electricsheepafrica/africa-synth-cancer-childhood-cancers-east-africa-all

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

Childhood Cancers - East Africa

Abstract

This synthetic dataset represents pediatric oncology data from east african registries and is designed to address the significant data gap in cancer research for sub-Saharan Africa. The dataset contains 2,000-3,000 per scenario records per scenario with key epidemiological parameters grounded in GLOBOCAN 2022 estimates, WHO reports, and peer-reviewed literature from the African Cancer Registry Network (AFCRN).

The age-standardized incidence rate (ASIR) of Childhood Cancers in the target population is approximately 15.0 per 100,000 population (GLOBOCAN 2022). This dataset provides training data for cancer epidemiology modeling, health systems research, and machine learning applications in oncology.

1. Introduction

1.1 Problem Statement

Cancer incidence in sub-Saharan Africa is rising rapidly, with estimated new cases reaching over 1 million annually by 2030. However, the region faces a critical shortage of granular cancer data for research, policy development, and health system planning. Population-based cancer registries cover less than 5% of the African population, creating significant gaps in understanding the true burden of disease.

1.2 Data Gap

• Limited population-based registry data outside major cities
• Missing survival and outcome data from most facilities
• Underrepresentation of pediatric and rare cancers
• Lack of treatment access and outcome metrics

1.3 Purpose

This dataset supports:

• Cancer burden estimation and projection modeling
• Health system capacity planning
• Machine learning for risk prediction and triage
• Epidemiological research on cancer patterns
• Policy development for cancer control programs

2. Methodology

2.1 Target Population

• Geographic scope: East Africa
• Population represented: Urban and rural populations
• Time period: Variable by data source (2010-2025)

2.2 Variable Selection

Variables were selected based on:

• IARC/WHO cancer registry standards
• Data availability in African cancer registries
• Clinical relevance for cancer control

2.3 Epidemiological Parameterization

All parameters are derived from:

• GLOBOCAN 2022 (IARC)
• WHO Cancer Reports
• African Cancer Registry Network (AFCRN)
• DHS/MICS survey data
• Peer-reviewed literature

2.4 Scenario Design

Scenario	Description	Records
low_burden	Low cancer burden setting	Varies by dataset
moderate_burden	Standard burden setting	Varies by dataset
high_burden	High burden / late presentation	Varies by dataset

2.5 Generation Process

Generation follows a conditional sampling approach based on directed acyclic graphs (DAGs) representing causal relationships between variables:

1. Sample demographic variables (age, sex, location)
2. Sample cancer type conditional on demographics
3. Sample clinical variables (stage, morphology, grade)
4. Sample treatment and outcome variables
5. Derive survival times from outcome models

3. Dataset Description

3.1 Key Variables

Pediatric oncology data from East African registries

3.2 Data Quality

• All categorical distributions validated against published literature
• Continuous variables modeled with appropriate statistical distributions
• Survival times based on exponential models with literature-derived parameters

4. Validation

4.1 Prevalence Verification

All prevalence values are validated against GLOBOCAN 2022 and published registry reports.

4.2 Distribution Quality

• Age and sex distributions match expected patterns
• Cancer type frequencies align with regional estimates

4.3 Clinical Plausibility

• No biologically impossible combinations
• Treatment patterns consistent with resource-limited settings

5. Usage

5.1 Loading with HuggingFace

from datasets import load_dataset
ds = load_dataset("electricsheepafrica/childhood-cancers-east-africa", "moderate_burden")

5.2 Loading from CSV

import pandas as pd
df = pd.read_csv("childhood_cancers_east_africa_moderate_burden.csv")

6. Limitations

• Synthetic data: Generated from aggregated statistics, not individual patient records
• Simplified correlations: May not capture complex dependencies
• Not for clinical use: Designed for research and ML training only

7. References

1. GLOBOCAN 2022. IARC Cancer Observatory.
2. African Cancer Registry Network (AFCRN).
3. WHO Cancer Control Reports.
4. DHS/MICS Survey Data.

Citation

@dataset{childhood_cancers_east_africa,
  title={Childhood Cancers - East Africa},
  author={Electric Sheep Africa},
  year={2025},
  publisher={HuggingFace},
  dataset_url={https://huggingface.co/datasets/electricsheepafrica/childhood-cancers-east-africa}
}

License

CC-BY-4.0