electricsheepafrica/africa-synth-nutrition-nutrition-rehabilitation-all

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

Nutrition Rehabilitation & Therapeutic Feeding Dataset

Abstract

This dataset provides 30,000 synthetic records (10,000 per scenario) of children aged 6–59 months admitted to CMAM nutrition programmes in sub-Saharan Africa. Each record contains 40 variables including anthropometry (MUAC, WHZ, oedema), SAM/MAM classification, medical complications, therapeutic food type (RUTF/F75/F100), treatment protocol, weight gain, and discharge outcomes benchmarked against Sphere standards. Three programme types: stabilisation centre (complicated SAM), OTP (uncomplicated SAM), and SFP (MAM).

1. Introduction

An estimated 45.4 million children under 5 suffer from wasting globally, with 13.9 million in SSA (UNICEF 2023). SAM is defined as MUAC <115mm, WHZ <-3, or bilateral oedema (WHO 2013). CMAM achieves cure rates >75% with death <10% and default <15% (Sphere Standards 2018; Collins et al., Lancet 2006). Systematic antibiotics reduce SAM mortality (Trehan et al., NEJM 2013). RUTF enables outpatient treatment of uncomplicated SAM.

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

2. Methodology

2.1 Parameterization

Parameter	Value	Source
Global wasting (under-5)	45.4M	UNICEF 2023
SAM definition	MUAC <115mm / WHZ <-3 / oedema	WHO 2013
Sphere cure target	>75%	Sphere Standards 2018
Sphere death target	<10%	Sphere Standards 2018
OTP recovery time	35–49 days median	Kerac et al., PLoS One 2014
Weight gain target	5–10 g/kg/day	WHO/UNICEF guidelines

2.2 Scenario Design

Scenario	Programme	Classification	Cure	Death	Default
Stabilisation centre	Inpatient SC	Complicated SAM	59%	30%	11%
OTP programme	Outpatient	Uncomplicated SAM	82%	5%	14%
SFP programme	Supplementary	MAM	87%	1%	12%

3. Schema

Column	Type	Description
id	int	Unique identifier
age_months	int	Age (6–59 months)
sex	categorical	M / F
caregiver	categorical	mother / grandmother / father / other
breastfeeding	binary	Currently breastfeeding
classification	categorical	SAM / MAM
muac_mm	int	Mid-upper arm circumference (mm)
whz	float	Weight-for-height z-score
bilateral_oedema	binary	Bilateral pitting oedema
oedema_grade	categorical	none / + / ++ / +++
waz	float	Weight-for-age z-score
haz	float	Height-for-age z-score
stunted	binary	HAZ < -2
weight_kg	float	Admission weight
appetite_test	categorical	pass / fail
complicated	binary	Complicated SAM
dehydration	categorical	none / some / severe
diarrhoea	binary	Diarrhoea present
vomiting	binary	Vomiting
fever	binary	Fever
pneumonia	binary	Pneumonia
malaria_positive	binary	Malaria RDT positive
hiv_status	categorical	positive / negative / unknown
tb_suspected	binary	TB suspected
anaemia_clinical	binary	Clinical anaemia
skin_lesions	binary	Skin lesions
programme_type	categorical	SC / OTP / SFP
therapeutic_food	categorical	F75 / F100 / RUTF / RUSF / CSB_plus / local_RUTF
systematic_antibiotics	binary	Systematic antibiotics given
antibiotic_type	categorical	amoxicillin / amoxicillin_plus_metronidazole / ampicillin_gentamicin / none
vitamin_a	binary	Vitamin A given
folic_acid	binary	Folic acid given
deworming	binary	Deworming given
antimalarial	binary	Antimalarial given
length_of_stay_days	int	Duration in programme
weight_gain_gkgday	float	Weight gain (g/kg/day)
muac_discharge_mm	int	MUAC at discharge
weekly_visits_attended	int	Weekly visits attended
outcome	categorical	cured / died / defaulted / non_responder

4. Validation

Key validation checks:

• OTP Sphere compliance: Cure 82% (>75% ✓), Death 5% (<10% ✓), Default 14% (<15% ✓)
• SFP Sphere compliance: Cure 87%, Death 1%, Default 12% ✓
• MUAC thresholds: SAM <115mm, MAM 115–124mm correctly distributed ✓
• Weight gain: OTP 5.5 g/kg/day, SC 8.0 g/kg/day ✓
• Oedema gradient: SC 34% > OTP 15% > SFP 0% ✓

5. Usage

from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/nutrition-rehabilitation", "otp_programme")
df = dataset["train"].to_pandas()

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

6. Limitations

• Synthetic: Not derived from real CMAM programme data.
• No longitudinal tracking: Single admission record, no weekly follow-up series.
• Simplified weight gain: Single summary statistic, not weekly trajectory.
• No dietary data: No detailed food intake or feeding frequency.
• SC death rate high: Stabilisation centre scenario reflects complicated SAM with high baseline mortality.

7. References

1. WHO (2013). Guideline: updates on management of SAM in children.
2. Collins S, et al. (2006). Community-based therapeutic care. Lancet, 368(9551):1977–1987.
3. UNICEF (2023). Levels and trends in child malnutrition.
4. Trehan I, et al. (2013). Antibiotics as part of SAM management. NEJM, 368(5):425–435.
5. Kerac M, et al. (2014). SAM outcomes. PLoS One, 9(6):e100573.
6. Sphere Association (2018). Sphere Handbook: minimum standards.
7. Maust A, et al. (2015). RUTF vs local foods for MAM. Am J Clin Nutr, 101(3):632–637.
8. WHO/UNICEF (2009). WHO child growth standards for SAM.

Citation

@dataset{esa_nutrition_2025,
  title={Nutrition Rehabilitation and Therapeutic Feeding Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/nutrition-rehabilitation}
}

License

CC-BY-4.0