Aim: Over 50% of people with obesity-associated prediabetes fail to sustain long-term glycemic improvements, even after successful weight-loss maintenance through lifestyle interventions. This study aimed to identify baseline lipidomic signatures predictive of personalized glycemic changes following a 3-year structured lifestyle intervention.
Methods: Serum lipidomics was conducted at baseline and annually over 3 years, in 100 participants from the Australian sub-cohort of the PREVIEW randomized controlled lifestyle trial. Longitudinal lipidome dynamics were characterized using multivariate analyses and Fuzzy C-Means clustering. Lipid-glycemic associations across 4 time points were assessed using Response Screening and Bayesian curve fitting. Baseline lipid predictors were identified using the best-performing model, Boosted Neural Networks, among 6 advanced machine learning algorithms.
Results: Lifestyle interventions induced significant serum lipidome remodeling, marked by reductions in ceramides and lysophospholipids, increases in complex sphingolipids and ethanolamine-based lipids, and altered ratios between biologically related lipid subclasses. Several lipid species, particularly from dihydroceramides and lysophospholipids, exhibited longitudinal associations with glycemic changes over time. Notably, a distinct subset of baseline lipids only bearing saturated C14:0 –C24:0 fatty acyl chains accounted for 27% of the leading predictors of 3-year glycemic changes.
Conclusions: Discovered lipidomic dynamics and lipid-glycemic associations illustrate how lifestyle interventions broadly reshape the metabolic landscape, informing potential targets for long-term glycemic monitoring and regulation. In addition, this study demonstrates the utility of advanced machine learning in identifying robust lipid predictors of personalized glycemic outcomes, facilitating early identification of individuals who are less likely to benefit from lifestyle interventions and thus supporting precision prediabetes care.