electricsheepafrica/african-preeclampsia-dataset

African Preeclampsia/Eclampsia Synthetic Dataset

Risk Prediction and Maternal Mortality Prevention

Version: 1.0
Release Date: November 2024
Context: Sub-Saharan African Maternal Health
License: Research & Educational Use

---

Abstract

We present synthetic datasets for preeclampsia and eclampsia prediction in Sub-Saharan Africa, generated using literature-based odds ratios from WHO systematic reviews and African cohort studies. Preeclampsia affects 5.6-9.1% of African pregnancies (higher than global 2-8%) and accounts for 14% of maternal deaths. The datasets incorporate validated risk factors with documented ORs: previous preeclampsia (OR 7.19), chronic kidney disease (OR 5.6), chronic hypertension (OR 5.1), nulliparity (OR 2.91), obesity (OR 2.47), and family history (OR 2.90). Six datasets (21,000 samples total, 5.8 MB) provide configurations for risk stratification, severity classification (mild vs severe preeclampsia vs eclampsia), and HELLP syndrome prediction. Models trained on these data are expected to achieve AUC-ROC >0.85 for preeclampsia prediction and >0.80 for severe complications, serving as proof-of-concept for antenatal screening algorithms in resource-limited African settings where preeclampsia remains a leading preventable cause of maternal mortality.

Keywords: Preeclampsia, Eclampsia, HELLP Syndrome, Maternal Health, Pregnancy Complications, African Health, Risk Stratification

---

Key Features

Clinical Parameters

• Blood pressure: Systolic/diastolic measurements with severity thresholds (140/90 mild, ≥160/110 severe)
• Proteinuria: 24-hour measurements (300-2000 mg mild, >2000 mg severe)
• Laboratory values: Platelets, AST/ALT, creatinine, LDH, uric acid, hemoglobin
• HELLP syndrome markers: Hemolysis, elevated liver enzymes, low platelets

Risk Factors (Evidence-Based ORs)

• Previous preeclampsia: OR 7.19 (strongest predictor)
• Chronic kidney disease: OR 5.6
• Chronic hypertension: OR 5.1
• Nulliparity: OR 2.91
• Obesity (BMI >30): OR 2.47
• Family history: OR 2.90
• Multiple pregnancy: OR 2.93
• Pre-gestational diabetes: OR 3.56
• Advanced maternal age (>35): OR 1.96

African-Specific Factors

• Late presentation: 60-70% present in 3rd trimester (limited antenatal care access)
• Rural vs urban disparities: Rural areas 1.3× higher risk, fewer antenatal visits
• Magnesium sulfate availability: Modeled at 70% urban, 35% rural coverage
• Eclampsia rates: 10-20× higher than high-income countries
• Maternal mortality: 3-5% of eclampsia cases in Africa vs <1% globally

Complications Modeled

• HELLP syndrome: 10-20% of severe preeclampsia
• Eclampsia: 1-2% of preeclampsia cases (seizures)
• Maternal death: 0.1-5% depending on severity, access, treatment
• Fetal growth restriction: 10-25%
• Placental abruption: 1-4%
• Acute kidney injury: 1-5%
• Stroke: 0.1-0.3%

---

Dataset Collection

Dataset Inventory

Dataset	N	PE Cases	Severe %	Eclampsia	Deaths	Use Case
preeclampsia_africa_baseline_1000	1,000	404 (40%)	80%	13	4	Prototyping
preeclampsia_africa_large_5000	5,000	2,007 (40%)	79%	49	17	Main training
preeclampsia_africa_extra_large_10000	10,000	3,941 (39%)	79%	78	34	Deep learning
preeclampsia_africa_high_risk_2000	2,000	993 (50%)	79%	14	8	High-risk cohort
preeclampsia_africa_rural_2000	2,000	887 (44%)	78%	16	9	Rural settings
preeclampsia_africa_test_2000	2,000	779 (39%)	76%	9	2	Validation

Note: Prevalence enriched from baseline 7.5% to 39-50% to enable model training.

---

Expected Model Performance

Preeclampsia Prediction

• Logistic Regression: AUC 0.82-0.86
• Random Forest: AUC 0.85-0.90
• XGBoost: AUC 0.87-0.92

Severity Classification

• Multi-class F1: 0.75-0.82 (mild/severe/eclampsia)
• HELLP Prediction: AUC 0.78-0.85

Top Predictive Features (Expected)

1. Previous preeclampsia history
2. Chronic hypertension
3. Blood pressure measurements
4. Parity (nulliparous)
5. BMI category
6. Proteinuria levels
7. Gestational age at presentation
8. Laboratory values (platelets, creatinine, LFTs)

---

Usage Example

import pandas as pd
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

# Load data
train = pd.read_csv('preeclampsia_africa_large_5000.csv')
test = pd.read_csv('preeclampsia_africa_test_2000.csv')

# Features
features = ['age', 'parity', 'bmi', 'chronic_hypertension', 
            'pregestational_diabetes', 'previous_preeclampsia',
            'systolic_bp_mmhg', 'diastolic_bp_mmhg', 'proteinuria_24h_mg',
            'platelets_k_ul', 'creatinine_mg_dl']

X_train = train[features]
y_train = train['preeclampsia_status'].map({'Positive': 1, 'Negative': 0})

# Train model
model = RandomForestClassifier(n_estimators=100, class_weight='balanced')
model.fit(X_train, y_train)

# Expected AUC: 0.85-0.90

---

Limitations

What This Dataset IS:

• ✅ Antenatal risk stratification training
• ✅ Severity prediction modeling
• ✅ Healthcare access disparity analysis
• ✅ Sample size planning for prospective studies

What This Dataset IS NOT:

• ❌ Replacement for blood pressure monitoring
• ❌ Including longitudinal BP trajectories
• ❌ Capturing postpartum preeclampsia
• ❌ Validated for clinical decision-making without prospective trials

Mandatory Next Steps:

1. Prospective validation in African antenatal clinics
2. Integration with routine BP monitoring protocols
3. Healthcare worker training on interpretation
4. Ethical approval for clinical deployment

---

Citation

African Preeclampsia Synthetic Dataset (2024)
Risk-based prediction for maternal mortality prevention
Version 1.0, Generated November 2024
Organization: Electric Sheep Africa
License: CC-BY-4.0

Primary Sources:

• WHO Systematic Review: Preeclampsia Risk Factors (ORs)
• African Maternal Health Cohorts (Nigeria, Ghana, Kenya)
• HELLP Syndrome Meta-Analysis
• Eclampsia Mortality Data (Sub-Saharan Africa)

---

Quick Reference

Core Features (11):

features = ['age', 'parity', 'bmi', 'chronic_hypertension',
            'systolic_bp_mmhg', 'diastolic_bp_mmhg', 
            'proteinuria_24h_mg', 'platelets_k_ul',
            'ast_u_l', 'creatinine_mg_dl', 'previous_preeclampsia']

Targets:

• Binary: preeclampsia_status (Positive/Negative)
• Severity: severity (None/Mild/Severe)
• Complications: eclampsia, hellp_syndrome, maternal_death

Expected Performance:

• Preeclampsia prediction: AUC 0.85-0.92
• Severe complication risk: AUC 0.78-0.86

---

Status: Research Use Only - Prospective Validation Required
Last Updated: November 7, 2024
Contact: https://huggingface.co/electricsheepafrica