Estimating Visceral Fat with Petal

Statistical Methods and Reference Data for a DEXA-Based Estimation Model

1. Overview

VAT is the fat depot that surrounds our internal organs; it is distinct from subcutaneous fat. Elevated VAT compromises metabolic performance and is often associated with diabetes and cardiovascular disease. This whitepaper describes the predictive statistical model used by Petal in estimating visceral fat or visceral adipose tissue (VAT) based on midsection fat (or trunk fat mass), chronological age, body weight, and height. This model returns a predicted VAT alongside age-matched population range.

2. Data & Statistical Methods

2.1 Survey Overview

The foundational model was constructed on data in the National Health and Nutrition Examination Survey (NHANES), conducted by the National Center for Health Statistics (NCHS), Centers for Disease Control and Prevention (CDC). NHANES uses a complex, multistage probability sampling design to produce nationally representative estimates for the US civilian non-institutionalised population. DXA scan data from this survey was pooled to create an analytic subset with 5,610 women aged 20 - 59 and with a mean VAT of 478 g (SD 269 g).

2.2 Model Details

The model is a piecewise ordinary least squares (OLS) regression. Knots for the piecewise OLS were selected through an exhaustive grid search run across plausible ranges with 10-fold cross-validated root mean squared error (RMSE).

3. Model Performance

Fitted on n = 5,610 women pooled across four NHANES cycles, the model achieves an in-sample R² of 0.7227 (72.3% of variance in VAT mass explained), an in-sample RMSE of 147.4 g, and a 10-fold cross-validated RMSE of 147.7 g (SD = 2.7 g across folds). The near-identical in-sample and cross-validated RMSE indicates minimal overfitting. For age > 59, a simple linear extrapolation is utilized. In general, prediction error grows with age, reflecting increasing biological variability during the perimenopausal decade. 

Age-specific NHANES percentiles provide the reference against which each predicted VAT value is compared. Percentiles were computed using a five-year rolling window centred on each single year of age — for example, the norm at age 40 draws on all women aged 38–42 in the pooled dataset. This avoids the artificial discontinuities of decade-based groupings. At each age, the mean and standard deviation of observed VAT were computed. Percentile rank was then estimated by treating the age-specific distribution as approximately normal.

3.1 VAT Categories

The model also categorizes VAT as normal or elevated. A conservative approach is utilized to reflect the worst plausible outcome within model uncertainty:

  • Healthy range: the upper bound of the predicted VAT range (prediction + 1 RMSE) falls below the 75th NHANES percentile for the user's age.
  • Elevated range: the upper bound reaches or exceeds the 75th NHANES percentile for the user's age.

The 75th percentile threshold is consistent with the epidemiological literature associating VAT above this level with elevated cardiometabolic risk, including insulin resistance, dyslipidaemia, hypertension, and cardiovascular disease (Despres and Lemieux 2006; Lemieux et al. 2000).

4. Clinical Validation

A two-part clinical validation was performed with VAT index from a commercial body composition scale and a pilot study where volunteers obtained DEXA scans.

4.1 Comparisons with body composition scale measurements

VAT indices were measured in 51 unique volunteers across 119 sessions by Tipre Body Composition scale. The participants ranged in age from 21 to 71 years (mean 38.1, median 37) with heights spanning 144.78 to 177.8 cm and weights from 89.8 to 221.7 lb. Key body composition metrics include body fat percentage (mean 30.5%, range 15.6–65.6%), visceral fat index (mean 7.07, range 2–18), trunk fat mass (mean 20.43 lb, range 6.4–45 lb). The figure below compares the VAT index from the scale with the VAT categorization from the model described here.

4.2 Comparisons with DEXA Scans

DEXA scans were obtained from 10 volunteers from accredited labs, The participants ranged in age from 30 to 65 years, heights spanning 144.78 to 177.8 cm and weights from 89.8 to 221.7 lb. The figure below compares VAT measurements from Petal with the DEXA based measurements.

5. Limitations

  • Women only. The model was developed exclusively on female NHANES participants and should not be applied to men.
  • Ages 20–59 only. The NHANES android/gynoid DXA protocol covers ages 8–59. Above 59, a simple linear extrapolation is utilized without specific uncertainty adjustment.
  • US population norms. NHANES represents the US civilian non-institutionalised population. Percentile norms may not directly apply to populations with substantially different body composition distributions.
  • DEXA required for trunk fat. The primary predictor, trunk fat mass, is a DEXA measurement and is not obtainable from routine clinical anthropometry alone.
  • Prediction interval, not measurement. The ±1 RMSE range covers approximately 68% of the prediction distribution under normality, leaving a one-in-three probability that the true VAT value falls outside the displayed range.
  • No longitudinal validation. The model was developed and validated cross-sectionally. Its accuracy for tracking within-individual VAT change over time has not been assessed.

6. References and Data Citations

6.1 NHANES Survey Data

National Center for Health Statistics. National Health and Nutrition Examination Survey Data. Hyattsville, MD: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, 2011–2018. Available at: https://wwwn.cdc.gov/nchs/nhanes/

6.2 VAT Risk Threshold Context

Despres JP, Lemieux I, Bergeron J, et al. Abdominal obesity and the metabolic syndrome: contribution to global cardiometabolic risk. Arteriosclerosis, Thrombosis, and Vascular Biology. 2008;28(6):1039–1049.

Lemieux I, Pascot A, Couillard C, et al. Hypertriglyceridemic waist: a marker of the atherogenic metabolic triad in men? Circulation. 2000;102(2):179–184.

6.3 Statistical Methods

Abramowitz M, Stegun IA. Handbook of Mathematical Functions. Washington, DC: National Bureau of Standards; 1964. Formula 26.2.17 (normal CDF polynomial approximation).

This document is for informational and educational purposes only. The model and its outputs are not a clinical diagnostic tool and should not be used for medical decision-making without professional interpretation.