Global Geographic, Seasonal, Environmental and Demographic Drivers of Variations in Longitudinal Cardiac Physiology in 5 million Fitbit Users

Resting Heart Rate (RHR) and Heart Rate Variability (HRV) are key biomarkers of cardiovascular health and fitness, but they are typically measured only opportunistically during clinical visits or studies. These sporadic observations make it challenging to identify population-level trends or understand how geographic, environmental and seasonal factors influence cardiac physiology globally. Past research has been limited to small cohorts in selected countries, focusing primarily on RHR while largely overlooking HRV. The widespread adoption of consumer wearables, capable of continuous heart-rate monitoring, makes it possible to finally model and understand these variations on a global scale, but requires a unique access to large volume longitudinal data. Here, we present the first worldwide models describing the influence of objective, external drivers on cardiac physiology variation, using longitudinal data from 4.9 (RHR) and 2.1 (HRV) million consenting Google Health (Fitbit) users collected across 217 countries globally throughout 2022, representing an unprecedented 60 million person-months dataset. We employ Linear Mixed-Effects models to quantify seasonal (month, day length), geographical (longitude, latitude), and environmental (mean, maximum, and minimum temperature) effects on RHR and HRV while adjusting for demographics (age, sex). We use a novel, automated model search framework across five stages that systematically ablates included drivers, driver encodings, and interaction terms across 20 candidate models per metric. Retaining the best specifications at each stage ensures optimal model parsimony and explanatory power for both RHR and HRV individually. Our results reveal stable external drivers of RHR variations, and for the first time, of HRV variations as well. For both measures, the optimal model included age, sex, month, longitude, latitude, and minimum and maximum temperature, explaining 5.3\% of the population-level variance in RHR and 21.0\% in HRV. The magnitude of these fixed effects aligns with previous large-scale physiological studies on sleep, confirming that while these environmental signals are subtle, they represent meaningful and robust population-level drivers. By isolating these drivers, this study provides the first global, monthly-level models for cardiac physiology. Crucially, our modelling framework enables the estimation of physiological variation even in data-sparse regions. Additionally, it provides valuable context for personalised, AI-driven health recommendations in future health assistants, and for normalising disparate national health survey data to a common reference