Methodological considerations for identifying multiple plasma proteins associated with all-cause mortality in a...

Area of Science:

• Biochemistry and Molecular Biology
• Genomics and Proteomics
• Epidemiology and Public Health

Background:

• Plasma proteome characterization advances allow large-scale longitudinal studies.
• Identifying robust protein-disease associations remains challenging due to proteome complexity.
• Novel biomarkers are crucial for improving risk stratification for early mortality.

Purpose of the Study:

• To investigate associations between 138 plasma proteins and long-term all-cause mortality risk.
• To assess the reproducibility of protein-mortality associations using a two-step random-split approach.
• To evaluate the predictive value of plasma proteins for mortality risk stratification.

Main Methods:

• Analysis of 138 plasma proteins in 3,918 participants from the Malmö Diet and Cancer Study.
• Utilized a two-step random-split approach for discovery and replication cohort simulation.
• Employed Cox regression, stepwise Cox regression, Lasso-Cox regression, and random survival forest (RSF) for analysis.

Main Results:

• Eight plasma proteins associated with all-cause mortality after adjusting for established risk factors and Bonferroni correction.
• Three proteins (GDF-15, NT-proBNP, ESE4) were consistently identified across four methods in both random samples.
• Inclusion of GDF-15 improved the C-statistic of a risk model from 0.7222 to 0.7284.

Conclusions:

• Several plasma proteins are associated with increased all-cause mortality risk, independent of established factors.
• Growth/differentiation factor-15 (GDF-15) shows potential as a predictive biomarker for mortality.
• Methods for identifying multiple protein biomarkers may be prone to overfitting, limiting reproducibility.