Machine Learning on Prediagnostic Metabolite Data Identifies Etiologic Endotypes of Exfoliation Glaucoma in United States Health Professionals.

Purpose: Exfoliation glaucoma (XFG) etiology is poorly understood. Metabolomics-based etiologic endotypes of XFG may provide novel etiologic insights. We aimed to use unsupervised machine learning on prediagnostic plasma metabolites to characterize etiologic XFG endotypes. Design: Prospective case-only analysis. Participants: Among Nurses' Health Study and Health Professionals Follow-up Study participants, 205 (174 female and 31 male) incident XFG cases diagnosed with an average of 11.8 years following blood collection (1989-1995) were included. Methods: We identified and confirmed incident cases of XFG or XFG suspect (collectively called "XFG" henceforth) through 2016 with medical record review. Liquid chromatography-mass spectrometry was used to profile 341 plasma metabolites. After preprocessing prediagnostic metabolites with adjustment for season, time of blood draw, and fasting status, we computed a distance matrix using Pearson distance and computed gap statistics to identify distinct endotypes. Main Outcome Measures: Metabolomics-based XFG etiologic endotypes. Metabolomic profiles were compared across endotypes; false discovery rate (FDR) was used to account for multiple comparisons in Metabolite Set Enrichment Analyses. Exfoliation glaucoma environmental risk factors (e.g., lifetime ultraviolet (UV) exposure, folate consumption), a genetic risk score incorporating 8 major single nucleotide polymorphisms for exfoliation syndrome, and clinical presentations were compared across endotypes. Results: We identified 3 distinct XFG metabolomic endotypes. Compared with the most common endotype 2 (reference group [n = 90; 43.9%]), endotype 1 (n = 56; 27.3%) tended to include more male southern US residents with greater UV exposure and were the least likely to have cardiovascular disease; among women, a higher percentage were postmenopausal. Endotype 3 (n = 59; 28.8%) was associated with being a male northern US resident; a higher prevalence of cardiovascular disease and risk factors such as higher body mass index, diabetes, hypertension, and dyslipidemia; and the lowest genetic susceptibility score. There were no differences in ophthalmic characteristics (e.g., maximum intraocular pressure, bilaterality, age at diagnosis) across endotypes ( Conclusions: Integrated metabolomic profiling can identify distinct XFG etiologic endotypes, suggesting different pathobiological mechanisms. Financial Disclosures: Proprietary or commercial disclosure may be found in the Footnotes and Disclosures at the end of this article.