Predicting Catecholamine Therapy Initiation

Repository with datasets and models from our paper: "Towards actionable hypotension prediction: Predicting catecholamine therapy initiation in the intensive care unit".

We demonstrate the feasibility of predicting catecholamine therapy initiation using MAP measurements and additional clinical variables with robust overall model performance.

Setup

1. Clone this repository

   git clone https://github.com/AI4HealthUOL/actionable-hypotension.git
   cd actionable-hypotension

2. Install dependencies

   pip

   pip install -r requirements.txt

   Conda (managed environment) → alternative, don’t install both

   conda env create -f environment.yml
   conda activate actionable-hypotension

3. Install project modules (editable / dev mode)

   pip install -e .

4. Set up MIMIC-III in Postgres
   Follow the official [MIT-LCP buildmimic/postgres instructions](https:/ github.com/MIT-LCP/mimic-code/tree/main/mimic-iii/buildmimic/postgres).

   These instructions also cover how to download the MIMIC-III CSVs from PhysioNet and load them into Postgres.

5. Configure database connection
   Copy the env template .env.example → .env and set exactly one DATABASE_URI (no quotes):

   cp .env.example .env

   Example:

   DATABASE_URI=postgresql+psycopg2://USERNAME:PASSWORD@HOST:PORT/DBNAME
   

Usage

• To try pre-trained models → run 1 and 3.
• To retrain models → run 1, 2, 3.

1. Prepare Datasets

Run the SQL files in order:

• patient_metadata

  • all_mv_icu_stays.sql
  • chartevents_metadata.sql
  • mv_metadata_binned.sql
  • mv_drug_item_map.sql

• context_windows

  • linkorder_treatment_events.sql
  • all_mv_context_target_windows.sql
  • all_mv_labeled_windows.sql
  • additional_meds_in_windows.sql

• map_values

  • all_mv_labeled_windows_with_map_values.sql
  • invasive_noninvasive_mix_windows.sql
  • <datasetname>_statistical_features.sql

• split_data

  • split_all_subjects.sql

• merge_final_datasets

  • dataset_inv_ninv_mix.sql

2. Train and Calibrate Models (skip if using pretrained models)

• XGBoost:

  • Train: training/train_xgboost_models.ipynb
  • Calibrate: calibration/calibrate_xgboost_models.ipynb (set model name in marked cell)

• Baseline:

  • Train & Calibrate: baseline_model/train_and_calibrate_baseline_model.ipynb

3. Evaluate Models

In each script you need to set the model name in the marked cell.

• Performance plots: ce_approach/evaluation/evaluate_models.ipynb

• DeepROC:

  • Precompute: ce_approach/evaluation/deeproc_cli.py (⚠ long runtime)
  • Plot: ce_approach/evaluation/deeproc_plot.ipynb

• Metric tables: ce_approach/evaluation/calculate_tables.ipynb

• Subgroup analysis:

  • Precompute: subanalysis/notebooks/create_test_full.ipynb
  • Create Table: subanalysis/notebooks/subgroup_analysis.ipynb