A Spatio-Temporal Modelling of Climate Variability on Respiratory Disease Morbidity in Kampala City, Uganda

Technical Summary: Spatio-Temporal Modelling of Climate Variability on Respiratory Disease Morbidity in Kampala City, Uganda

Problem Statement
Climate change and rapid urbanization pose significant public health risks, particularly in rapidly growing urban areas of sub-Saharan Africa like Kampala, Uganda. These factors exacerbate respiratory diseases such as asthma, chronic obstructive pulmonary disease (COPD), and pneumonia through complex interactions involving temperature, humidity, precipitation, and air quality. While global literature links climatic variability to respiratory outcomes, there is a scarcity of evidence regarding how these environmental exposures jointly influence the spatial and temporal distribution of respiratory diseases in Ugandan urban settings. Traditional statistical models often fail to capture the nonlinear, non-uniform, and spatio-temporal complexities inherent in climate-health relationships, necessitating more flexible analytical frameworks.

Methodology
The study employed a retrospective ecological design analyzing monthly aggregated data from 2021 to 2024 across the five administrative divisions of Kampala City (Central, Kawempe, Makindye, Nakawa, and Rubaga).

• Data Sources:

  • Health Data: Aggregated monthly counts of respiratory illnesses (asthma, COPD, pneumonia) were extracted from the District Health Information System 2 (DHIS2) managed by the Ministry of Health.
  • Climate Data: Monthly averages for temperature, precipitation, and humidity were obtained via the Health Information System Program (HISP) Uganda, utilizing ERA5 and CHIRPS datasets.
  • Air Quality Data: Particulate matter (PM2.5) concentrations were sourced from AirQo devices (Makerere University) and Kampala Capital City Authority (KCCA) monitoring networks.
  • Spatial Data: Administrative boundaries were obtained from the Uganda Bureau of Statistics (UBOS).

• Statistical Approach:
  The core analytical framework utilized Spatio-Temporal Generalized Additive Models (ST-GAMs) with a negative binomial distribution to account for overdispersion in disease counts. The models incorporated:

  • Smooth terms: Non-parametric functions for climate variables (temperature, humidity, precipitation, PM2.5) to capture nonlinear relationships.
  • Spatial smoothing: Two-dimensional smooth functions of longitude and latitude to model spatial variation.
  • Temporal smoothing: Smooth functions for time (month and year) to capture trends and seasonality.
  • Interaction terms: Tensor-product smooths to model spatio-temporal interactions (e.g., how seasonal patterns vary across locations).
  • Offset: An offset term based on the estimated monthly population size for each division to adjust for population risk.

  Model selection was guided by the Akaike Information Criterion (AIC), deviance explained, and Generalized Cross-Validation (GCV). Model diagnostics included randomised quantile residuals and checks for concurvity using the mgcv package in R.

Key Results

• Descriptive Statistics: A total of 89,253 respiratory illnesses were recorded. Pneumonia constituted the majority (65%), followed by asthma (23%) and COPD (12%). PM2.5 levels averaged 27.80 µg/m³, with significant seasonal peaks in January and February.
• Disease-Specific Drivers:
  • Asthma: Showed a significant, nonlinear association with PM2.5 (edf ≈ 4.5, p < 0.05) and precipitation. Cases increased markedly at PM2.5 concentrations above 30 µg/m³ and when precipitation exceeded 3 mm. Asthma exhibited significant spatial clustering in the Central and Nakawa divisions and seasonal peaks between July and October.
  • Pneumonia: Demonstrated a significant nonlinear relationship with precipitation (edf ≈ 2.1, p < 0.05), with cases rising rapidly after 3 mm of rainfall. Spatial clustering was observed in the Central and Kawempe divisions. Seasonal peaks occurred between March–May and September–November.
  • COPD: Unlike asthma and pneumonia, COPD patterns were not significantly driven by the specific climatic variables tested in the final model. Instead, COPD morbidity was strongly influenced by spatial location and temporal trends. The best-fitting model for COPD excluded climatic and air quality variables, retaining only spatial and temporal components, suggesting that entrenched local exposures (e.g., household biomass use, unregulated urban emissions) may be more dominant drivers. COPD cases peaked seasonally between May and August, with a steady increase in burden from 2022 onwards.
• Spatio-Temporal Patterns: All three diseases exhibited significant spatial heterogeneity. The interaction between month and year was significant for asthma and pneumonia, indicating interannual variability in seasonal peaks. However, the interaction between seasonality and space was generally not significant, suggesting that seasonal patterns remained relatively consistent across the different divisions of Kampala.

Key Contributions

1. Methodological Application: The study successfully applied ST-GAMs with tensor-product interactions to model climate-health relationships in an African urban context, demonstrating the utility of these flexible models in capturing nonlinear and spatio-temporal complexities that traditional linear models miss.
2. Disease-Specific Insights: The research distinguishes the environmental drivers of different respiratory conditions, identifying PM2.5 as a critical modifiable risk factor for asthma and precipitation for pneumonia, while highlighting that COPD in this context is driven more by spatial and temporal structural factors than by the specific meteorological variables analyzed.
3. Spatial Granularity: By mapping disease burdens to specific administrative divisions, the study identifies persistent hotspots (Central, Kawempe, Makindye, Nakawa), providing granular evidence for targeted public health interventions.

Significance and Claims
The paper claims that its findings provide empirical support for the use of flexible spatio-temporal modelling frameworks in environmental epidemiology within low- and middle-income settings. The study concludes that PM2.5 is a key modifiable risk factor for respiratory diseases in Kampala, particularly for asthma. It emphasizes that environmental exposures affect disease patterns differently depending on the disease type, location, and season.

The authors assert that the identification of persistent spatial clusters and specific seasonal peaks supports the need for precision public health interventions. They argue that resources should be allocated to high-risk divisions (Central, Kawempe, Makindye, Nakawa) and that interventions should be tailored to the specific environmental drivers of each disease (e.g., air quality improvements for asthma, sanitation and flood management for pneumonia). The study positions itself as a step toward climate-resilient health strategies in rapidly urbanizing African cities, though it acknowledges limitations regarding the use of aggregated data and the exclusion of other pollutants like NO₂ and O₃.