Spatiotemporal Dataset and Reproducible Pipeline for Adolescent Suicide Mortality in Colombia (2015–2024)
Description
Version 4 provides the data, code, and computational outputs for a nationwide ecological study of registered adolescent suicide mortality in Colombia, 2015–2024. The balanced analytical panel contains 1,320 department–year–sex–age cells: 33 territorial units (32 departments and Bogotá, D.C.), ten years, two recorded-sex categories, and two age groups (10–14 and 15–19 years). It includes 4,234 deaths after three documented exclusions from a frozen historical input containing 4,237 INMLCF records. Zero-death cells are retained. Mortality records originate from the Instituto Nacional de Medicina Legal y Ciencias Forenses (INMLCF). Population denominators and Multidimensional Poverty Index (MPI) inputs originate from DANE. Numerators represent department of occurrence, whereas denominators represent projected resident populations. Departmental rates therefore describe occurrence-based ecological mortality, and MPI associations should not be interpreted as individual-level or causal effects. The repository includes frozen INMLCF inputs, harmonized data, analytical adjacency graphs, R-INLA scripts, posterior draws and summaries, model diagnostics, sensitivity analyses, supplementary tables S1–S30, figures, and audit records. Additional analyses address MPI measurement uncertainty, within- and between-department associations, posterior predictive calibration, blocked cross-validation and stacking, insular connectivity, archipelago influence, aggregate DANE EEVV triangulation, and simultaneous posterior exceedance. The historical mortality input remains the basis of the reported results. A later official export is retained solely to document source-vintage differences. DANE EEVV individual records and cartographic vector layers are not redistributed. Their source references, file hashes, and retrieval/restoration procedures are provided. The deposited results can be validated without these external inputs; complete source-to-figure reconstruction requires their authorized local acquisition. V4 updates release documentation, citation and rights information, dependency records, file-integrity checks, and Figure S1 export metadata while preserving the primary analytical panel and model results. It succeeds V3 (doi: 10.17632/4b3fmpdn93.3). Original code and documentation retain CC BY 4.0. INMLCF source material and adapted mortality data retain CC BY-SA 4.0. Other third-party materials retain their applicable provider rights, as detailed in LICENSE.md and RIGHTS.md.
Files
Steps to reproduce
1. Use the extracted data_analysis directory as the working directory. Preserve an untouched copy of the deposit because rerunning stages updates generated files and audit reports. 2. Prepare the computational environment. The reported analysis used R 4.6.0 and INLA 26.06.08, displayed by packageVersion() as 26.6.8. Consult 00_documentation/SOFTWARE_ENVIRONMENT.md, SOFTWARE_PACKAGES.csv, and renv.lock. Restore the recorded R packages with renv::restore(lockfile = "renv.lock") and make the restored library available to Rscript through R_LIBS_USER. The exact INLA archive URL is documented. Spatial packages also require compatible system libraries. 3. Verify the deposited files before regenerating outputs: Rscript --vanilla 02_code/validation/98_verify_release_files.R This checks file sizes and SHA-256 hashes against FILE_MANIFEST.csv. 4. Validate the included results and inspect execution prerequisites: Rscript --vanilla 02_code/validation/99_validate_reproducibility.R Rscript --vanilla run_workflow.R --profile=preflight Stage 99 checks the frozen input, analytical panel, tables, workbook, figure-source relationships, and computational lineage. Preflight does not execute scientific stages. INFO entries identify intentionally external inputs. 5. For DANE triangulation, acquire and verify the external EEVV sources: python3 02_code/data_preparation/00_download_dane_eevv.py python3 02_code/data_preparation/00_download_dane_eevv.py --verify-only Python 3.8+ is required. The script retrieves official archives and verifies their extracted CSVs against the annual manifest. 6. For spatial-input reconstruction and redrawing Figures 2–3, obtain authorized local copies of the historical cartographic vectors identified in 00_documentation/EXTERNAL_FILES.csv. Follow EXTERNAL_DATA.md to restore and verify them. The included panel, MPI inputs, and adjacency graph support primary-model fitting without these vectors. 7. After restoring all external prerequisites, execute the complete pipeline: Rscript --vanilla run_workflow.R --profile=full The runner executes preparation, models, posterior summaries, sensitivity analyses, tables, figures, and final validation in order. Full fitting and blocked cross-validation are computationally intensive. Available subsets are listed with: Rscript --vanilla run_workflow.R --list-profiles 8. Review outputs in directories 03–06 and validation reports in 07_audit_records/validation. The frozen panel must contain 1,320 cells and 4,234 deaths, and Stage 99 must report PASS. Detailed procedures and verification limits are documented in REPRODUCIBILITY.md and RELEASE_VALIDATION.md.