INTERPRETABLE AI-BASED RISK PROFILING OF METABOLIC SYNDROME AND SIMULATED KLOTHO MODULATION USING REAL-WORLD NHANES DATA (2007-2023)
Author(s)
Achilles Saxby, MS1, Karthik Chandrakant, BS1, Pramod Koujalagi, BS1, Chandra Ranganathan, MBA1, Dahlia Musa, PhD1, Paola Dama, PhD2, Vinodh Balaraman, MBA1;
1KolateAI PharmaTech Inc, New York, NY, USA, 2NAMina Bio, New York, NY, USA
1KolateAI PharmaTech Inc, New York, NY, USA, 2NAMina Bio, New York, NY, USA
OBJECTIVES: Metabolic Syndrome (MetS) is a heterogeneous condition linked to elevated cardiometabolic risk. Traditional models relying on isolated biomarkers often miss patient‑level complexity relevant to clinical decisions and trial design. This study used interpretable artificial intelligence (AI) to identify key predictors of MetS in real‑world data and to evaluate Klotho, a pleiotropic protein involved in metabolic and aging pathways, as a contextual modifier of metabolic risk.
METHODS: A retrospective analysis was performed using National Health and Nutrition Examination Survey (NHANES) data from 2007-2023. Of 71,775 participants, 4,855 met inclusion criteria (age >60 years, serum Klotho measurements, and complete metabolic biomarker data). An interpretable AI model was developed using KolateAI’s novel precision‑medicine platform to generate individual MetS risk predictions, rank feature importance, identify clinically meaningful thresholds, and simulate counterfactual Klotho‑related perturbations through digital‑twin modeling. Simulations were intended for hypothesis generation and stratification rather than efficacy assessment. Future translational evaluation will use human‑relevant New Approach Methodologies (NAMs).
RESULTS: The model achieved 92% accuracy. Strong positive correlates included sagittal abdominal diameter (SAD), body mass index (BMI), glycohemoglobin (HbA1c), fasting insulin, Apolipoprotein B, and Mefox oxidation product. Negative correlates included sex hormone-binding globulin (SHBG), nervonic acid, and Klotho. Key thresholds included SAD >25 cm (69.3% MetS prevalence), SHBG <30 nmol/L (65.7%), and HbA1c ≥6.5% (79%). Klotho showed an inverse association with MetS and clustered with favorable metabolic profiles, though it was not a dominant predictor. Simulated Klotho modulation enabled stratification into likely responder (~53%) and non‑responder (~47%) groups.
CONCLUSIONS: Interpretable AI enables transparent, patient‑level MetS risk profiling. Klotho acts as a negative correlate and contextual modifier rather than a primary driver. Simulation‑based stratification supports hypothesis generation, patient enrichment, and precision trial design, offering a NAMs‑aligned, human‑relevant pathway for downstream validation.
METHODS: A retrospective analysis was performed using National Health and Nutrition Examination Survey (NHANES) data from 2007-2023. Of 71,775 participants, 4,855 met inclusion criteria (age >60 years, serum Klotho measurements, and complete metabolic biomarker data). An interpretable AI model was developed using KolateAI’s novel precision‑medicine platform to generate individual MetS risk predictions, rank feature importance, identify clinically meaningful thresholds, and simulate counterfactual Klotho‑related perturbations through digital‑twin modeling. Simulations were intended for hypothesis generation and stratification rather than efficacy assessment. Future translational evaluation will use human‑relevant New Approach Methodologies (NAMs).
RESULTS: The model achieved 92% accuracy. Strong positive correlates included sagittal abdominal diameter (SAD), body mass index (BMI), glycohemoglobin (HbA1c), fasting insulin, Apolipoprotein B, and Mefox oxidation product. Negative correlates included sex hormone-binding globulin (SHBG), nervonic acid, and Klotho. Key thresholds included SAD >25 cm (69.3% MetS prevalence), SHBG <30 nmol/L (65.7%), and HbA1c ≥6.5% (79%). Klotho showed an inverse association with MetS and clustered with favorable metabolic profiles, though it was not a dominant predictor. Simulated Klotho modulation enabled stratification into likely responder (~53%) and non‑responder (~47%) groups.
CONCLUSIONS: Interpretable AI enables transparent, patient‑level MetS risk profiling. Klotho acts as a negative correlate and contextual modifier rather than a primary driver. Simulation‑based stratification supports hypothesis generation, patient enrichment, and precision trial design, offering a NAMs‑aligned, human‑relevant pathway for downstream validation.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR203
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Diabetes/Endocrine/Metabolic Disorders (including obesity)