MIMIC Code Repository

Extract patient vitals using 'vitalsign.sql'.

Calculate hourly 'SOFA' scores using the concepts directory.

Label sepsis onset using the 'sepsis3' definitions provided in the repo.

Train a Random Forest or LSTM model on the resulting time-series data.

Extract de-identified clinical notes from the 'note' table.

Use the 'preprocess_notes.py' script to clean headers and timestamps.

Tokenize and prepare datasets for instruction tuning.

Fine-tune models like Llama-3 or Mistral on domain-specific medical summaries.

Query the 'icustays' table for historical occupancy rates.

Join with 'admissions' to identify seasonal trends.

Apply the 'length_of_stay' prediction script.

Generate a dashboard showing predicted vs. actual bed availability.

Use the 'prescriptions' table to isolate antibiotic orders.

Link with 'microbiologyevents' to check for culture results.

Run the 'antibiotic_timing' script to assess delay from infection onset.

Identify patterns of inappropriate prescribing.

Generate the 'mortality' outcome labels for a specific cohort.

Calculate the 'OASIS' score as a baseline predictor.

Run the comparative analysis scripts in R.

Evaluate the AUC-ROC of the new model against the OASIS baseline.

Extract respiratory parameters using 'ventilation.sql'.

Map patient IDs to the 'mimic-cxr' image database.

Cluster patients based on both structured data and image features.

Validate clusters against 30-day mortality outcomes.

Use the repository's schema to define the target data structure.

Train a GAN or Diffusion model on the real MIMIC distributions.

Generate synthetic records that maintain the relational integrity defined in the SQL scripts.

Validate synthetic data against the repo's 'data_quality' checks.