AFRIDIARRHEA is an ongoing computational epidemiology and artificial intelligence project that develops a multimodal fusion framework for estimating childhood diarrheal disease burden among children under five years of age in Africa. The platform integrates Bayesian epidemiological modeling, machine learning, temporal forecasting, geospatial analytics, pathogen-attribution modeling, environmental intelligence, uncertainty quantification, and multimodal data fusion within a unified decision-support architecture.
The framework combines epidemiological, climatic, environmental, nutritional, healthcare-access, WASH (Water, Sanitation and Hygiene), geospatial, and pathogen-specific information to estimate mortality, morbidity, hospitalization burden, pathogen-attributed disease burden, and associated uncertainties. By linking multiple data streams into a single analytical framework, AFRIDIARRHEA provides geographically resolved, uncertainty-aware estimates capable of supporting public health planning, vaccine prioritization, climate-health adaptation, outbreak preparedness, and child survival programs in resource-constrained settings.
To develop a scalable and reproducible multimodal fusion framework capable of accurately estimating childhood diarrheal disease burden while integrating epidemiological, environmental, spatial, temporal, healthcare, and pathogen-specific determinants of disease transmission.
The framework aims to:
Estimate under-five diarrheal morbidity and mortality burden.
Quantify severe disease and hospitalization burden.
Attribute disease burden to major enteric pathogens.
Integrate environmental and climate-related disease drivers.
Generate uncertainty-aware burden estimates.
Identify disease hotspots and vulnerable populations.
Support evidence-based public health policy and intervention planning.
A hierarchical epidemiological baseline model estimates disease burden while accounting for uncertainty, country effects, spatial heterogeneity, environmental risk, healthcare access, and temporal variability.
Advanced machine-learning algorithms capture nonlinear interactions among multiple disease determinants, including:
XGBoost-style boosting models
LightGBM-style boosting models
Ensemble prediction architectures
Seasonal and lag-dependent disease dynamics are modeled to capture temporal patterns in diarrheal disease transmission and burden.
Spatial risk models identify geographic variations in disease burden and highlight regions with elevated transmission risk.
The model estimates pathogen-specific contributions to childhood diarrheal mortality and morbidity for:
Rotavirus
Shigella
Norovirus GII
ST-ETEC
Cholera
Cryptosporidium
Campylobacter
Adenovirus 40/41
Environmental and climate-related variables incorporated into the framework include:
Temperature
Rainfall
Humidity
Flood risk
Water quality indicators
Sanitation access
Satellite-derived environmental risk metrics
The framework accounts for pathogen co-occurrence and coinfection effects, improving pathogen-attributed burden estimation.
Outputs from epidemiological, machine-learning, temporal, geospatial, environmental, and pathogen-attribution models are fused into a single integrated prediction framework.
Prediction intervals and uncertainty calibration provide confidence bounds around disease-burden estimates, improving interpretability and decision support.
The framework identified substantial heterogeneity in childhood diarrheal disease burden across Kenya, Somaliland, and Zimbabwe, demonstrating the ability to capture geographic variation in disease outcomes.
Zimbabwe exhibited the highest modeled under-five diarrheal mortality burden, followed by Somaliland and Kenya.
Zimbabwe also demonstrated the highest estimated diarrheal morbidity burden, while Somaliland showed a similarly elevated disease burden.
Somaliland recorded the highest hospitalization burden, suggesting a greater proportion of clinically severe disease requiring healthcare services.
Rotavirus emerged as the dominant contributor to pathogen-attributed childhood diarrheal mortality, followed by Shigella.
Norovirus GII, Cholera, and Cryptosporidium contributed moderate levels of mortality burden, while ST-ETEC, Campylobacter, and Adenovirus 40/41 contributed smaller but still significant burdens.
The multimodal fusion model consistently outperformed individual component models and the Bayesian baseline, demonstrating improved predictive accuracy and robustness.
Observed-versus-predicted analyses showed stronger agreement between model predictions and observed disease outcomes compared with conventional approaches.
Most observed disease burdens fell within predicted 95% uncertainty intervals, indicating reliable uncertainty estimation and model stability.
Accurate estimation of childhood diarrheal disease burden remains a major challenge in many African countries due to limited surveillance infrastructure, incomplete mortality reporting, pathogen-attribution uncertainty, climate variability, and healthcare disparities.
AFRIDIARRHEA addresses these challenges by integrating epidemiological modeling, machine learning, environmental intelligence, geospatial analytics, and pathogen attribution within a unified multimodal architecture.
The framework provides a powerful tool for:
Disease burden estimation
Vaccine prioritization
WASH intervention planning
Outbreak preparedness
Climate-health adaptation
Healthcare resource allocation
Child survival initiatives
Public health policy development
particularly in low-resource settings where high-quality epidemiological data may be limited.
XGBoost
LightGBM
Ensemble Learning
Multimodal Fusion Architectures
Bayesian Hierarchical Models
Disease Burden Estimation Frameworks
Pathogen Attribution Models
Coinfection Adjustment Models
Satellite-Derived Environmental Indicators
Climate Risk Modeling
Geospatial Risk Mapping
Spatial Epidemiology
World Health Organization (WHO)
UNICEF
World Bank
Global Burden of Disease (GBD)
Environmental and Climate Datasets
WASH Indicators
Healthcare Accessibility Metrics
Python
NumPy
Pandas
Scikit-Learn
XGBoost
LightGBM
Matplotlib
Geospatial Analytics Libraries
Current Status: Manuscript completed and submitted as a proof-of-concept multimodal burden-estimation framework.
Integration of real-world epidemiological surveillance data.
Incorporation of laboratory-confirmed pathogen datasets.
Expansion to additional African countries.
Integration of satellite-derived climate and environmental intelligence.
Development of county-level burden-estimation dashboards.
Real-time disease surveillance capability.
Enhanced uncertainty quantification using full Bayesian inference.
Integration with public-health decision-support systems.
Validation against national mortality and morbidity databases.
To establish a continent-wide AI-powered disease burden estimation and forecasting platform capable of supporting vaccine policy, outbreak preparedness, climate-health adaptation, child survival programs, and evidence-based public health planning across Africa.