Work

Interpretable AI links routine clinical gout phenotypes to multi-omic signatures and divergent health trajectories

Year: 2026
Type: article
Abstract: Integrative frameworks that connect scalable clinical phenotypes with molecular mechanisms are needed to resolve disease heterogeneity. We developed and externally evaluated an interpretable gout clinical-subset framework using routine data from 21,235 participants across UK Biobank (UKB; n = 10,599), the multicentre Chinese GoutRe registry ( n = 8617), and MIMIC-IV ( n = 2019). Unsupervised learning on six prespecified routine laboratory variables identified three subsets: C1, a low-burden state with preserved reserve; C2, a urate-metabolic inflammatory state with the greatest recurrence liability and a greater cardiometabolic and hepatic comorbidity burden; and C3, a state characterized by systemic vulnerability and higher mortality. In UKB, death-competing recurrence analysis showed a higher recurrence incidence for C2 versus C1 (subdistribution hazard ratio [sHR] 1.896, 95% CI 1.710–2.101), whereas C3 did not differ from C1 (sHR 0.994, 95% CI 0.887–1.112). A six-variable CRT model reconstructed the locked subset labels using nested 10-fold out-of-fold evaluation in UKB and external application without retraining in GoutRe and MIMIC-IV. One-versus-rest AUCs were 0.965/0.945/0.980 in UKB, 0.906/0.914/0.953 in GoutRe, and 0.912/0.898/0.962 in MIMIC-IV for C1/C2/C3, respectively. Genetic, proteomic, and metabolomic analyses supported distinct urate-inflammatory and systemic-vulnerability axes, although the C3 senescence-related genetic signal was nominal and requires replication. Routine laboratory data can identify gout clinical subsets with distinct molecular profiles and patterns of recurrence and mortality; whether these subsets can guide clinical management requires prospective evaluation.
Language: English
Subfield: Nephrology
Field: Medicine
Open Access status: gold
Estimated APC paid: $4,090
OpenAlex ID: W7234550753
FWCI:
Cites: 35
Cited by:
Related to: 10