Serum metabolomic signal in dry eye disease in a family aware twin cohort

Abstract Dry eye disease is common and heterogeneous, but whether retrospective blood metabolite profiles capture dry-eye signal remains unclear. We performed a retrospective machine-learning analysis of untargeted serum metabolomics in TwinsUK, using approximately 4,200 serum samples from approximately 1,400 participants and two questionnaire-derived outcomes: a broad Women’s Health Study dry-eye definition and a more specific highly symptomatic phenotype. Serum samples were collected from 1996 to 2014, whereas dry-eye outcomes were defined from 2013 questionnaires; the median sample-to-outcome gap in the main repeated-row analysis was 7.0 years. In family-aware analyses, metabolomics added statistically detectable but limited discrimination beyond covariates, increasing receiver operating characteristic area under the curve from 0.43 to 0.57 for the broader outcome and from 0.42 to 0.60 for the highly symptomatic phenotype. Preprocessing strongly influenced discrimination. After false-discovery correction, 13 metabolites were associated with highly symptomatic dry eye, and none with the broader outcome; five remained significant in low-value imputation sensitivity analyses. These findings suggest modest dry-eye-related serum metabolomic signal, shaped by preprocessing and most evident for the highly symptomatic phenotype.