Repository release: v1.1.0
Computational checkpoint: v0.7.0
Author: Yuzhan Zhang
ORCID: 0009-0000-3121-7972
This repository contains the code, evidence tables, frozen simulation outputs, tests, and research-design analyses supporting the manuscript:
Evidence-Calibrated Simulation of Thromboprophylaxis After Femoral Shaft Fracture: Time-Varying Risk and Value of Information
The project treats thromboprophylaxis after femoral shaft fracture (FSF) as a time-varying thrombosis, observation, bleeding, mobility, treatment-effect, and research-design problem. It is a methodological/reproducibility resource, not a clinical decision tool.
The state-space simulation separates:
- occult/pre-existing thrombosis from new postoperative latent DVT;
- latent thrombosis from surveillance-dependent clinical ascertainment;
- thrombotic pulmonary embolism from other post-traumatic embolic syndromes;
- clinician-cleared assisted-upright mobility from unrestricted injured-limb weight bearing;
- natural-history calibration from uncertainty in prophylactic treatment efficacy;
- longitudinal phenotyping information from evidence that can identify a causal treatment-duration effect.
The decision layer compares fixed 7-, 14-, 21-, 28-, and 35-day policies using a provisional acute loss index and value-of-information analysis. Duration rankings are deliberately treated as model stress tests rather than treatment recommendations.
The final computational checkpoint is v0.7.0. Release v1.1.0 is a reporting, packaging, and journal-migration release; it does not change the model mechanism, calibrated parameters, or numerical results from the frozen scientific baseline.
The principal scientific conclusion is negative but informative: currently available aggregate FSF evidence does not identify a clinically interpretable individualized prophylaxis-stopping rule. Longitudinal phenotyping can reduce structural uncertainty, but observational phenotyping alone cannot identify the causal effect of prophylaxis duration.
src/dptfsf/ Python model, calibration, decision and EVSI modules
scripts/ Reproduction scripts for historical checkpoints
configs/ Frozen model configurations
tests/ Automated regression/structural tests
evidence/ Evidence matrices and parameter registries
results/ Frozen simulation/calibration/decision outputs
figures/ Publication-facing figures generated from frozen outputs
docs/ Core model, evidence, safety and adversarial-review documents
deferred/ Optional MIMIC-IV external-validation bridge; not used in the study
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
python -m pip install --upgrade pip
python -m pip install -e ".[test]"Python 3.10 or later is required.
Core commands:
python scripts/run_v0.4.0.py
python scripts/run_v0.7.0.py
pytest -qThe frozen v1.1.0 release passed 22/22 automated tests before archiving.
The calibration and decision outputs used by the final manuscript are retained under results/v0.4.0/ through results/v0.7.0/. Publication-facing figures are under figures/publication_v1.1.0/.
See REPRODUCIBILITY.md for the checkpoint map and the distinction between scientific and reporting versions.
The assisted-upright branch applies only after individual assessment and clearance by the treating clinician, with a stable perioperative course, adequate fixation for the permitted activity, and no mobilization contraindication. Assisted upright mobility is not equivalent to unrestricted weight bearing.
The model does not instruct any patient to start or stop anticoagulation, change dose, stand, ambulate, or bear weight. It has not undergone external or prospective clinical validation.
A MIMIC-IV feasibility bridge was prepared but deliberately deferred. No MIMIC-IV data were used in the manuscript or in the results contained here. The archived bridge is retained under deferred/ solely as a future research interface.
GitHub should expose a Cite this repository control from CITATION.cff. Zenodo/GitHub release metadata are pre-populated in .zenodo.json. After a Zenodo DOI is minted, the DOI and repository URL should be added to CITATION.cff in a small metadata-only follow-up release.
Until then, cite this software release as:
Zhang Y. Evidence-Calibrated Simulation of Thromboprophylaxis After Femoral Shaft Fracture: Reproducibility Code. Version 1.1.0. 2026.
Please also cite the associated manuscript once its bibliographic record is available.
Source code and original repository documentation are released under the MIT License. Third-party publications cited in the evidence tables remain the property of their respective rights holders and are not redistributed here.