electricsheepafrica/congenital-anomalies-birth-defects

Congenital Anomalies & Birth Defects Dataset

Abstract

This dataset provides 30,000 simulated congenital anomaly records (10,000 per scenario) of neonates and children with birth defects across sub-Saharan African settings. Each record contains 40+ variables including anomaly type (10 categories with subtypes), organ system, severity, prenatal detection, surgical intervention, and outcomes including mortality and disability. Three settings: specialist centre (16% mortality, 45% surgery), district hospital (31% mortality, 7% surgery), and rural birth (43% mortality, 0% surgery).

1. Introduction

Congenital anomalies account for ~240,000 neonatal deaths annually worldwide (WHO 2024). Prevalence in SSA is modelled at ~2% of births, likely underestimated (Tekendo-Ngongang et al., 2020). CHD is the most common (25%), followed by NTDs (15%) and cleft lip/palate (15%). NTDs have the largest surgical amenability (76%), yet folic acid periconceptional coverage in SSA is only 10–20%. Surgery can reduce mortality by >60% for many conditions, but paediatric surgical capacity in SSA is severely limited.

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

2. Methodology

2.1 Parameterization

Parameter	Value	Source
Prevalence (SSA)	~2% of births	Tekendo-Ngongang et al., 2020
CHD proportion	25% of anomalies	WHO 2024
NTD proportion	15%	March of Dimes
Cleft proportion	15% (1:700 births)	Smile Train 2024
NTD surgical amenability	76%	Adeleye et al., 2023
Folic acid coverage (SSA)	10–20%	WHO 2024
Abdominal wall defect mortality	40–60%	Poenaru 2012

2.2 Scenario Design

Scenario	Facility	Surgery	NICU	Mortality	Prenatal
Specialist centre	Paediatric surgical	45%	Yes	16%	20%
District hospital	Basic neonatal	7%	No	31%	8%
Rural birth	Home/health centre	0%	No	43%	1%

3. Schema

Column	Type	Description
id	int	Unique identifier
gestational_age_weeks	int	Gestational age
preterm	binary	Preterm (<37 weeks)
sex	categorical	M / F
birth_weight_kg	float	Birth weight
low_birth_weight	binary	<2.5 kg
multiple_birth	binary	Multiple birth
maternal_age	int	Maternal age
anc_visits	int	Antenatal care visits
folic_acid_periconceptional	binary	Periconceptional folic acid
consanguinity	binary	Consanguinity
maternal_diabetes	binary	Maternal diabetes
maternal_epilepsy_meds	binary	Maternal epilepsy medications
family_history_anomaly	binary	Family history
anomaly_category	categorical	10 types: CHD / NTD / cleft / clubfoot / etc.
organ_system	categorical	cardiovascular / nervous / craniofacial / etc.
anomaly_subtype	categorical	Specific subtype
severity	categorical	mild / moderate / severe
isolated	binary	Isolated anomaly
associated_anomalies	int	Number of associated anomalies
surgically_amenable	binary	Surgically amenable
prenatal_detection	binary	Prenatal detection
age_at_detection_days	int	Age at detection
diagnosed_clinically	binary	Clinical diagnosis
imaging_done	binary	Imaging performed
echocardiogram	binary	Echocardiogram
genetic_testing	binary	Genetic testing
surgery_performed	binary	Surgery performed
surgery_type	categorical	cardiac_repair / cleft_repair / ponseti / etc.
age_at_surgery_days	int	Age at surgery
referred_tertiary	binary	Referred to tertiary
nicu_admission	binary	NICU admission
rehabilitation	binary	Rehabilitation
outcome	categorical	survived / died
disability	categorical	motor / cognitive / feeding / limb / none

4. Validation

Key validation checks:

• Mortality gradient: 16% → 31% → 43% across scenarios ✓
• CHD most common: 25% of all anomalies ✓
• Surgery impact: 45% at specialist vs 0% rural, with clear mortality reduction ✓
• Abdominal wall defects: Highest mortality (~45%) ✓
• Prenatal detection: 20% at specialist vs 1% rural ✓
• Folic acid coverage: ~15% consistent with SSA estimates ✓

5. Usage

from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/congenital-anomalies-birth-defects", "district_hospital")
df = dataset["train"].to_pandas()

python generate_dataset.py --all-scenarios --n 10000 --seed 42

6. Limitations

• Simulated: Not derived from real birth defect registries.
• No imaging: No ultrasound images or echocardiograms.
• No longitudinal follow-up: Snapshot, no developmental milestones.
• No cost data: No surgical or rehabilitation cost information.
• Simplified genetics: No karyotype or molecular genetics data.

7. References

1. WHO (2024). Congenital disorders fact sheet.
2. March of Dimes. Global report on birth defects.
3. Adeleye AO, et al. (2023). Surgical amenability of congenital anomalies SSA.
4. Poenaru D (2012). Surgical conditions in SSA neonates. World J Surg, 36(5):1101.
5. WHO (2024). Folic acid supplementation guidelines.
6. Tekendo-Ngongang C, et al. (2020). Congenital anomalies SSA surveillance. PLoS Med.
7. Smile Train (2024). Cleft lip and palate: global statistics.

Citation

@dataset{esa_congenital_anomalies_2025,
  title={Congenital Anomalies and Birth Defects Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/congenital-anomalies-birth-defects}
}

License

CC-BY-4.0