Baseline BMI Category and Incident Treated Hypertension in Australian Men

Analysis code for a prospective cohort study of the association between baseline body mass index (BMI) category and incident treated hypertension in Australian men, using the Ten to Men Australian Longitudinal Study on Male Health linked with Pharmaceutical Benefits Scheme (PBS) and Medicare Benefits Schedule (MBS) records.

The pipeline estimates the BMI–hypertension association with progressive confounder adjustment, quantifies incremental discrimination and clinical utility, and stress-tests the findings against detection bias, missing data, latency, effect modification by age, and competing risks.

---

Summary of the analysis

• Design: prospective cohort, 11-year follow-up (baseline 1 Jan 2014, censor 28 Feb 2025)
• Primary cohort: 12,742 men aged ≥18, free of treated hypertension at baseline, with a valid Wave 1 BMI category
• Exposure: baseline BMI category (Underweight, Normal [reference], Overweight, Obese)
• Outcome: incident treated hypertension (first PBS dispensing of an antihypertensive, ATC C02/C03/C07/C08/C09)
• Primary model: Cox proportional hazards with progressive adjustment
• Key secondary analyses: bootstrap C-index, decision curve analysis, calibration, multiple imputation, E-values, BMI × age interaction, competing risks

Headline results

BMI category	Fully adjusted HR (95% CI)	E-value (CI bound)
Underweight	1.43 (1.16–1.77)	2.21 (1.59)
Overweight	1.42 (1.22–1.67)	2.19 (1.74)
Obese	1.88 (1.59–2.22)	3.17 (2.56)

• Incremental discrimination of BMI over age alone: ΔC = 0.028 (95% CI 0.020–0.038)
• Fully adjusted model C-index: 0.754 (95% CI 0.740–0.769)
• BMI × age interaction: χ² = 12.79, df = 3, p = 0.005
• 10-year cumulative incidence: 7.5% (Normal) → 20.5% (Obese)

---

Repository structure

ttm-bmi-hypertension/
├── README.md                  # this file
├── LICENSE                    # MIT licence (code)
├── CITATION.cff               # how to cite this repository
├── renv.lock                  # pinned package versions (optional, see Reproducibility)
├── .gitignore
├── R/
│   ├── config.R               # central paths, constants, seed
│   ├── 00_run_all.R           # master runner (sources 01–13 in order)
│   ├── 01_data_prep.R         # cohort construction + outcome derivation
│   ├── 02_table1.R            # Table 1 descriptive statistics
│   ├── 03_cox_primary.R       # Cox models M1–M3 + Schoenfeld test
│   ├── 04_cindex_bootstrap.R  # C-index with bootstrap 95% CIs
│   ├── 05_calibration_dca.R   # calibration + DCA (point estimates)
│   ├── 06_figures.R           # Figures 1–3 (+ superseded 4,5)
│   ├── 07_sensitivity.R       # age strata, landmark, 90-day, combined outcome
│   ├── 08_multiple_imputation.R   # MICE (m=20) pooled Cox
│   ├── 09_interaction_evalue_cuminc.R  # interaction, E-values, cumulative incidence
│   ├── 10_schoenfeld_plot.R   # Figure 6 (Schoenfeld residuals)
│   ├── 11_dca_bootstrap.R     # Figure 4 (DCA with bootstrap ribbons)
│   ├── 12_competing_risks.R   # cause-specific Cox + Fine–Gray
│   └── 13_calibration_deciles.R   # Figure 5 (decile calibration)
├── data/
│   └── README.md              # data access + the UTF-8 conversion step
├── figures/                   # generated figures (git-ignored except .gitkeep)
└── manuscript/
    └── README.md              # pointers to the manuscript + figure mapping

---

Requirements

• R ≥ 4.3
• CRAN packages: dplyr, tidyr, readr, survival, rms, boot, ggplot2, survminer, cowplot, mice, cmprsk, scales

Install them with:

install.packages(c(
  "dplyr", "tidyr", "readr", "survival", "rms", "boot", "ggplot2",
  "survminer", "cowplot", "mice", "cmprsk", "scales"
))

(See R/install_packages.R for a one-shot installer.)

---

How to run

1. Obtain the data. This study uses linked Ten to Men / PBS / MBS data, which are not publicly distributable (see data/README.md and the Data Availability statement). Place your UTF-8 converted input file at:

   data/ttmcombined_utf8.csv
   

2. Set the working directory to the repository root (or open the project in RStudio at the root).

3. Run the whole pipeline:

   Rscript R/00_run_all.R

   or, interactively:

   source("R/00_run_all.R")

   Intermediates (.rds) are written to data/; figures (.png) to figures/.

You can also run any single script (after its prerequisites have been run once), e.g. Rscript R/04_cindex_bootstrap.R.

Approximate runtime

Most scripts run in seconds. The bootstrap steps dominate: 04_cindex_bootstrap.R (~1–2 min, B = 500) and 11_dca_bootstrap.R (~1 min, B = 150). The full pipeline completes in well under 10 minutes on a typical laptop.

---

Reproducibility

• A global seed (SEED = 2026) is set in R/config.R and applied before every stochastic step (bootstrap, multiple imputation, competing-risks imputation).
• All analysis constants (dates, horizon, ATC codes, bootstrap replicates) live in R/config.R — change them in one place.
• An renv.lock is provided for exact package-version pinning; run renv::restore() to reconstruct the environment. (Optional — the pipeline runs on any R ≥ 4.3 with the package versions listed above.)

---

Data availability

The Ten to Men data are managed by the Australian Institute of Family Studies and are available to researchers under a data access agreement. PBS/MBS linkage data are provided under separate approvals. None of these data are included in this repository. See data/README.md for the access process and the one-line encoding conversion required before running the pipeline.

---

Citation

If you use this code, please cite the associated manuscript and this repository. See CITATION.cff.

Licence

Code is released under the MIT Licence (LICENSE). The licence applies to the analysis code only, not to the underlying data.