters
A single plasma proteomics panel could potentially consolidate and replace diverse clinical and laboratory tests to forecast broad multi-disease risk and mortality.
Limits
The study relied entirely on internal validation within the UK Biobank, which features known healthy-volunteer selection biases and limited ancestral diversity. Absolute risk calibration, specific discrimination metrics across all 45 individual conditions, and prospective external validation in diverse real-world clinical cohorts were not reported in the abstract.
Cited by
- supports Data from the UK Biobank with 14 years of follow-up demonstrates that high baseline GDF15 levels are associated with higher rates of mental illness, cardiovascular disease, hypertension, and all-cause mortality.