Brandl · American journal of physiology. Endocrinology and metabolism 2026 · cross-sectional machine learning prediction study · n=454

Predicting resting metabolic rate in healthy adults: a comparative analysis using the enable cohort.

Cited 1 times in the scientific literature.

Level 4 - case-series / case-control

Cross-sectional predictive modeling and validation study

PubMed 41494660 · doi:10.1152/ajpendo.00375.2025 · record verified 2026-08-29

What was done

Machine learning models (linear and nonlinear) were trained to predict resting metabolic rate (RMR) using data from 454 adult participants in the cross-sectional enable phenotyping platform (Freising and Nuremberg cohorts). Models were tested using either a full set of 94 variables—encompassing anthropometry, clinical blood markers, gut microbiota, fecal short-chain fatty acids (SCFAs), and mean outdoor temperature—or a reduced panel of standard clinical features (sex, age, body weight, fat mass, and fat-free mass). Performance was evaluated by cross-validation and independent test datasets, and feature stability was assessed with repeated cross-validation and marginal variance decomposition.

What was found

Lasso regression achieved the best performance, explaining 76.8% of RMR variance in the Freising cohort. Key predictive features included fat-free mass, body weight, and mean outdoor temperature. Clinical blood parameters contributed marginally, whereas gut microbiota variables and fecal short-chain fatty acids did not contribute to explaining RMR.

Why it matters

Incorporating environmental variables such as mean outdoor temperature alongside clinical measures improves standard anthropometric RMR prediction. Gut microbiota composition and fecal SCFAs appear irrelevant for explaining interindividual variance in resting metabolic rate.

Limits

The abstract reports variance explained (76.8%) but provides no absolute error metrics (e.g., mean absolute error, root-mean-square error) nor specific numerical performance figures for the independent test cohorts. The study population was restricted to healthy adults, limiting applicability to individuals with clinical metabolic disorders.

Cited by