ICU Mortality Risk Prediction Model

Disclaimer: Not for clinical use — educational purposes only.

A prognostic model for ICU mortality risk prediction using the eICU Collaborative Research Database. This report compares traditional logistic regression with LASSO penalized regression to predict in-hospital mortality for ICU patients. The final model prioritizes interpretability over algorithmic complexity.

Project Structure

icu_mortality/
├── scripts/                    # Analysis workflow
│   ├── 01_data_load_and_connect.R    
│   ├── 02_clean_data.R               
│   ├── 03_feature_selection.R        
│   ├── 04_run_glm_model.R            
│   ├── 05_run_lasso_model.R         
│   ├── 06_compare_models.R           
│   ├── 07_validate_model.R           
│   ├── Appendix.R                    
│   └── sql/                          
├── data/                    # Data                    
│   └── README.md        
├── docs/                                          
│   ├── annotated_bibliography.md     
│   ├── images/                       
└── renv/                       

Running the Analysis

1. Clone and setup:

   # Restore package environment
   renv::restore()

2. Execute analysis pipeline:

   # Run scripts in order:
   source("scripts/01_data_load_and_connect.R")
   source("scripts/02_clean_data.R")
   source("scripts/03_feature_selection.R")
   source("scripts/04_run_glm_model.R")
   source("scripts/05_run_lasso_model.R")
   source("scripts/06_compare_models.R")
   source("scripts/07_validate_model.R")

References

This project follows established clinical prediction modeling guidelines and draws from key literature in ICU mortality prediction. See docs/annotated_bibliography.md for detailed references.

License

This project uses the eICU Collaborative Research Database under the PhysioNet Credentialed Health Data License 1.5.0. The analysis code is available for research and educational purposes.