CEA Pipeline

Cost-effectiveness analysis scaffolding for the lab — Markov, decision tree, PSA, CEAC, AMA-styled output. Drummond 2015 + CHEERS 2022 standards.

Active pilots

Reusable infrastructure for every CEA project in the Holubar Lab sandbox. Modeled after Leeds 2019/2022 methodology (PMIDs 31596763, 35277799), implemented in Python.

Cost-Effectiveness Analyses/
├── _shared/                   ← you are here
│   ├── Scripts/               ← reusable model code
│   ├── Inputs/                ← canonical parameter library w/ PMID provenance
│   ├── Templates/             ← memo + manuscript + notebook templates
│   ├── Docs/                  ← methodology + CHEERS checklist + this README
│   └── Validation/            ← reference-case validation artifacts
├── VTE_Prophylaxis_Apixaban_vs_Enoxaparin/   ← Pilot 1
└── IBD-PSC_Surgical_Strategy_CEA/            ← Pilot 2

Scripts (_shared/Scripts/)

All scripts include self-tests in __main__ blocks. Run:

cd "/Volumes/T1 Claude/000 Claude Sandbox/Active Research and Manuscripts/Cost-Effectiveness Analyses/_shared/Scripts"
python3 cea_utils.py        # reproduces Leeds 2022 ICER
python3 markov_engine.py    # reference-case Markov + time-varying check
python3 decision_tree.py    # Leeds-2022-style tree + PSA wrapper
python3 plotting.py         # generates 4 canonical figures (CEA plane, CEAC, tornado, Markov trace)
python3 reporters.py        # ICER text + .docx + CHEERS snippet

How to scaffold a new CEA project

1. Create the project folder with standard subfolder layout:

``bash PROJECT="MyNewCEA" cd "/Volumes/T1 Claude/000 Claude Sandbox/Active Research and Manuscripts/Cost-Effectiveness Analyses" mkdir -p "$PROJECT"/{Scripts,Data,Figures/drafts,Lit} find "$PROJECT" -type d -exec touch {}/z.txt \; ``

1. In Scripts/main_analysis.py, import from _shared/Scripts/ via

sys.path injection:

``python from pathlib import Path import sys sys.path.insert(0, str(Path(__file__).resolve().parents[2] / "_shared" / "Scripts")) from cea_utils import Strategy, icer_table, ceac_curve, evpi, sample_beta, sample_gamma from markov_engine import MarkovModel from decision_tree import DecisionNode, ChanceNode, TerminalNode, evaluate_psa from plotting import plot_cea_plane, plot_ceac, plot_tornado, plot_markov_trace from reporters import psa_summary, write_icer_table_docx, cheers_results_snippet ``

1. Load inputs from _shared/Inputs/:

``python import pandas as pd shared_inputs = Path(__file__).resolve().parents[2] / "_shared" / "Inputs" probs = pd.read_csv(shared_inputs / "probabilities.csv").set_index("parameter") costs = pd.read_csv(shared_inputs / "costs.csv").set_index("parameter") utils = pd.read_csv(shared_inputs / "utilities.csv").set_index("parameter") ``

1. Verify all parameters used are VERIFIED before running the analysis

that will go into the manuscript. TODO_* rows are allowed during model building but must be resolved before final results.

1. Build the model, run PSA, generate figures + tables, draft the

analytic memo. Each project's Data/ folder holds frozen inputs + results JSON; Figures/ holds draft PNGs + clean TIFFs.

1. Cascade: per project-structure.md, results JSON → notebook →

analytic memo → manuscript → figures. Use /pc to commit.

Validation requirements (per project)

Each project's Validation/ folder (or Scripts/validate.py) must include:

• Face validity — clinical-expert structure review note
• Internal validity — reproduction of a published reference case OR

hand-calculation match at known inputs

• External validity — comparison to independent observational data

when available

• Sensitivity to alternative model structures — if both decision tree

and Markov are plausible for the same question, run both and report agreement

Conventions enforced

• 3%/yr discount rate, half-cycle correction ON, $100K/QALY primary WTP,

10,000 PSA iterations (per _shared/Docs/METHODOLOGY.md)

• Every parameter value has a PMID source or is flagged TODO_*
• Every figure is TIFF 600 DPI LZW + PPTX editable copy
• Every table follows AMA three-line border style
• Every manuscript carries the CHEERS 2022 checklist (Supplementary File)
• Costs and QALYs both discounted by default
• No DOIs in references — PMIDs only

Reference papers (in Odin/PDF Articles/)

• Leeds 2019 — PMID 31596763 — "Cost-Benefit Limitations of Extended VTE

Prophylaxis After Surgery for Crohn's Disease" (DCR)

• Leeds 2022 — PMID 35277799 — "Cost-Effectiveness of Aspirin for Extended

VTE Prophylaxis After Major Surgery for IBD" (J Gastrointest Surg) — SDH co-author

• Neumann 2016 — PMID 27623463 — Second Panel on CEA in Health and Medicine
• Husereau 2022 — PMID 35017145 — CHEERS 2022 reporting guideline
• Briggs et al. — Decision Modelling for Health Economic Evaluation (2006)

— foundational textbook (not in corpus; see CCF library)