What are Active People Made of? Identifying Key Biological Correlates of Physical Activity in Nationally Representative Databases

Authors

Isaac R. Ramos, University of Texas at AustinFollow
Pratik Patil, University of Texas at AustinFollow
Ayland C. Letsinger, University of Texas at AustinFollow

Abstract

Despite widespread awareness of the health benefits of physical activity, more than three-quarters of U.S. adults fail to meet national physical activity guidelines. Biological correlates may contribute to persistent differences in physical activity engagement beyond sociodemographic factors. PURPOSE: To identify biological correlates of physical activity levels in nationally representative U.S. adults. METHODS: Data were collected from the National Health and Nutrition Examination Survey (NHANES) 2021-2023 cycle. Physical activity was assessed using the Global Physical Activity Questionnaire and was expressed as minutes per week of leisure time moderate to vigorous physical activity (MVPA), with vigorous activity weighted by a factor of two. Reported activity frequencies were standardized to a weekly metric, and implausible values (>10,080 min·wk⁻¹) were excluded. Candidate biological, demographic, and socioeconomic predictors were derived from NHANES data modules; variables with >30% missingness or no variability were excluded. Weekly physical activity was log-transformed prior to analysis. Missing predictor values were imputed using a random-forest–based approach (missRanger). Random Forest models were utilized to estimate permutation feature importance for all predictors, with multiple linear regression and Least Absolute Shrinkage and Selection Operator (LASSO) models evaluated for comparative model performance using (out-of-sample) R² and root mean squared error (RMSE). RESULTS: A total of 2,783 adults were included in the final analyses.  Random forest permutation feature importance analyses identified total testosterone (PFI = 0.147), follicle-stimulating hormone (FSH; PFI = 0.05), and estradiol (PFI = 0.048) as the strongest biological correlates of weekly physical activity. The strongest sociodemographic and examination factors were body mass index (BMI; PFI = 0.086) and gender (PFI = 0.077).  Random forest models demonstrated greater explanatory power than multiple linear regression and LASSO models, with higher explained variance (Random Forest R² = 0.107, multiple linear regression R² = -0.067, LASSO R² = 0.031) and lower prediction error (Random Forest RMSE = 1.464, multiple linear regression RMSE = 1.6, LASSO RMSE = 1.525). CONCLUSION: Sex hormone biomarkers (testosterone, FSH, estradiol) ranked among the most important predictors of leisure time MVPA and, in this dataset, testosterone showed greater relative importance than BMI and gender. Given the cross-sectional design, these findings are associational and experimental studies are warranted to determine the directionality and causal nature of these relationships. Nevertheless, these findings parallel extensive animal data and emerging human intervention studies linking sex hormones to physical activity regulation.

Recommended Citation

Ramos, Isaac R.; Patil, Pratik; and Letsinger, Ayland C. (2026) "What are Active People Made of? Identifying Key Biological Correlates of Physical Activity in Nationally Representative Databases," International Journal of Exercise Science: Conference Proceedings: Vol. 2: Iss. 18, Article 108.
Available at: https://digitalcommons.wku.edu/ijesab/vol2/iss18/108