Circulating lipid signature can predict patient survival. To predict patient outcomes, a machine learning model was developed for lipidomics data, with and without the addition of clinical parameters. After training the algorithm on 80% of the PMH data set, a holdout testing subset (20%) of the PMH data set was used to evaluate the model performance. (A) Lipids with clinical parameters compared similarly to lipids alone, and lipids vastly outperform models developed for metabolites or clinical parameters alone. (B) The testing PMH data subset was also used to evaluate patient survival based on the lipid signature derived from the machine learning model. Overall patient survival, based on the abundance of the top lipid in the lipid signature alone; SM (d44:1) outcomes in the entire PMH data set (C) and the entire UC data set (D).