DOI: 10.1161/circ.148.suppl_1.18875 ISSN: 0009-7322

Abstract 18875: Using the Proteome to Detect Biomarkers of Cardiovascular Comorbidities

Yolanda Hagar, Jessica Chadwick
  • Physiology (medical)
  • Cardiology and Cardiovascular Medicine

Introduction: Concomitant non-cardiovascular disorders in individuals with cardiovascular disease contribute to morbidity and mortality. However, commonly used statistical approaches for identifying biomarkers of disease typically only examine the relationship between the biomarker and one endpoint at a time, ignoring potential correlation across diseases. Use of more advanced statistical tools that model two endpoints simultaneously can provide additional insight beyond what is provided by commonly used methods.

Hypothesis: Biomarker identification can be made more powerful through the use of Gaussian Process (GP) Bayesian analyses. This method quantifies how one or more biomarkers may predict two biological endpoints (which may or may not be correlated) simultaneously. It is hypothesized that this will allow for the detection of additional biomarkers of comorbidities that cannot be detected using standard univariate approaches.

Methods: Plasma from 5,281 participants with cardiovascular disease and with or without at least one comorbidity (CKD and/or T2D) from the Atherosclerosis Risk in Communities study visit 5 were assayed for 5,284 proteins using the SomaScan® assay. The GP model was applied in a univariate context, quantifying the relationship between each individual protein and cardiovascular disease plus one either comorbidity (CKD or T2D), characterizing the relationship using t-tests and AUC. Results from the univariate Gaussian process models were compared to results from standard univariate logistic regression models.

Results: The GP method identified an additional 283 proteins that were significant for both CVD and CKD simultaneously. These proteins were not identified using logistic regression. Similarly, in addition to the 969 found significant using logistic regression models, the GP method identified an additional 190 proteins that were significant for both CVD and T2D. Among all additional proteins identified, the average AUC increased 25-30%.

Conclusions: Utilizing GP and the proteome, we can discover biomarkers and characterize their signal for comorbid endpoints associated with cardiovascular disease.

More from our Archive