Spatially-informed Bayesian Distributed Lag Non-Linear Models (SB-DLNM) for Influenza in NSW

📋 Overview

This repository contains the complete implementation of Spatially-informed Bayesian Distributed Lag Non-Linear Models (SB-DLNM) for investigating the complex relationships between meteorological factors and influenza transmission across New South Wales (NSW), Australia.

Our framework quantifies the delayed and non-linear effects of temperature, relative humidity, and rainfall on influenza notifications, analyzing over 1.2 million laboratory-confirmed cases across 15 Local Health Districts (LHDs) from 2000-2023. The model captures both non-linear and delayed effects of weather exposures while accounting for spatial dependencies between neighboring health districts, providing crucial insights for public health planning and disease surveillance.

🎯 Key Features

• Bayesian DLNM Framework: Captures non-linear and lagged effects up to 4 weeks using cross-basis splines
• Spatial Integration: Conditional autoregressive (CAR) priors enable neighboring LHDs to borrow strength
• Comprehensive Analysis: Over 1.2 million laboratory-confirmed cases (2000-2023) linked to district-level meteorological data
• Multiple Model Comparisons: Four designs compared (case-crossover vs. time-series, with/without spatial pooling)
• Early Warning Capability: Framework supports district-level surge prediction weeks in advance
• Publication-Ready Outputs: Automated generation of figures, tables, and diagnostic plots

👥 Authors

Mohammad Afzal Khan¹, Oyelola Adegboye², Shiyang Lyu¹, Kiki Maulana Adhinugraha³, Theophilus I. Emeto⁴⁵, and David Taniar¹

¹ Faculty of Information Technology, Monash University, Melbourne, VIC 3800, Australia
² Menzies School of Health Research, Charles Darwin University, Darwin, NT 0800, Australia
³ School of Computing and Information Technology, La Trobe University, Melbourne, VIC 3086, Australia
⁴ Australian Institute of Tropical Health and Medicine, James Cook University, Townsville, QLD 4811, Australia
⁵ College of Public Health, Medical and Veterinary Sciences, James Cook University, Townsville, QLD 4811, Australia

*Corresponding authors

📊 Data

Data Sources

• Influenza Data: NSW Health Notifiable Conditions Information Management System (NCIMS)
• Meteorological Data: Australian Bureau of Meteorology (BoM)
• Spatial Boundaries: NSW Ministry of Health

Data Description

• Study Period: January 2000 - December 2023 (24 years)
• Total Cases: Over 1.2 million laboratory-confirmed influenza notifications
• Geographic Coverage: 15 Local Health Districts (LHDs) in NSW
• Temporal Resolution: Monthly aggregation for analysis
• Variables:
  • Laboratory-confirmed influenza cases (Types A, B, and total)
  • Daily mean temperature (°C)
  • Daily mean relative humidity (%)
  • Daily total rainfall (mm)

🚀 Quick Start

Prerequisites

• R ≥ 4.0.0
• Required R packages (see requirements.R)

Key Dependencies

• Statistical Modeling: dlnm, splines, survival, coda
• Spatial Analysis: sf, spdep, ape
• Data Processing: dplyr, tidyr, lubridate, data.table
• Visualization: ggplot2, plotly, viridis, patchwork

Installation

```bash

Clone the repository

git clone https://github.com/afzal0/SB-DLNM-InfluenzaNSW.git cd SB-DLNM-InfluenzaNSW

Install required packages

Rscript requirements.R ```

Running the Analysis

For a complete analysis pipeline:

```r

Run the entire analysis

source("Data Prepration.R") source("Data Modelling.R") source("AccuracyAssessment.R") source("Visualisation.R") source("Opt-Vis1-DLNMCurves.R") source("Opt-Vis2-Cumulative Effects.R") ```

For detailed workflow instructions, see WORKFLOW.md.

📁 Repository Structure

SB-DLNM-Influenza_NSW/ ├── data/ # Input data files ├── spatial_data/ # Spatial boundary files ├── output/ # Processed data outputs ├── new_output/ # Model results and diagnostics ├── new_output_acc/ # Model accuracy assessments ├── Visualisations/ # Generated figures ├── Data Prepration.R # Data preprocessing script ├── Data Modelling.R # Main modeling script ├── Accuracy_Assessment.R # Model evaluation ├── Visualisation.R # Enhanced visualizations ├── Opt-Vis1-DLNM_Curves.R # Response curve analysis ├── Opt-Vis2-Cumulative Effects.R # Cumulative effects ├── requirements.R # Package installation ├── WORKFLOW.md # Detailed workflow guide └── README.md # This file

