electricsheepafrica/childhood-cancer-late-diagnosis

Childhood Cancer & Late Diagnosis Dataset

Abstract

This dataset provides 30,000 simulated childhood cancer records (10,000 per scenario) of children 0–15 years diagnosed with cancer across sub-Saharan African settings. Each record contains 40+ variables including cancer type (14 types including GICC index cancers), stage at diagnosis, diagnostic delay, pathology confirmation, chemotherapy/surgery/radiotherapy, treatment abandonment, complications, and 2-year outcomes. Three settings: paediatric oncology unit (29.5% 2-year survival), general paediatric ward (13.6%), and late referral (4.7%).

1. Introduction

Childhood cancer survival in SSA averages approximately 41% at 2 years, ranging from 27% (ALL) to 69% (HL), compared to >80% in HICs (CANCaRe Africa, Pediatr Blood Cancer 2025). Treatment abandonment affects nearly 25% of patients and significantly impacts outcomes. The WHO GICC targets 60% global survival for six index cancers. Burkitt lymphoma is the most common childhood cancer in SSA (30%). Advanced stage at presentation (60–80%) and diagnostic delays (median 12 weeks) are major barriers to improved outcomes.

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

2. Methodology

2.1 Parameterization

Parameter	Value	Source
2-year EFS (SSA overall)	40.8%	CANCaRe Africa, PBC 2025
Treatment abandonment	24%	CANCaRe Africa 2025
BL EFS	39.1%	CANCaRe Africa 2025
ALL EFS	27.4%	CANCaRe Africa 2025
WT EFS	51.5%	CANCaRe Africa 2025
RB EFS	39.9%	CANCaRe Africa 2025
HL EFS	68.5%	CANCaRe Africa 2025
Advanced stage at presentation	60–80%	Population registries, Lancet Oncol 2021
Diagnostic delay (median)	12 weeks	Paintsil et al., BMC Cancer 2020

2.2 Scenario Design

Scenario	Facility	Pathology	Chemo	Abandonment	2-Year Survival
Oncology unit	Dedicated unit	80%	Yes	28%	29.5%
General ward	Basic diagnostics	31%	Limited	48%	13.6%
Late referral	Minimal	30%	No	58%	4.7%

3. Schema

Column	Type	Description
id	int	Unique identifier
age_years	int	Child age
sex	categorical	M / F
distance_to_centre_km	int	Distance to treatment centre
health_insurance	binary	Has health insurance
caregiver_education	categorical	none / primary / secondary / tertiary
hiv_positive	binary	HIV status
nutritional_status	categorical	normal / moderate / severe malnutrition
cancer_type	categorical	14 types including BL, ALL, WT, RB, HL
gicc_index_cancer	binary	WHO GICC index cancer
diagnostic_delay_weeks	int	Weeks from symptom onset to diagnosis
stage_at_diagnosis	categorical	early / locally_advanced / metastatic
pathology_confirmed	binary	Pathology confirmation
imaging_done	categorical	ultrasound / xray / ct / none
referred_from_other	binary	Referred from another facility
traditional_medicine_first	binary	Sought traditional medicine first
chemotherapy	binary	Chemotherapy received
chemo_protocol	categorical	Protocol type
surgery	binary	Surgery performed
radiotherapy	binary	Radiotherapy
blood_transfusion	binary	Blood transfusion
supportive_care	binary	Supportive care
palliative_care	binary	Palliative care
febrile_neutropenia	binary	Febrile neutropenia
treatment_related_infection	binary	Treatment-related infection
tumour_lysis_syndrome	binary	Tumour lysis syndrome
chemo_toxicity	binary	Chemotherapy toxicity
treatment_abandoned	binary	Treatment abandoned
two_year_outcome	categorical	alive / died / lost_to_follow_up

4. Validation

Key validation checks:

• Survival gradient: Oncology 29.5% → General 13.6% → Late 4.7% ✓
• Burkitt lymphoma most common: ~25% of cases ✓
• Treatment abandonment: 28% → 48% → 58% across scenarios ✓
• Advanced stage: ~75% across all scenarios ✓
• GICC index cancers: ~66% of all cancers ✓
• HL best survival, ALL worst: Consistent with CANCaRe ✓

5. Usage

from datasets import load_dataset
dataset = load_dataset("electricsheepafrica/childhood-cancer-late-diagnosis", "paediatric_oncology_unit")
df = dataset["train"].to_pandas()

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

6. Limitations

• Simulated: Not derived from real cancer registries.
• No histopathology details: Cancer type is categorical, no grading/subtype.
• No treatment timeline: Snapshot, no cycle-by-cycle chemotherapy data.
• No imaging: Staging is categorical, no radiology images.
• No cost data: No out-of-pocket expenditure or catastrophic spending.

7. References

1. CANCaRe Africa (2025). GICC index cancers SSA. Pediatr Blood Cancer, e31997.
2. WHO GICC (2018). Global Initiative for Childhood Cancer.
3. Israels T, et al. (2021). Childhood cancer SSA. Lancet Oncol, 22(4):e169.
4. Stefan DC (2015). Childhood cancer Africa. Ecancermedicalscience, 9:587.
5. Paintsil V, et al. (2020). Diagnostic delay Ghana. BMC Cancer, 20:409.
6. Molyneux E, et al. (2012). Burkitt lymphoma SSA. Lancet Oncol, 13(12):e554.
7. Population registries (2021). Stage & survival SSA. Lancet Oncol, 22(1):e48.

Citation

@dataset{esa_childhood_cancer_2025,
  title={Childhood Cancer and Late Diagnosis Dataset},
  author={Electric Sheep Africa},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/electricsheepafrica/childhood-cancer-late-diagnosis}
}

License

CC-BY-4.0