Personalized concurrent resistance–endurance training in obesity: A bilevel framework for dose and sequence optimization
Intelligence-Based Medicine, cilt.16, 2026 (Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 16
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.ibmed.2026.100508
- Dergi Adı: Intelligence-Based Medicine
- Derginin Tarandığı İndeksler: Scopus, EMBASE, Directory of Open Access Journals
- Anahtar Kelimeler: Bilevel optimization, Concurrent training, In silico modeling, Obesity, Personalized exercise prescription, Quadratic programming
- İstanbul Gelişim Üniversitesi Adresli: Evet
Özet
This study presents a bilevel optimization framework for personalized concurrent resistance–endurance training in adults with obesity. The framework is motivated by a decision mismatch that conventional single-level prescription models may not explicitly preserve: the training designer selects a nominal program, whereas its feasible realization depends on participant-specific adherence, fatigue, recovery, and response dynamics. The National Health and Nutrition Examination Survey (NHANES) 2013–2014 was used only to calibrate a realistic baseline phenotype space for 1963 adults with obesity, and the framework was evaluated in silico on a calibrated synthetic cohort of 3000 participants. The upper level optimizes weekly dose, modality allocation, progression, and sequencing, while the lower level returns the participant-specific optimal realization of that prescription through adherence, fatigue, and response dynamics. The upper-level search couples a warm-started, surrogate-assisted mixed-variable stage with CMA-ES refinement, and each lower-level problem is a convex quadratic program whose curvature and optimistic selection are characterized explicitly. Relative to heuristic progression, population-average, random, fixed equal-dose, and fixed-sequence prescriptions, the framework increased the mean simulated objective by 3.4%, 10.6%, 13.9%, 22.6%, and 28.0%, respectively, and exceeded heuristic progression for every participant, with all strategies remaining feasible. Personalization acted through participant-specific dose bounds and a resistance share that increased with fatigue sensitivity. A benchmark against genetic, tree-structured Parzen, and CMA-ES searches and a KKT reformulation, together with cohort reweighting and latent-trait sensitivity analyses, showed that the ordering of the strategies was stable, whereas the gain over the heuristic required joint refinement of all weekly decisions. These in silico findings depend on the assumed reduced-order dynamics and do not imply clinical validation.