electricsheepafrica/africa-synth-maternal-health-maternal-health-pregnancy-all

⚠️ Synthetic dataset — Parameterized from published SSA literature, not real observations. Not suitable for empirical analysis or policy inference.

Synthetic Maternal Health & Pregnancy Complications Dataset

The complete bundle — including the full dataset (35,000 rows), trained xgboost model (AUC-ROC: 0.990), fully-executed notebook, and full Paper— is available on Gumroad:

👉 Get the full bundle on Gumroad for $30

---

Abstract

This dataset provides 30,000 synthetic records (10,000 per scenario) of pregnant women attending antenatal care (ANC) in LMIC facility settings. Each record contains 16 clinically relevant variables including demographics (age, gravidity, parity), clinical measurements (blood pressure, hemoglobin, fasting glucose, BMI, proteinuria), risk factors (HIV status), and outcomes (primary complication, pregnancy outcome, risk level). All distributions are parameterized from WHO ANC guidelines, the Lancet Maternal Health series, UNAIDS, IDF Diabetes Atlas, and DHS surveys. Three burden scenarios (low, moderate, high) capture the spectrum from well-resourced urban facilities to under-resourced high-HIV settings.

1. Introduction

Maternal mortality remains unacceptably high in LMICs, with approximately 287,000 deaths annually (WHO 2023). Hypertensive disorders (preeclampsia/eclampsia), obstetric hemorrhage, and sepsis account for over 50% of maternal deaths (Say et al., 2014). Anemia affects 40-60% of pregnant women in Sub-Saharan Africa, and gestational diabetes prevalence is rising across LMICs.

Open-access clinical datasets of maternal health from LMIC contexts are scarce due to privacy regulations and fragmented health information systems. This synthetic dataset fills that gap for:

• Training ML models for pregnancy risk stratification
• Benchmarking complication prediction algorithms
• Prototyping ANC clinical decision support tools
• Educational use in global maternal health curricula

This dataset is entirely synthetic. It must not be used for clinical decision-making.

2. Methodology

2.1 Target Population

Pregnant women aged 15-49 presenting for ANC or delivery at LMIC health facilities.

2.2 Epidemiological Parameterization

Parameter	Value	Source
Preeclampsia prevalence	2-10% by setting	Abalos et al., Hypertension in Pregnancy 2013
Eclampsia incidence	0.1-1.5%	Abalos et al., 2013
GDM prevalence	3-8% in LMIC	IDF Diabetes Atlas 2021
Obstetric hemorrhage	2-8%	Say et al., Lancet Global Health 2014
Anemia in pregnancy (Hb<11)	40-65% in SSA	Stevens et al., Lancet Global Health 2013
Severe anemia (Hb<7)	2-15% by setting	Stevens et al., 2013
HIV prevalence (women)	4-18% by setting	UNAIDS 2023
C-section rate (LMIC)	5-18%	DHS Program; Vogel et al., 2014

2.3 Scenario Design

Scenario	Context	Preeclampsia	Eclampsia	GDM	Hemorrhage	Severe Anemia	HIV
Low burden	Urban LMIC, functional ANC	4.2%	0.4%	6.8%	2.6%	2.9%	4.2%
Moderate burden	District hospital	7.5%	1.0%	10.6%	3.9%	5.8%	8.0%
High burden	Under-resourced / high-HIV	11.6%	2.1%	13.5%	6.0%	9.9%	18.1%

2.4 Risk Factor Modelling

Complication probabilities are adjusted by individual risk factors:

• Preeclampsia: OR ~1.8 for age >35, OR ~2.0 for BMI >30, OR ~1.5 for primigravida (WHO 2016)
• GDM: OR ~1.5 for age >30, OR ~2.5 for BMI >25 (IDF 2021)
• Blood pressure: Conditional on complication, with age and BMI adjustments
• Hemoglobin: Background anemia prevalence plus complication-specific shifts

