Decade Preceding Diseases

To trace the trajectories of metabolites prior to disease diagnosis, we conducted a nested case-control study. We identified individuals diagnosed with the disease during the follow-up as “cases”. To facilitate meaningful comparisons, each case was paired with five controls, carefully matched based on propensity scores that took into account factors like sex, baseline age, race, BMI, and TDI using the nearest neighbor approach.

We computed the duration preceding clinical diagnosis for every case and assigned a corresponding “proxy” time to the matched controls. We considered multiple time points leading up to the diagnosis (e.g., 15 years, 12 years, 10 years before diagnosis, etc.). Case-control pairs were grouped based on their time to diagnosis, and we fitted a LOESS regression model, based on Euclidean distance, for each metabolite across the years preceding the clinical diagnosis. The analysis were performed through built-in function“loess” in R.