electricsheepafrica/ssa-body-composition-women

SSA Body Composition Dataset (Women, Multi-ancestry, Synthetic)

Dataset summary

This dataset provides a synthetic body composition cohort of 10,000 women across multiple ancestry groups with a focus on sub-Saharan Africa (SSA). It includes:

• Body mass index (BMI) and BMI categories.
• Waist-to-hip ratio (WHR) and high-risk WHR categories.
• Body fat percentage and body fat categories.
• Mammographic breast density (BI-RADS A–D) and dense/non-dense flags.

All individuals are fully synthetic. Distributions are informed by DHS/SSA data, WHO/obesity statistics, NHANES body composition data, and breast density references (BI-RADS) but do not represent real patients.

Cohort design

Sample size and populations

• Total N: 10,000 synthetic women.

• Populations:

  • SSA_West: 2,000
  • SSA_East: 2,000
  • SSA_Central: 1,500
  • SSA_Southern: 1,500
  • AAW (African American women, admixed): 1,500
  • EUR (European reference): 1,000
  • EAS (East Asian reference): 500

• Sex: all Female.

• Age: 18–80 years; age distributions differ modestly by population (e.g. slightly older EUR/AAW on average) but remain realistic for an adult screening/clinical cohort.

These population labels align with other Electric Sheep Africa synthetic datasets to enable cross-dataset analyses (e.g., linking body composition with genomic or pharmacogenomic features).

Body composition features

Body mass index (BMI)

• Continuous bmi in kg/m².
• Categorical bmi_category using standard cut points:
  • underweight – BMI < 18.5.
  • normal – 18.5 ≤ BMI < 25.
  • overweight – 25 ≤ BMI < 30.
  • obese – BMI ≥ 30.

Per-population BMI category prevalences are tuned to reflect:

• Elevated overweight/obesity in SSA women in some regions (DHS/SSA analyses).
• High overweight and obesity prevalence globally (WHO fact sheet, ~43% adults overweight, rising obesity).
• Differences between SSA, AAW, EUR, EAS reference groups.

Waist-to-hip ratio (WHR)

• Continuous whr (waist circumference / hip circumference).
• Categorical whr_category:
  • high – WHR ≥ 0.85 (WHO high-risk cutoff for women).
  • normal – WHR < 0.85.

Target high WHR prevalence by population is specified so that SSA and AAW groups carry a higher central-obesity burden, while EUR/EAS groups maintain substantial but somewhat lower high-WHR fractions, reflecting global cardiovascular risk literature.

Body fat percentage

• Continuous body_fat_percent, modeled as a function of age.
• Categorical body_fat_category:
  • lean
  • average
  • overweight
  • obese

Age bands and mean values draw on NHANES DXA data (Borrud et al.), where mean female body fat percentage increases with age:

• 18–29 years: ~30%.
• 30–39 years: ~32%.
• 40–49 years: ~34%.
• 50–59 years: ~36%.
• 60–80 years: ~38%.

A modest standard deviation introduces inter-individual variability while preserving realistic age trends.

Breast density (BI-RADS)

• breast_density_birads – BI-RADS density category:

  • A – almost entirely fatty.
  • B – scattered fibroglandular densities.
  • C – heterogeneously dense.
  • D – extremely dense.

• breast_density_is_dense – boolean flag for dense breasts (C or D).

The age- and ancestry-specific dense breast fractions are informed by:

• SBI white paper on breast density and supplemental screening (~40% of women over 40 have dense breasts overall).
• BI-RADS atlas definitions and qualitative patterns:
  • Higher density in younger women.
  • Decreasing density with age across all groups.

The configuration sets dense fractions by group and age band (SSA+AAW, EUR, EAS), with SSA/AAW generally showing slightly higher dense fractions at younger ages than EUR, while EAS sits between or near EUR.

