Mortality counts, population, temperatures, regions, and capacity values are synthetic. They are not observed Belgian mortality, Belgian temperatures, official population estimates, or measured health-system capacity. Coefficients are assumptions or fits to the synthetic data. The study does not show that raising capacity would reduce heat-related deaths.
Why this question
Heat and mortality are studied together often. A harder question for health-services research is whether the services around a population change that relationship, and whether the change is the same for every age group. This prototype asks that question inside a simulation, so the data-generating process is known and the limits of the model can be inspected.
Design
| Component | Specification |
|---|---|
| Period | 1 January 2019 to 31 December 2025 |
| Regions | 3 simulated regions |
| Age groups | 6 |
| Population | Hypothetical total of 1,000,000 |
| Records | 46,026 region–age–day observations |
| Seed | 20261016 |
| Mortality | Poisson simulation |
| Main model | Illustrative Poisson regression |
Temperature follows a synthetic seasonal process. Baseline mortality and heat vulnerability differ by age. Regions are given hypothetical capacity values. Two heat definitions sit side by side and should not be collapsed: the mortality equation treats heat as exposure above 20°C, while the weather summary marks a heat day above 25°C.
What the scripts examine
simulation_pipeline_publication.py is the entry point for the dataset, model output, summaries, and figures. Other scripts check properties of the generated data, compare heat specifications, try to recover parameters across repeated simulations, score prediction on a held-out synthetic set, vary the heat threshold, and audit whether regional capacity can be separated from within-region variation. Interaction scripts explore whether an interaction can be recovered under the scenarios that were specified.
The repository is explicit that a clean-environment installation and a full run of every script are still required before a complete reproducibility claim. Individual scripts can depend on files produced earlier.
Limits that belong in the reading
Associations in synthetic data follow the equation that generated them. Prediction scores on held-out synthetic days do not describe accuracy for observed mortality. A later empirical analysis would be a different project, with its own permissions and data agreement. Private research files are kept out of the public repository.
Repository: climate-demography-synthetic-prototype.