Temperature-Mortality → Life Expectancy & Lifespan Inequality

A pipeline that computes temperature-attributable mortality from climate projections and decomposes its impact on life expectancy (LE) and lifespan inequality (LI) by age and temperature range.

Pipeline

Notebook	Step
notebook/LI1_AN.Rmd	Attributable numbers (ANs) by age group and temperature range
notebook/LI2_disaggregate.Rmd	PCLM disaggregation of ANs from wide groups to single ages
notebook/LI3_analysis.Rmd	Combine ANs with population and all-cause mortality; period life tables
notebook/LI4_decomposition.Rmd	Decompose ΔLE and ΔLI by age and temperature range

Quick start

# 1. Install dependencies
install.packages(c("data.table", "arrow", "dlnm", "splines",
                   "ggplot2", "scales", "ungroup", "MASS", "eurostat"))

# 2. Render all notebooks (data is downloaded on the fly via Eurostat API)
notebooks <- c("LI1_AN", "LI2_disaggregate", "LI3_analysis", "LI4_decomposition")
for (nb in notebooks) {
  rmarkdown::render(file.path("notebook", paste0(nb, ".Rmd")),
                    output_dir = "results/demo",
                    knit_root_dir = ".")
}

Configuration

Each notebook has a config block at the top. Change city, city_code, ssp_label, and demo_gcm to run for a different European city or scenario. All outputs are written to results/demo/ with filenames derived from the city name.

Production pipeline (multi-city, multi-GCM, multi-SSP)

After validating with the notebooks, run the full pipeline for all 854 cities:

Rscript scripts/00_RunAll.R

This uses a cascading source chain (Masselot-style) and parallel foreach loops over cities. See scripts/ for the individual numbered scripts.

Method

Implements the standard attributable-risk framework (Gasparrini & Leone 2014, Masselot et al. 2023) using city-specific exposure-response functions from the MCC study, daily CMIP6 temperature projections, and EUROPOP2019 mortality projections. Life tables follow standard demographic methods. Lifespan inequality is measured via the standard deviation of age at death. Decomposition uses a stepwise replacement approach attributable by age and cause.

Data sources

Files committed to git (in data/)

File	Source	Description
coefs.csv	Masselot et al. 2023 — Zenodo	B-spline coefficients (b1–b5) per city × age group for reconstructing ERFs
vcov.csv	Same Zenodo record	Lower-triangle 5×5 variance–covariance per city × age group
city_results.csv	Same Zenodo record	City metadata, population, deaths, MMT, MMP, RR, historical attributable fractions

Files that must be downloaded separately (in data/, ignored by git)

File	Size	Source	Description
tmeanproj.gz.parquet	3.2 GB	ISIMIP3b CMIP6 (see Masselot et al. 2023 data notice)	Daily mean temperature for 854 cities, 21 GCMs, 3 SSPs, 1990–2099
coef_simu.csv	470 MB	Same Zenodo record as coefs.csv	1000 simulated coefficient vectors per city × age group for empirical CIs

Data downloaded on the fly via Eurostat API (no file needed)

Dataset	API code	Source	Description
Population projections	proj_19np	Eurostat EUROPOP2019	Single-age population by sex, country, year (2019–2100). Used to derive mortality improvement trends.
Life tables	demo_mlifetable	Eurostat	Historical age-specific death rates (DEATHRATE) by sex and country (1960–2024). Used as baseline mx.

The notebooks call load_eurostat_mortality(country_code, sex) in R/load_data.R, which downloads both datasets via get_eurostat() and combines them into a projected mx time series for any European country and gender.

Repository structure

.
├── data/                          # Input data (see table above)
│   ├── coefs.csv
│   ├── vcov.csv
│   ├── city_results.csv
│   ├── tmeanproj.gz.parquet       (download)
│   └── coef_simu.csv              (download)
├── notebook/                      # Validation notebooks (R Markdown)
│   ├── LI1_AN.Rmd
│   ├── LI2_disaggregate.Rmd
│   ├── LI3_analysis.Rmd
│   └── LI4_decomposition.Rmd
├── scripts/                       # Production pipeline (R scripts, Masselot-style)
│   ├── 00_Packages_Parameters.R
│   ├── 01_PrepData.R
│   ├── 02_ComputeAN.R
│   ├── 03_Disaggregate.R
│   ├── 04_AnalysisDataset.R
│   ├── 05_LifeTables.R
│   ├── 06_Decomposition.R
│   └── 00_RunAll.R               # Cascading master
├── R/                             # Shared helper functions
│   ├── rr_basis.R
│   ├── impact.R
│   ├── simulation.R
│   ├── load_data.R
│   ├── load_coefficients.R
│   ├── period_lifetable.R
│   ├── cohort_lifetable.R
│   ├── epv.R
│   ├── isimip3.R
│   └── utils.R
├── references/                    # Original reference code (Gasparrini, Masselot)
├── results/                       # Outputs (gitignored, except .gitkeep)
└── README.md

References

• Gasparrini A, Leone M. "Attributable risk from distributed lag models." BMC Medical Research Methodology 14:55, 2014. DOI: 10.1186/1471-2288-14-55
• Masselot P et al. "Excess mortality attributed to heat and cold: a health impact assessment study in 854 cities in Europe." The Lancet Planetary Health 7(4):e271–e281, 2023. DOI: 10.1016/S2542-5196(23)00023-2
• Rizzi S, Gampe J, van der Gaag N. "An estimator for the pairwise score function." Demographic Research 32:625–656, 2015. DOI: 10.4054/DemRes.2015.32.21