marinhodomingosm/post-stroke-aphasia-risk-analysis
🩺 Analyze post-stroke aphasia risks by investigating medication patterns and mental health impacts to improve patient outcomes and reduce hospital readmissions.
What's novel
🩺 Analyze post-stroke aphasia risks by investigating medication patterns and mental health impacts to improve patient outcomes and reduce hospital readmissions.
Code Analysis
11 files read · 4 roundsA clinical data analysis pipeline that builds a stroke+aphasia cohort from an OHDSI/OMOP Redshift database, engineers medication/mental-health features, and runs predictive (LR/Lasso/XGBoost+SHAP), multivariate statistical, and counterfactual clinical-impact analyses of PIM exposure and hospital rea
Strengths
Genuine analytical depth: the clinical_impact module implements a real counterfactual g-computation decomposition (differential PIM exposure vs. direct aphasia effect vs. interaction) with bootstrap CIs and marginalization, and the PIM polypharmacy logic computes concurrent-medication streaks over date ranges. The pipeline is well-organized with a clear Makefile, modular clinical_impact package, and reproducible seeded modeling.
Weaknesses
No tests at all, and the README is largely misleading — it markets a downloadable 'application' with a 'user-friendly interface' and install steps, but the code is a research script pipeline with no UI, and the download link points to a zip inside an Observable cache folder. There is also heavy code duplication (concurrent-med calc repeated for PIMs vs all meds), hardcoded OMOP schema/table names and a config module with plaintext DB credentials, and a performance-inefficient per-date loop for p
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
17
Contributors
3
Files
138
Active weeks
3
Repository
Language
Python
Stars
1
Forks
0
License
MIT