File and schema

body_composition_data.parquet / body_composition_data.csv

One row per synthetic individual with:

• Demographics / ancestry

  • sample_id – synthetic ID.
  • population – SSA_West, SSA_East, SSA_Central, SSA_Southern, AAW, EUR, EAS.
  • region – SSA subregion name or Non_SSA.
  • is_SSA – boolean.
  • is_reference_panel – True for AAW/EUR/EAS.
  • sex – Female.
  • age – integer years (18–80).

• BMI

  • bmi – continuous BMI.
  • bmi_category – underweight, normal, overweight, obese.

• WHR

  • whr – continuous waist-to-hip ratio.
  • whr_category – high (≥0.85) or normal.

• Body fat

  • body_fat_percent – continuous body fat percentage.
  • body_fat_category – lean, average, overweight, obese.

• Breast density

  • breast_density_birads – A, B, C, or D.
  • breast_density_is_dense – True for C/D, False for A/B.

Generation

The dataset is generated with:

• body_composition/scripts/generate_body_composition.py

using configuration in:

• body_composition/configs/body_composition_config.yaml

and literature curated in:

• body_composition/docs/LITERATURE_INVENTORY.csv

Key modeling choices:

1. Age sampled from population-specific normal distributions and truncated to 18–80 years.
2. BMI
   • BMI category sampled according to population-specific prevalences.
   • Continuous BMI sampled from category-specific means/SDs, with global min/max bounds.
3. WHR
   • WHR sampled from population-specific normal distributions, with plausible hard limits.
   • high assigned using the WHO cutoff 0.85.
4. Body fat %
   • Age band determined from age, then percentage sampled around age-band means.
   • Category derived from percentage using fixed cutpoints.
5. Breast density
   • Age band and population group (SSA+AAW, EUR, EAS) used to determine dense fraction.
   • Dense vs non-dense assigned probabilistically, followed by specific BI-RADS A–D category.

Validation

Validation follows the GENOMICS Synthetic Data Playbook and is performed by:

• body_composition/scripts/validate_body_composition.py

Key checks:

• C01–C02 – Sample size and population counts
  • Confirms N = 10,000 and population counts match the configuration within ±10% relative deviation.
• C03 – BMI category distributions
  • Compares observed BMI category fractions by population to the configured targets.
• C04 – WHR high-risk prevalence
  • Compares observed high-WHR prevalence (WHR ≥ 0.85) by population to configuration.
• C05 – Body fat mean percent by age band
  • Compares observed age-band means to NHANES-inspired config values.
• C06 – Breast density dense fractions
  • Compares observed dense (C/D) fractions by group and age band to config.
• C07 – Missingness in key variables
  • Ensures negligible missingness in demographics and body composition measures.

The validator outputs:

• body_composition/output/validation_report.md

For the released version, all checks have an overall status of PASS (with at most minor WARN-level deviations acceptable for synthetic data).

Intended use

This dataset is intended for:

• Methods development in epidemiology and risk modeling that require realistic but non-identifiable body composition data.
• Teaching and demonstrations of:
  • BMI/WHR/body fat distributions by age and region.
  • The impact of dense breasts on screening strategies.
• Benchmarking of risk-scoring pipelines that incorporate anthropometric and breast density features.

It is not suitable for:

• Clinical decision-making.
• Estimating real-world prevalence or trends.
• Individual-level inference.

Ethical considerations

• No real patient data are used; all individuals are synthetic.
• Population labels are for simulation and methodologic realism only.
• Users should avoid over-interpreting results as statements about specific countries or real-world groups.

License

• License: CC BY-NC 4.0.
• Free to use for non-commercial research, teaching, and methods development with attribution.

Citation

If you use this dataset, please cite:

Electric Sheep Africa. "SSA Body Composition Dataset (Women, Multi-ancestry, Synthetic)." Hugging Face Datasets.

and, where appropriate, relevant underlying body composition and breast density literature (DHS/SSA analyses, WHO obesity fact sheets, NHANES body composition publications, BI-RADS atlas, and SBI breast density white paper).