electricsheepafrica/africa-synth-child-health-childhood-pneumonia-diarrhoea-all

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

Childhood Pneumonia & Diarrhoeal Disease Dataset

Abstract

This dataset provides 30,000 synthetic records (10,000 per scenario) of children aged 2–59 months presenting with pneumonia, diarrhoea, or both at LMIC health facilities in sub-Saharan Africa. Each record contains 33 variables including demographics, nutritional status, WHO IMCI pneumonia classification (pneumonia / severe / very severe), vital signs (respiratory rate, SpO2, temperature), danger signs, WHO dehydration assessment (none / some / severe), pathogen etiology (from GEMS-informed distributions including rotavirus, Shigella, Cryptosporidium, ST-ETEC), treatment received (ORS, zinc, antibiotics, oxygen, IV fluids), vaccination status, and clinical outcome. Three burden scenarios span mortality from 1.7% (low burden, good IMCI coverage) to 10.4% (high burden, weak health systems).

1. Introduction

Pneumonia and diarrhoea are the two leading infectious killers of children under 5 globally. In 2015, pneumonia caused an estimated 0.9 million deaths and 138 million episodes in children <5y (McAllister et al., Lancet Glob Health 2019). Diarrhoea kills approximately 525,000 children annually (WHO/UNICEF 2022). The GEMS study (Kotloff et al., Lancet 2013) identified four pathogens responsible for most moderate-to-severe diarrhoea: rotavirus, Cryptosporidium, ST-ETEC, and Shigella. Despite this burden, structured open datasets combining IMCI-classified pneumonia with dehydration-assessed diarrhoea and pathogen data are virtually non-existent.

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

2. Methodology

2.1 Epidemiological Parameterization

Parameter	Value	Source
Pneumonia episodes globally (2015)	138M (95% UI 86–226M)	McAllister et al., Lancet Glob Health 2019
Pneumonia deaths (2015)	0.9M (0.8–1.1M)	McAllister et al., 2019
Top 4 diarrhoea pathogens	Rotavirus 28%, Shigella 15%, Cryptosporidium 12%, ST-ETEC 10%	GEMS, Kotloff et al., Lancet 2013
ORS coverage in SSA	49% (range 24–66%)	DHS 30-country analysis, 2022
Zinc coverage in SSA	~15–25%	DHS multi-country, 2022
IMCI fast breathing: 2–12mo	≥50 bpm	WHO IMCI 2014
IMCI fast breathing: 12–59mo	≥40 bpm	WHO IMCI 2014
Severe pneumonia CFR (malnourished)	20–30%	Chisti et al., Lancet 2010
Diarrhoea deaths odds (MSD vs control)	OR 8.5 (95% CI 5.8–12.5)	GEMS, 2013

2.2 WHO IMCI Classification Criteria

Classification	Criteria	Source
Pneumonia	Fast breathing (age-specific threshold)	WHO IMCI 2014
Severe pneumonia	Chest indrawing	WHO IMCI 2014
Very severe disease	Danger signs: unable to drink, convulsions, lethargy, stridor	WHO IMCI 2014

2.3 WHO Dehydration Classification

Status	Signs	Source
No dehydration	Alert, drinks normally, skin pinch normal	WHO Pocket Book 2013
Some dehydration	Restless/irritable, drinks eagerly, skin pinch slow	WHO Pocket Book 2013
Severe dehydration	Lethargic, unable to drink, skin pinch very slow	WHO Pocket Book 2013

2.4 Scenario Design

Scenario	Setting	Severe Pneumonia	Severe Dehydration	ORS Coverage	Mortality
Low burden	Urban, good IMCI, vaccines	12%	4%	65%	1.7%
Moderate burden	Semi-urban, partial IMCI	18%	7%	49%	4.1%
High burden	Rural, weak health system	25%	12%	30%	10.4%