📈 Results

Key Findings

• Temperature as dominant driver: Cumulative relative risk (RR) peaked at 1.9 (95% CI: 1.4-2.5) near 21°C
• Protective effects at extremes: RR < 1.0 for temperatures <10°C or >28°C
• Humidity effects: Modest, location-specific effects (RR 1.3-1.6 between 55-75%)
• Rainfall: Only sporadically associated with influenza risk
• Spatial patterns: Coherent coastal hot-spots identified through spatial pooling
• Lag structure: Effects manifest over 0-4 weeks post-exposure

Model Performance

| Model | Description | DIC | |-------|-------------|-----| | Model 1 | Case-crossover without spatial pooling | - | | Model 2 | Time-series without spatial pooling | - | | Model 3 | Case-crossover with spatial pooling | 153 (Best) | | Model 4 | Time-series with spatial pooling | - |

Practical Implications

• Spatial pooling removed implausible extremes in data-sparse western districts
• Framework enables early-warning dashboards for district-level surge prediction
• Temperature monitoring crucial for influenza preparedness in NSW

🛠️ Methodology

This implementation builds upon the Spatial Bayesian Distributed Lag Non-Linear Models (SB-DLNM) framework developed by Quijal-Zamorano et al. (2024), adapting it for influenza surveillance in NSW.

Statistical Framework

• Distributed Lag Non-Linear Models (DLNM): Cross-basis splines capturing non-linear and delayed effects (0-4 weeks)
• Bayesian Framework: Provides uncertainty quantification through credible intervals
• Spatial Component: Conditional autoregressive (CAR) priors enable neighboring LHDs to borrow strength
• MCMC Sampling: 10,000 iterations with 5,000 burn-in
• Model Selection: Deviance Information Criterion (DIC) for model comparison
• Temporal Resolution: Monthly aggregation of daily meteorological and case data

Model Specifications

1. Model 1: Independent B-DLNM with case-crossover design
2. Model 2: Independent B-DLNM with time-series design
   
3. Model 3: Spatially pooled DLNM with case-crossover design
4. Model 4: Spatially pooled DLNM with time-series design

Key Methodological Contributions

• Analysis of over 1.2 million laboratory-confirmed influenza cases across 24 years
• Integration of multiple meteorological factors with identification of temperature as dominant driver
• Spatial pooling to address data sparsity in western districts and identify coastal hot-spots
• Development of early-warning framework for district-level surge prediction
• Quantification of optimal temperature range (21°C) for influenza transmission in temperate Australia

🌟 Applications

This framework can be adapted for: - Public Health Surveillance: Real-time monitoring and early warning systems - Resource Planning: Hospital capacity and vaccine distribution optimization - Climate-Health Research: Understanding weather-disease relationships in other regions - Policy Development: Evidence-based interventions for influenza control - Forecasting Models: Integration with machine learning for enhanced prediction

📝 Citation

If you use this code or data in your research, please cite both our work and the original SB-DLNM methodology:

Our Implementation

bibtex @software{khan2024sbdlnm, title = {Spatially-informed Bayesian Distributed Lag Non-Linear Models for Influenza in NSW}, author = {Khan, Mohammad Afzal and Adegboye, Oyelola and Lyu, Shiyang and Adhinugraha, Kiki Maulana and Emeto, Theophilus I. and Taniar, David}, year = {2024}, url = {https://github.com/afzal0/SB-DLNM-Influenza_NSW}, version = {1.0.0} }

Original SB-DLNM Methodology

bibtex @article{quijal2024sbdlnm, title = {Spatial Bayesian distributed lag non-linear models (SB-DLNM) for small-area exposure-lag-response epidemiological modelling}, author = {Quijal-Zamorano, Marcos and Martinez-Beneito, Miguel A and Ballester, Joan and Marí-Dell'Olmo, Marc}, journal = {International Journal of Epidemiology}, volume = {53}, number = {3}, pages = {dyae061}, year = {2024}, doi = {10.1093/ije/dyae061} }

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

We welcome contributions! Please feel free to submit a Pull Request.

📞 Contact

For questions or collaborations: - Mohammad Afzal Khan: mkha0168@student.monash.edu - Repository Issues: GitHub Issues

🙏 Acknowledgments

• NSW Health for providing influenza surveillance data
• Australian Bureau of Meteorology for meteorological data
• Monash University for computational resources

Methodological Foundation

This work builds upon the Spatial Bayesian Distributed Lag Non-Linear Models (SB-DLNM) framework:

Quijal-Zamorano, M., Martinez-Beneito, M. A., Ballester, J., & Marí-Dell'Olmo, M. (2024). Spatial Bayesian distributed lag non-linear models (SB-DLNM) for small-area exposure-lag-response epidemiological modelling. International Journal of Epidemiology, 53(3), dyae061. https://doi.org/10.1093/ije/dyae061

We acknowledge and thank the authors for making their methodology available, which enabled this adaptation for influenza surveillance in NSW.