Supplementary file 1_No evidence that serum metabolite slopes improve held-out prediction of clinical progression in the Alzheimer’s disease neuroimaging initiative.zip

Models using 88 phosphatidylcholine-family metabolite slopes or 598 full-panel slopes were exploratory.</p>Results<p>The clinical model achieved an area under the receiver operating characteristic curve (AUROC) of 0.758 (95% CI, 0.675–0.838). Adding PC aa C42:1 changed the AUROC by +0.0004 (−0.013 to 0.017). Both partially penalized lasso models retained no metabolite slopes at the selected penalty.

A post-hoc evaluation restricted to 233 participants with mild cognitive impairment, including 38 events, also found no evidence of incremental discrimination. Other sensitivity analyses yielded similar results.</p>Conclusion<p>We found no evidence that annualized serum metabolite slopes improved held-out prediction in this sample. The 26 development events limited estimation and feature selection, so these results cannot distinguish between a lack of incremental prognostic information and insufficient data to identify such information.

Larger development samples and independent validation are needed to assess the value of serial metabolomics.</p>