electricsheepafrica/ssa-hormonal-profiles

SSA Hormonal Profiles Dataset (Multi-ancestry, Synthetic)

Dataset summary

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

• Estradiol (E2) and progesterone (P4) in women and men (by menopausal status for women).
• Testosterone (T) in women and men (by age band).
• Fasting insulin levels (by population).
• IGF-1 levels (by age band).

Values are informed by published reference intervals and age/sex trends for these hormones, but all individuals and measurements are fully synthetic.

Cohort design

Sample size and populations

• Total N: 10,000 synthetic adults.

• Populations:

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

• Sex distribution:

  • Male: ~45%
  • Female: ~55%

• Age: 18–80 years with population-specific means/SDs, loosely aligned with prior cardiovascular and reproductive-history datasets.

Hormonal variables

Menopausal status (women)

Variable:

• menopausal_status – for women:
  • premenopausal
  • perimenopausal
  • postmenopausal
• For men: menopausal_status = "NA".

Status is assigned by age band (18–39, 40–49, 50–80) with probabilities reflecting typical menopause timing:

• 18–39: mostly premenopausal.
• 40–49: mix of pre-, peri-, and postmenopausal.
• 50–80: predominantly postmenopausal.

Estradiol (E2)

Variable:

• estradiol_pg_ml – estradiol in pg/mL.

Design anchors:

• Premenopausal women have higher and more variable estradiol (mean ~80 pg/mL, SD ~40), capturing the broad range across the menstrual cycle.
• Perimenopausal women have intermediate values (mean ~50 pg/mL).
• Postmenopausal women have low values (mean ~15 pg/mL) with an approximate upper bound of 20–30 pg/mL.
• Men have low-to-moderate estradiol (mean ~35 pg/mL, SD ~15).

These values are informed qualitatively by clinical reference intervals and reviews of estradiol levels in pre- vs postmenopausal women and men.

Progesterone (P4)

Variable:

• progesterone_ng_ml – progesterone in ng/mL.

Design anchors:

• Premenopausal women: mean ~3 ng/mL (large SD), representing random-cycle values with luteal peaks.
• Perimenopausal women: mean ~1.5 ng/mL.
• Postmenopausal women: mean ~0.5 ng/mL.
• Men: mean ~0.5 ng/mL.

These reflect the general pattern of high-cycle variation in premenopausal women and low steady values after menopause or in men.

Testosterone (T)

Variable:

• testosterone_ng_dl – total testosterone in ng/dL.

Age bands:

• 18–29, 30–39, 40–49, 50–59, 60–80 years.

Design anchors (mid-range values from reference summaries):

• Men

  • 18–29: mean ~600 ng/dL.
  • 30–39: ~550 ng/dL.
  • 40–49: ~500 ng/dL.
  • 50–59: ~450 ng/dL.
  • 60–80: ~400 ng/dL.

• Women

  • 18–29: mean ~40 ng/dL.
  • 30–39: ~35 ng/dL.
  • 40–49: ~30 ng/dL.
  • 50–59: ~25 ng/dL.
  • 60–80: ~20 ng/dL.

Testosterone decreases with age in both sexes, roughly consistent with published age-specific reference data.

Fasting insulin

Variable:

• insulin_fasting_uIU_ml – fasting serum insulin in uIU/mL.

Design anchors:

• Healthy adult reference intervals suggest 2–12 uIU/mL as typical.
• Population-specific means reflect metabolic differences:
  • SSA and AAW have slightly higher means (~8.5–10 uIU/mL).
  • EUR/EAS have somewhat lower means (~7–7.5 uIU/mL).

Values are truncated between 1 and 40 uIU/mL.

IGF-1

Variable:

• igf1_ng_ml – insulin-like growth factor 1 in ng/mL.

Age bands and means reflect age-related decline (≈15% per decade) from young adulthood:

• 18–29: mean ~250 ng/mL.
• 30–39: ~210 ng/mL.
• 40–49: ~180 ng/mL.
• 50–59: ~150 ng/mL.
• 60–80: ~130 ng/mL.

Values are truncated between 40 and 500 ng/mL.

File and schema

hormonal_profiles_data.parquet / hormonal_profiles_data.csv

One row per synthetic individual with:

• sample_id
• population, region, is_SSA, is_reference_panel
• sex (Male, Female)
• age (18–80 years)
• menopausal_status (for women; NA for men)
• estradiol_pg_ml
• progesterone_ng_ml
• testosterone_ng_dl
• insulin_fasting_uIU_ml
• igf1_ng_ml

Generation

The dataset is generated with:

• hormonal_profiles/scripts/generate_hormonal_profiles.py

using configuration in:

• hormonal_profiles/configs/hormonal_profiles_config.yaml

and literature curated in:

• hormonal_profiles/docs/LITERATURE_INVENTORY.csv

Key modeling steps:

1. Sample generation – multi-ancestry sample with age/sex distribution.
2. Menopausal status assignment – by age band in women.
3. Hormone sampling – for each hormone, sample from normal distributions with age/sex/status-specific means and SDs and truncate to plausible ranges.

Validation

Validation is implemented in:

• hormonal_profiles/scripts/validate_hormonal_profiles.py

Checks include:

• C01–C02 – Sample size and population counts vs config.
• C03 – Estradiol means by sex and menopausal status vs config.
• C04 – Progesterone means by sex and menopausal status vs config.
• C05 – Testosterone means by sex and age band vs config.
• C06 – Insulin means by population vs config.
• C07 – IGF-1 means by age band vs config.
• C08 – Missingness in key variables.

The validator writes:

• hormonal_profiles/output/validation_report.md

For the released version, the validator reports overall status PASS (with some checks at WARN level where sampling noise produces minor deviations from target means).

Intended use

This dataset is intended for:

• Methods development in endocrine modeling and multi-modal risk prediction.
• Exploring how age, sex, and menopausal status relate to key hormonal axes.
• Serving as a companion to the reproductive history, body composition, and cardiovascular metrics datasets.

It is not suitable for:

• Clinical decision-making or hormone therapy dosing.
• Deriving exact clinical reference intervals.
• Individual-level risk prediction.

All hormone levels are synthetic.

Ethical considerations

• No real patient data are used.
• Population labels are for simulation and methodological realism only.
• Analyses should be interpreted as methodological demonstrations rather than statements about specific groups.

License

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

Citation

If you use this dataset, please cite:

Electric Sheep Africa. "SSA Hormonal Profiles Dataset (Multi-ancestry, Synthetic)." Hugging Face Datasets.

and consider citing relevant endocrine reference works that informed the design (estradiol/progesterone reference intervals, age-related testosterone and IGF-1 studies, and fasting insulin reference interval papers).