cmdpredict_bayesian

Bayesian prediction models for cardiometabolic disease (CMD) using data from the SCAPIS cohort. Current pipeline version: v07.

---

Folder Structure

cmdpredict_bayesian/
├── Main analysis/     # Steps 1-5: the primary pipeline
├── Stepwise/          # Nested/domain models (m1-m9)
├── Interaction/       # Interaction-term models (5 pairs)
└── Stan/              # Stan model files

Start in Main analysis/. Stepwise/ and Interaction/ reuse the data and fold indices created by Main analysis/ Step 1, so run that first. See the README in each subfolder.

---

File Naming Convention

cmdpredict_{step}_logit({any|multi})[_{variant}]_v{version}.R

{step} — position in the analysis pipeline:

Step	Meaning
1	Pre-processing and 10-fold CV split
2	Model fitting across folds
3	Results compilation
4	Full-sample fitting and individual predictions
5	Subgroup analysis

logit({any|multi}) — outcome:

Tag	Outcome
logit(any)	cmd_any — any cardiometabolic disease (no CMD vs any CMD)
logit(multi)	cmd_multi — cardiometabolic multimorbidity (no CMD multimorbidity vs CMD multimorbidity)

{variant} — present only in the Stepwise/ and Interaction/ folders:

Tag	Meaning
m1-m9	Nested/domain model set (see Stepwise/)
interaction / interactions	Interaction-term models (see Interaction/)

Other tags appearing in script and output file names:

Tag	Meaning
cov	Complete-case analysis
lasso	Regularised model
cvresult	Cross-validated fold results

Examples:

• Main analysis/cmdpredict_2_logit(any)_v07.R — Step 2, cmd_any
• Stepwise/cmdpredict_2_logit(multi)_m1-m9_v07.R — Step 2, cmd_multi, models m1–m9
• Interaction/cmdpredict_3_logit(any)_interactions_v07.R — Step 3, cmd_any, interaction models