electricsheepafrica/ssa-comorbidities

SSA Comorbidities Dataset (Adults, Multi-ancestry, Synthetic)

Dataset summary

This dataset provides a synthetic adult cohort (18–85 years) with key infectious and chronic comorbidities relevant to cancer and NCD risk in sub-Saharan Africa (SSA) and reference populations.

Included conditions:

• HIV prevalence with ART status and viral load distributions.
• Diabetes prevalence (Type 2 diabetes).
• Hypertension prevalence.
• Tuberculosis (TB) co-infection rates among people living with HIV.

Parameters are qualitatively anchored to UNAIDS (HIV/viral suppression), IDF Diabetes Atlas / WHO AFRO (diabetes), hypertension reviews in SSA, and TB–HIV integration literature, but all individuals 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 reference): 1,500
  • EUR (European reference): 1,000
  • EAS (East Asian reference): 500

• Sex:

  • Female, Male (~55% vs 45% overall).

• Age:

  • 18–85 years.
  • Slightly older mean ages in AAW, EUR, and EAS vs SSA clusters.

Population labels align with other Electric Sheep Africa synthetic datasets (risk factors, cardiovascular metrics, surgical outcomes, etc.).

Comorbidity variables

HIV and viral load

Variables:

• hiv_status – Negative or Positive.
• has_hiv – boolean flag.
• on_art – on antiretroviral therapy.
• viral_suppressed – suppression indicator based on the ART/suppression model.
• viral_load_copies_ml – HIV RNA copies/mL (continuous, log-normal).

Patterns reflect:

• HIV prevalence highest in SSA_Southern, followed by SSA_East and SSA_Central, lower in SSA_West, and much lower in AAW, EUR, and EAS.
• ART coverage among people living with HIV (PLHIV) in SSA around 80–90%.
• Viral load suppression among those on ART in the 80–90% range, consistent with progress toward the UNAIDS 95–95–95 targets.

Viral load modeling:

• Suppressed individuals: log10 viral load ~ N(1.6, 0.3²), i.e., median tens of copies/mL, mostly <1,000 copies/mL.
• Unsuppressed individuals: log10 viral load ~ N(4.5, 0.6²), typically tens of thousands of copies/mL.
• viral_suppressed is sampled according to population-specific suppression probabilities informed by care cascade literature.

Diabetes

Variables:

• diabetes_status – No or Type2.
• has_diabetes – boolean flag.

Prevalence is set using IDF Diabetes Atlas estimates for the Africa region (~4–6% adults) and higher burdens in high-income reference groups:

• SSA clusters: Type 2 diabetes prevalence typically 4–9%, slightly higher in southern African settings.
• AAW/EUR/EAS: higher prevalence (8–12%), reflecting known burdens in African American, European, and East Asian populations.

Hypertension

Variables:

• hypertension_status – No or Yes.
• has_hypertension – boolean flag.

Calibrated using systematic and narrative reviews placing adult hypertension prevalence in SSA around 30–31%:

• SSA clusters: hypertension prevalence ~30–35%.
• AAW/EUR/EAS: slightly higher prevalences (mid-30s to ~40%), consistent with global high-burden NCD settings.

TB co-infection among PLHIV

Variable:

• tb_coinfection_status – No or Active_TB.

Modeling assumptions:

• Only HIV-positive individuals are at risk for Active_TB in this synthetic dataset (TB independent of HIV is not modeled here).
• TB co-infection probabilities are higher in southern and eastern SSA and lower in reference populations, reflecting high TB–HIV overlap in high-burden countries.

This structure is inspired by reviews of TB–HIV treatment integration and WHO TB/HIV burden reports.

File and schema

comorbidities_data.parquet / comorbidities_data.csv

Each row represents one adult individual:

• Demographics

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

• HIV / ART / viral load

  • hiv_status
  • has_hiv
  • on_art
  • viral_suppressed
  • viral_load_copies_ml

• Diabetes

  • diabetes_status
  • has_diabetes

• Hypertension

  • hypertension_status
  • has_hypertension

• TB

  • tb_coinfection_status

Generation

The dataset is generated using:

• comorbidities/scripts/generate_comorbidities.py

with configuration in:

• comorbidities/configs/comorbidities_config.yaml

and literature inventory in:

• comorbidities/docs/LITERATURE_INVENTORY.csv

Key steps:

1. Sample generation – multi-ancestry adult cohort with age and sex distributions by population.
2. HIV module – sample hiv_status by population and sex, then assign on_art, viral_suppressed, and viral_load_copies_ml using population-specific coverage and suppression parameters.
3. Diabetes and hypertension – sample diabetes_status and hypertension_status by population and sex, consistent with IDF/WHO and hypertension review ranges.
4. TB co-infection – among PLHIV, assign tb_coinfection_status using region-specific probabilities reflecting higher burden in southern and eastern SSA.

Validation

Validation is performed with:

• comorbidities/scripts/validate_comorbidities.py

and summarized in:

• comorbidities/output/validation_report.md

Checks include:

• C01–C02 – Sample size and population counts vs config.
• C03 – HIV prevalence by population and sex.
• C04 – Viral suppression among HIV-positive on ART.
• C05 – Diabetes prevalence by population and sex.
• C06 – Hypertension prevalence by population and sex.
• C07 – TB co-infection prevalence among PLHIV.
• C08 – Missingness in key variables.

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

Intended use

This dataset is intended for:

• Joint modeling of communicable and non-communicable comorbidities in SSA.
• Simulation studies of multi-morbidity burden and intervention scenarios (e.g., improving viral suppression or hypertension control).
• Educational use in epidemiology, public health, and global health analytics.

It is not suitable for:

• Estimating real-world prevalence in any specific country or clinic.
• Clinical decision-making, risk counselling, or program evaluation without real-world data.

All data are synthetic.

Ethical considerations

• No real patient data are used; all records are simulated.
• Population differences are modeled for methodological realism and should not be over-interpreted or used to stigmatize communities.
• Users should contextualize any analysis with up-to-date surveillance and peer-reviewed data.

License

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

Citation

If you use this dataset, please cite:

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

and, as appropriate, foundational sources such as UNAIDS HIV fact sheets, IDF Diabetes Atlas, hypertension reviews in sub-Saharan Africa, and TB–HIV integration literature.