In:
Circulation, Ovid Technologies (Wolters Kluwer Health), Vol. 144, No. Suppl_1 ( 2021-11-16)
Abstract:
Introduction: The critical path for novel treatment and prevention strategies, both pharmacological and non-pharmacological, requires clinical outcomes trials, which are slow, expensive and retard patient access to successful therapies. For marketed drugs, unreliable residual risk stratification hinders precision-directed therapy and facilitates payor reticence. Hypothesis: Large-scale proteomics and machine learning can identify a reliable, individualized, and dynamic surrogate of cardiovascular risk, synoptic for multiple biologic mechanisms, with a resulting multimarker model that is more robust than individual biomarkers. Avoidance of immutable prognostic factors (e.g. age, sex, medical diagnoses) will enhance sensitivity to change in risk. Methods: We quantified ~5000 proteins in 〉 31,000 archived samples from 〉 21,000 higher risk individuals in nine clinical datasets, generating ~150M plasma protein measurements. We derived and validated a 27-protein model encompassing 10 biologic processes to predict the four-year likelihood of myocardial infarction, stroke, hospitalization for heart failure or death. We assessed the association of model-predicted change with observed outcomes in a variety of conditions in four clinical intervention trials. Results: See table for details. In validation, observed event rates were 7.8 fold higher in quintile 5 than quintile 1 for proteins vs. 2.9 for an optimized clinical factor model; the Net Reclassification Index was +0.43. AUCs were 0.73 for proteins vs. 0.63 for clinical, and c-statistics 0.71 vs. 0.62. The protein model also responded concordantly with outcomes in four intervention studies. Conclusions: A 27-protein model is more prognostic than clinical factors in higher risk populations and has directionally concordant responsiveness to all mechanistic challenges to date. These features are consistent with surrogacy, and may prove useful in clinical trials and precision medicine.
Type of Medium:
Online Resource
ISSN:
0009-7322
,
1524-4539
DOI:
10.1161/circ.144.suppl_1.11537
Language:
English
Publisher:
Ovid Technologies (Wolters Kluwer Health)
Publication Date:
2021
detail.hit.zdb_id:
1466401-X
Bookmarklink
