Abstract Background Metabolic dysregulation is hypothesized to contribute to the pathogenesis of cardiovascular disease in patients with psoriatic disease (PsO). Whether multi-biomarker panels can improve cardiovascular risk prediction in PsO is currently unknown. Purpose To develop and validate a metabolite-based AI score for cardiac risk prediction in patients with PsO. Methods We performed 1H-NMR metabolomics on plasma samples from 1,086 Mass General Brigham Biobank participants with a history of PsO and no known history of cardiovascular disease. The metabolomics platform (Nightingale Health) provided absolute quantification of 168 metabolites. In a random 50% training subset of the MGB Biobank, extreme gradient boosting was applied to construct an AI-driven risk score (MetaPsoAI) to predict the primary outcome of acute myocardial infarction (AMI) or all-cause death. MetaPsoAI was then validated in the 50% test subset of MGB Biobank participants not included in training and externally in 8,065 UK Biobank participants with PsO, using algorithmically defined outcomes. Models for the development and validation of MetaPsoAI were adjusted for age, gender, hypertension, prior stroke, prior MI, BMI, and lipid-lowering medications. Results This cohort’s mean age was 62±15 years and 57% were women. A total of 77 (7.1%) primary outcomes occurred over a median 7 years of follow-up (10 events per 1,000 person-years) in the MGB Biobank. Sixteen metabolites were significantly associated with the primary outcome of AMI or death (a). The 10 metabolites with the greatest contributions to the MetaPsoAI score are shown in the panel (b). Each 1-unit increase in MetaPsoAI was associated with a 1.28-times higher risk of death or AMI (Adj. HR: 1.28 1.01, 1.64). For an optimal log-rank cut-off of 4.69 (c), patients with high MetaPsoAI had a 3.2-fold higher adjusted risk for the primary outcome in the test subset (d). Applying MetaPsoAI in the full UK Biobank dataset revealed significantly higher scores in PsO patients compared to non-PsO individuals (e). Amongst 8,065 patients with PsO in the UK Biobank (1,192 15% primary outcomes), each 1-unit higher MetaPsoAI score was associated with a 1.10-times higher risk of death or AMI (Adj. OR: 1.10 1.05, 1.15, p0.0001) and a high MetaPsoAI score was associated with a 1.20-times higher risk of death or AMI (Adj. OR: 1.20 1.03, 1.38, p=0.01, f). Conclusions We introduce MetaPsoAI, an AI-driven risk score that integrates metabolomic data for long-term cardiac risk prediction in patients with PsO. In internal validation and external replication in the UK Biobank, MetaPsoAI demonstrated robust prognostic value for all-cause mortality and AMI. These findings highlight the potential of metabolomics-based AI models for personalized cardiovascular risk stratification in psoriatic disease.
Kotanidis et al. (Sat,) studied this question.