electricsheepafrica/africa-synth-vaccination-childhood-immunisation-all

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

Synthetic Childhood Immunisation Coverage & Dropout Dataset (0–23 months)

Abstract

This dataset provides 30,000 synthetic records (10,000 per scenario) of childhood immunisation status for children aged 0-23 months in LMIC settings. Each record contains 28 variables: demographics, socioeconomic determinants (wealth quintile, maternal education, urban/rural, distance to facility), individual vaccine doses (BCG, OPV0-3, Penta1-3, PCV1-3, Rota1-2, IPV1, MCV1-2), and derived indicators (fully immunised, zero-dose, dropout). Coverage and equity gradients are parameterized from WHO/UNICEF WUENIC estimates, Gavi zero-dose analytics, and DHS vaccination equity analyses. Three scenarios (high, moderate, low coverage) capture the spectrum from well-performing programmes to fragile/conflict-affected settings.

1. Introduction

Globally, 14.3 million children received no routine vaccines ("zero-dose") in 2022 (WUENIC 2023). Immunisation coverage inequities by wealth, geography, and education remain a central challenge for the Immunization Agenda 2030. Open-access individual-level vaccination datasets from LMICs are scarce—DHS microdata requires registration and is survey-weighted, making it unsuitable for direct ML training.

This synthetic dataset addresses this gap for:

• Training ML models for zero-dose identification and dropout prediction
• Equity analysis and coverage gap modelling
• Prototyping immunisation programme dashboards
• Educational use in vaccinology and public health informatics

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

2. Methodology

2.1 Vaccine Schedule

Based on WHO Expanded Programme on Immunization (EPI) recommendations:

Vaccine	Doses	Schedule	Disease Target
BCG	1	Birth	Tuberculosis
OPV	4 (0-3)	Birth, 6, 10, 14 weeks	Poliomyelitis
Penta (DTP-HepB-Hib)	3	6, 10, 14 weeks	Diphtheria, tetanus, pertussis, hepatitis B, Hib
PCV	3	6, 10, 14 weeks	Pneumococcal disease
Rotavirus	2	6, 10 weeks	Rotavirus diarrhoea
IPV	1	14 weeks	Poliomyelitis
Measles (MCV)	2	9, 15 months	Measles

2.2 Equity Determinants

Individual coverage probability is modulated by five equity determinants, each with literature-grounded multipliers:

Determinant	Effect	Source
Wealth quintile (1-5)	0.65x (Q1) to 1.30x (Q5)	Restrepo-Méndez et al., Bull WHO 2016
Urban/Rural	1.10x urban, 0.90x rural	DHS pooled estimates
Maternal education	0.70x (none) to 1.20x (tertiary)	Arsenault et al., Lancet Global Health 2017
Distance to facility	-1.2% per km	DHS access analyses
Individual random effect	N(1.0, 0.08)	Unobserved heterogeneity

2.3 Scenario Design

Scenario	Context	BCG	Penta3	MCV1	MCV2	Zero-dose
High coverage	Well-performing LMIC	75.5%	34.8%	64.1%	40.1%	4.8%
Moderate coverage	Average LMIC	66.5%	27.8%	55.4%	33.4%	11.6%
Low coverage	Fragile/conflict	40.5%	10.7%	33.1%	15.8%	33.9%

3. Dataset Description

3.1 Schema

Column	Type	Description
id	int	Unique identifier
sex	categorical (M/F)	Biological sex
age_months	float	Age in months (0-23.9)
region_type	categorical	Urban or rural
ses_quintile	int (1-5)	Socioeconomic status quintile (1=poorest)
maternal_education	categorical	None, primary, secondary, tertiary
distance_to_facility_km	float	Distance to nearest health facility
bcg, opv0-3, penta1-3, pcv1-3, rota1-2, ipv1, mcv1-2	binary (0/1)	Vaccine dose received
total_basic_doses	int	Sum of BCG+Penta1-3+OPV1-3+MCV1 (max 8)
fully_immunised	binary	All age-appropriate vaccines received
dropout_penta1_penta3	binary	Received Penta1 but not Penta3
dropout_penta1_mcv1	binary	Received Penta1 but not MCV1
zero_dose	binary	No vaccines received at all
immunisation_status	categorical	fully_immunised / partially_immunised / zero_dose

4. Validation

4.1 Diagnostic Plots

5. Usage

5.1 Loading with HuggingFace datasets

from datasets import load_dataset

dataset = load_dataset("electricsheepafrica/synthetic-childhood-immunisation-coverage-dropout-WUENIC", "moderate_coverage")
df = dataset["train"].to_pandas()

5.2 Regenerating

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

6. Limitations

• Synthetic: Not real programme data. Not for programme evaluation.
• No campaign vaccines: Only routine EPI; does not model supplementary immunisation activities (SIAs).
• Cross-sectional: Single snapshot; does not capture timeliness or catch-up dynamics.
• Simplified equity model: Real equity determinants are more complex and context-specific.

7. References

1. WHO/UNICEF (2023). WUENIC Estimates of National Immunization Coverage.
2. Gavi (2023). Zero-dose children: Key data and analytics.
3. Restrepo-Méndez MC, et al. (2016). Inequalities in full immunization coverage. Bull WHO, 94:794-805.
4. Arsenault C, et al. (2017). Equity in antenatal care quality. Lancet Global Health, 5(11):e1079-e1088.
5. WHO (2022). Immunization Agenda 2030.
6. DHS Program. Vaccination module, multiple countries 2015-2023.

Citation

@dataset{esa_immunisation_2025,
  title={Synthetic Childhood Immunisation Coverage and Dropout Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/synthetic-childhood-immunisation-coverage-dropout-WUENIC}
}

License

CC-BY-4.0