3. Dataset Description

3.1 Schema

Column	Type	Units	Range	Description
id	int	—	1-10000	Unique identifier
age_years	int	years	15-49	Maternal age
gravidity	int	—	1-16	Total pregnancies including current
parity	int	—	0-15	Previous deliveries
gestational_age_weeks	float	weeks	6.0-42.0	GA at clinical visit
bmi_pre_pregnancy	float	kg/m²	14.0-48.0	Pre-pregnancy BMI
systolic_bp_mmhg	int	mmHg	70-220	Systolic blood pressure
diastolic_bp_mmhg	int	mmHg	40-140	Diastolic blood pressure
hemoglobin_gdl	float	g/dL	3.0-17.0	Hemoglobin concentration
anemia_status	categorical	—	none/mild/moderate/severe	WHO pregnancy anemia classification
fasting_glucose_mgdl	int	mg/dL	45-250	Fasting blood glucose
proteinuria	ordinal	—	0-4	Urine protein (0=none, 4=≥+3)
hiv_status	binary	—	0/1	HIV serostatus
anc_visits	int	—	0-15	Number of ANC visits to date
delivery_mode	categorical	—	vaginal/caesarean	Mode of delivery
primary_complication	categorical	—	6 classes	Primary pregnancy complication
pregnancy_outcome	categorical	—	live_birth/stillbirth/maternal_death	Pregnancy outcome
risk_level	categorical	—	low/moderate/high	Composite risk classification

3.2 Classification Criteria

Classification	Criteria	Source
Anemia (mild)	Hb 10.0-10.9 g/dL	WHO 2011
Anemia (moderate)	Hb 7.0-9.9 g/dL	WHO 2011
Anemia (severe)	Hb < 7.0 g/dL	WHO 2011
Hypertension	SBP ≥140 or DBP ≥90 mmHg	WHO ANC 2016
Severe hypertension	SBP ≥160 or DBP ≥110 mmHg	WHO ANC 2016
GDM (fasting)	Fasting glucose ≥92 mg/dL	IADPSG/WHO 2013

4. Validation

4.1 Cross-Scenario Monotonicity

All adverse outcomes increase monotonically from low → moderate → high burden: anemia (48% → 58% → 65%), hypertension (5% → 9% → 14%), HIV (4% → 8% → 18%), stillbirth (0.6% → 0.8% → 1.4%).

4.2 Diagnostic Plots

5. Usage

5.1 Loading with HuggingFace datasets

from datasets import load_dataset

dataset = load_dataset("electricsheepafrica/synthetic-maternal-pregnancy-complications-WHO-ANC", "moderate_burden")
df = dataset["train"].to_pandas()

5.2 Loading directly from CSV

import pandas as pd

df = pd.read_csv("data/maternal_moderate_burden.csv")
high_risk = df[df['risk_level'] == 'high']
print(f"High risk: {len(high_risk)/len(df)*100:.1f}%")

5.3 Regenerating

pip install numpy pandas scipy matplotlib
python generate_dataset.py --all-scenarios --n 10000 --seed 42
python validate_dataset.py

6. Limitations & Ethical Considerations

• Synthetic data: No real patients. Not for clinical use.
• Simplified comorbidity: Each woman has one primary complication; real pregnancies often involve multiple concurrent conditions.
• No temporal modelling: Single timepoint snapshot; does not capture ANC trajectory or disease progression.
• HIV simplification: HIV status modelled as binary; does not capture viral load, ART status, or CD4 count.
• Geographic generalization: Parameters drawn from pooled LMIC estimates; may not represent any single country precisely.

7. References

1. WHO (2016). WHO Recommendations on Antenatal Care for a Positive Pregnancy Experience. Geneva.
2. Say L, et al. (2014). Global causes of maternal death. Lancet Global Health, 2(6):e323-333.
3. Abalos E, et al. (2013). Global and regional estimates of preeclampsia and eclampsia. Hypertension in Pregnancy, 32(sup1):36.
4. IDF (2021). IDF Diabetes Atlas, 10th edition.
5. Stevens GA, et al. (2013). Global, regional, and national trends in haemoglobin concentration. Lancet Global Health, 1(1):e16-25.
6. UNAIDS (2023). Global HIV & AIDS statistics fact sheet.
7. WHO (2023). Trends in maternal mortality 2000-2020. Geneva.
8. DHS Program. Demographic and Health Surveys, multiple countries.
9. Vogel JP, et al. (2014). Use of the Robson classification. Lancet Global Health, 2(5):e260-270.
10. Souza JP, et al. (2013). Moving beyond essential interventions. Lancet, 381(9879):1747-1755.

Citation

@dataset{esa_maternal_2025,
  title={Synthetic Maternal Health and Pregnancy Complications Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/synthetic-maternal-pregnancy-complications-WHO-ANC}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC-BY-4.0) license.