Multi-modal human digital twin for hydration

Hydration plays a crucial role in maintaining physical and mental wellbeing, yet individuals differ substantially in their daily hydration needs due to variations in diet, physiology, activity, and context. This pilot study explores the feasibility of developing a multi‑modal human digital twin for hydration, integrating continuous real‑world data from urine specific gravity (USG), nutritional intake, wearable‑based physiological signals, and ecological momentary assessments. Three participants were monitored across 20 non‑consecutive days, collecting USG at every toilet visit, logging food and drink consumption, and wearing Garmin VivoSmart5 and Chill+ sensors to capture heart rate, movement, and galvanic skin response. USG proved to be a sensitive hydration marker, revealing consistent diurnal patterns and partial alignment with subjective hydration reports, though substantial variability highlighted the limitations of self‑reported hydration. Multimodal analyses indicated a complex interplay among hydration, diet, physiological signals, and contextual factors, including strong effects of specific beverages such as Aquarius and coffee. Personalized machine‑learning models trained per participant showed reasonable predictive power and outperformed pooled models, underscoring the individualized nature of hydration dynamics. Preliminary in‑silico simulations demonstrated the potential of the digital‑twin to give specific drink recommendations. On the other hand, these also revealed limitations of window‑based modeling that cannot capture cumulative or long-term physiological dynamics, suggesting that the future hybrid approach that embeds mechanistic physiology alongside machine learning is necessary for realistic and actionable hydration digital twins. These findings support the feasibility and promise of a personalized hydration digital twin and lay the groundwork for future expansion toward more comprehensive and physiologically grounded models.