electricsheepafrica/ssa-breast-pathology-histology

SSA Breast Pathology & Histology Dataset (Women, Multi-ancestry, Synthetic)

Dataset summary

This dataset provides a synthetic cohort of invasive breast cancers in women across multiple ancestry groups, with emphasis on sub-Saharan Africa (SSA) and comparable reference populations.

Pathology features include:

• Tumor grade (Nottingham-like grades G1–G3).
• Histologic type (NST/ductal, lobular, medullary-like, mucinous, other special types).
• Lymphovascular invasion (LVI).
• Necrosis patterns (none, focal, extensive).
• Tumor-infiltrating lymphocytes (TILs) category (low, intermediate, high).

Distributions are anchored qualitatively to SEER-based histology series, Nottingham grade cohorts, LVI reviews, necrosis descriptions, and TNBC TIL literature, but all tumors are fully synthetic.

Cohort design

Sample size and populations

• Total N: 10,000 synthetic invasive breast cancers.

• Populations:

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

• Sex:

  • Predominantly Female, with a small fraction of male breast cancers (~1%).

• Age:

  • 18–90 years.
  • Older mean age at diagnosis in reference populations (EUR/EAS/AAW) relative to SSA clusters.

Pathology variables

Tumor grade (Nottingham-like)

Variable:

• tumor_grade – one of:
  • G1 – low grade.
  • G2 – intermediate grade.
  • G3 – high grade.

Distributions by population approximate published Nottingham histologic grade series:

• Typical ranges of ~15–25% G1, ~40–50% G2, ~30–40% G3.
• SSA and AAW populations have somewhat higher G3 fractions, reflecting higher prevalence of aggressive subtypes such as TNBC.
• EUR/EAS reference groups have relatively higher G1 and lower G3 proportions.

Histologic type

Variable:

• histologic_type – one of:
  • NST_ductal – invasive carcinoma of no special type (ductal/NST).
  • Lobular – invasive lobular carcinoma.
  • Medullary_like – medullary or medullary-like carcinomas.
  • Mucinous – mucinous/colloid carcinomas.
  • Other_special – other special histologic types grouped.

Anchored to SEER and WHO-based estimates:

• NST/ductal: ~75–80% of invasive cases.
• Lobular: ~8–12% (slightly more common in some EUR/EAS cohorts).
• Remaining medullary-like, mucinous, and other special types together account for ~10–15%.

Lymphovascular invasion (LVI)

Variable:

• lvi_status – Absent or Present.
• Derived flag: lvi_present (True if Present).

Modeled prevalence is based on LVI reviews (typically 15–35% in invasive breast cancer):

• Lower LVI prevalence in G1 tumors (~5–10%).
• Intermediate in G2 (~15–25%).
• Higher in G3 (~30–40%), with modest variation by population.

Necrosis patterns

Variable:

• necrosis_pattern – one of:
  • None
  • Focal
  • Extensive
• Derived flag: necrosis_any (True if Focal or Extensive).

Patterns align with descriptions of necrosis in high-grade DCIS and invasive tumors:

• G1: mostly None with occasional Focal necrosis; Extensive rare.
• G2: increased Focal and some Extensive necrosis.
• G3: highest proportion of Extensive necrosis (e.g., comedo/central necrosis), plus substantial Focal necrosis.

Tumor-infiltrating lymphocytes (TILs)

Variable:

• tils_category – one of:
  • Low
  • Intermediate
  • High
• Derived flag: tils_high (True if High).

Informed by TIL consensus papers and TNBC cohorts:

• TNBC-enriched populations (e.g., some SSA and AAW groups) are modeled with higher fractions of High TILs (~20–24%) and substantial Intermediate TILs.
• EUR/EAS cohorts have more Low TILs and fewer High TILs, reflecting higher ER+/HER2− prevalence and lower immunogenicity.

File and schema

pathology_histology_data.parquet / pathology_histology_data.csv

Each row represents an invasive breast cancer case:

• Demographics

  • sample_id
  • population
  • region
  • is_SSA
  • is_reference_panel
  • sex
  • age

• Primary pathology features

  • tumor_grade (G1, G2, G3)
  • histologic_type
  • lvi_status
  • necrosis_pattern
  • tils_category

• Derived indicators

  • lvi_present
  • necrosis_any
  • tils_high

Generation

The dataset is generated using:

• pathology_histology/scripts/generate_pathology_histology.py

with configuration in:

• pathology_histology/configs/pathology_histology_config.yaml

and literature inventory in:

• pathology_histology/docs/LITERATURE_INVENTORY.csv

Key steps:

1. Cohort construction – multi-ancestry invasive cancer cohort with age and sex distributions by population.
2. Tumor grade assignment – sample tumor_grade by population according to Nottingham-like distributions.
3. Histologic type assignment – sample histologic_type by population, with NST/ductal as the dominant category.
4. LVI and necrosis assignment – sample lvi_status and necrosis_pattern conditional on population and tumor_grade, with higher LVI and necrosis rates in high-grade tumors.
5. TILs assignment – sample tils_category by population, reflecting higher immunogenicity/high-TIL fractions in some SSA and TNBC-enriched groups.

Validation

Validation is performed with:

• pathology_histology/scripts/validate_pathology_histology.py

and summarized in:

• pathology_histology/output/validation_report.md

Checks include:

• C01–C02 – Sample size and population counts vs config.
• C03 – Tumor grade distributions by population.
• C04 – Histologic type distributions by population.
• C05 – LVI distributions by population and grade.
• C06 – Necrosis pattern distributions by population and grade.
• C07 – TIL category distributions by population.
• C08 – Missingness in key variables.

The released version passes all checks within the configured tolerance, yielding an overall validation status of PASS.

Intended use

This dataset is intended for:

• Pathology-centric modeling of tumor aggressiveness and microenvironment across ancestries.
• Integration with other Electric Sheep Africa synthetic datasets (e.g., genomics, comorbidities, environmental exposures) to build multi-modal cancer models.
• Educational use for understanding relationships between grade, histologic type, LVI, necrosis, and TILs.

It is not suitable for:

• Estimating true prevalence of grades or histologies in any specific registry or country.
• Direct clinical decision-making or quality benchmarking.

All tumors are synthetic and non-identifiable.

Ethical considerations

• No patient-level data are used; all records are simulated.
• Population differences in grade, histology, LVI, or TILs are modeled for methodological realism and should not be used to stigmatize groups.
• Analyses should be contextualized with up-to-date cancer registry and pathology data.

License

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

Citation

If you use this dataset, please cite:

Electric Sheep Africa. "SSA Breast Pathology & Histology Dataset (Women, Multi-ancestry, Synthetic)." Hugging Face Datasets.

and, as appropriate, key pathology literature on histologic types, Nottingham grade, LVI, necrosis, and TILs in breast cancer.