3. Schema

Column	Type	Description
id	int	Unique identifier
age_months	int	Age (2–59 months)
sex	categorical	M / F
weight_kg	float	Weight (kg)
nutritional_status	categorical	normal / MAM / SAM
rotavirus_vaccinated	binary	Rotavirus vaccine received
pcv_vaccinated	binary	Pneumococcal conjugate vaccine received
presenting_complaint	categorical	pneumonia_only / diarrhoea_only / both
temperature_c	float	Axillary temperature (°C)
cough	binary	Cough present
difficult_breathing	binary	Difficult breathing reported
respiratory_rate_bpm	int	Respiratory rate (breaths/min)
chest_indrawing	binary	Lower chest wall indrawing
nasal_flaring	binary	Nasal flaring present
stridor	binary	Stridor in calm child
spo2_pct	int	Oxygen saturation (%)
imci_pneumonia_class	categorical	none / pneumonia / severe / very_severe
diarrhoea_duration_days	int	Duration of diarrhoea (days)
stool_frequency_per_day	int	Stool frequency per day
bloody_stool	binary	Blood in stool (dysentery)
vomiting	binary	Vomiting present
sunken_eyes	binary	Sunken eyes
skin_pinch	categorical	normal / slow / very_slow
dehydration_status	categorical	none / some / severe
pathogen	categorical	Identified pathogen or not_tested
unable_to_drink	binary	Unable to drink or breastfeed
convulsions	binary	Convulsions
lethargy	binary	Abnormally sleepy or lethargic
received_ors	binary	Received ORS
received_zinc	binary	Received zinc supplementation
received_antibiotics	binary	Received antibiotics
received_oxygen	binary	Received oxygen therapy
received_iv_fluids	binary	Received IV fluids
outcome	categorical	discharged / admitted / referred / died

4. Validation

Key validation checks:

• Cross-scenario monotonicity: Mortality 1.7% → 4.1% → 10.4% ✓
• IMCI classification: Severe + very severe proportion increases with burden level ✓
• ORS coverage: Matches DHS pooled estimates (~49% in moderate) ✓
• Pathogen distribution: Rotavirus leading in <2y, Shigella in older children ✓
• SpO2: Hypoxemia prevalence increases with pneumonia severity ✓
• Malnutrition-mortality interaction: SAM cases have 3× higher CFR ✓

5. Usage

from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/childhood-pneumonia-diarrhoea-IMCI", "moderate_burden")
df = dataset["train"].to_pandas()

import pandas as pd
df = pd.read_csv("data/pneumonia_diarrhoea_moderate_burden.csv")

Regenerate with custom parameters:

python generate_dataset.py --scenario high_burden --n 20000 --seed 123
python generate_dataset.py --all-scenarios --n 10000 --seed 42

6. Limitations

• Synthetic: Not derived from real patient records. Must not be used for clinical decisions.
• No longitudinal data: Single presentation snapshot; no treatment response trajectory.
• Simplified etiology: Pathogen assignment is probabilistic; no co-infections modelled.
• No chest X-ray: IMCI classification only; no radiological confirmation of pneumonia.
• No seasonal variation: Rotavirus and pneumonia have seasonal peaks not captured here.
• ORS/zinc as binary: Real-world adherence is partial; dosing not modelled.
• No HIV co-morbidity: HIV+ children have higher pneumonia CFR; not modelled.

7. References

1. McAllister DA, et al. (2019). Global estimates of pneumonia morbidity & mortality in children <5y, 2000–2015. Lancet Glob Health, 7(1):e47–e57. PMID: 30497986
2. Kotloff KL, et al. (2013). Burden & aetiology of diarrhoeal disease (GEMS). Lancet, 382(9888):209–222. PMID: 23680352
3. WHO (2014). Revised IMCI classification of pneumonia in children.
4. WHO (2013). Pocket book of hospital care for children (2nd edition).
5. Agbagwa OE, et al. (2022). Prevalence and predictors of ORS, zinc, and other treatments for diarrhoea in SSA. PLoS ONE. PMID: 36227873
6. Liu L, et al. (2015). Global, regional, national causes of child mortality 2000–2013. Lancet, 385(9966):430–440.
7. WHO/UNICEF (2022). Diarrhoea: key facts. ~525,000 child deaths/yr.
8. Rudan I, et al. (2008). Epidemiology & etiology of childhood pneumonia. Bull WHO, 86(5):408–416.
9. Chisti MJ, et al. (2010). Pneumonia in severely malnourished children. Lancet, 375(9709):57.
10. Levine MM, et al. (2020). Diarrhoeal disease and subsequent risk of death in LMICs. Lancet Infect Dis.
11. WHO (2014). IMCI Chart Booklet.

Citation

@dataset{esa_pneumonia_diarrhoea_2025,
  title={Childhood Pneumonia and Diarrhoeal Disease Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/childhood-pneumonia-diarrhoea-IMCI}
}

License

CC-BY-4.0