Abstract
Aim: The aim of this study was to evaluate if the pre-operative risk stratification of post-operative atrial fibrillation can be improved by combining multiple omics layers. Methods: Patients with cardiac surgery who had not had a previous diagnosis of Atrial Fibrillation (AF) underwent pre-operative multi-omics profiling by four modalities, clinical, genomic, plasma proteomic and metabolomic. Cross validation was used to develop the per-layer risk models and the stacking integration model was used to combine the models. Each single-omics model and the clinical score were compared to the integrated model using the area under receiver operating characteristic curve, and the common directions of variation extracted through multi-omics factor analysis. Results: In total, 760 patients had post-operative AF in 35%. It was discriminated by the integrated multi-omics model with area under the curve of 0.82 with highly significant differences compared with best single-omics model and clinical score (p<0.001). This shared latent axis, which mainly consisted of inflammatory proteins and metabolic markers, discerned the patients developing atrial fibrillation, factor analysis found. Protein biomarkers and metabolite biomarkers were the top features of the signature. Conclusion: The combined clinical, genome, proteome and metabolome data performed better than any single layer and better than the clinical score alone in predicting postoperative atrial fibrillation, and uncovered an inflammatory and metabolic signature by the omics, demonstrating the potential of multi-omic integration in improving the clinical risk stratification of perioperatively.
Keywords: Postoperative atrial fibrillation, Cardiac surgery, Multi-omics, Proteomics, Metabolomics, Risk prediction, Data integration