Out of variety of possible use cases from MIMIC dataset we focus on single use case to predict the number of days a patient stays in the Intensive Care Unit (ICU) to further generate synthetic data.
This usecase seemed important - it is a benefit for payer to predict the tendency of the length of stay of a patient. It helps in changing the premium charged by the payer according to the comparison of predictions and baseline (the defined no. of days covered by a particular plan of the patient). For this use case the model utilises the total number of diagnosis that occured for different disease category for each patient.
We used the default implementation of CTGAN which is available here.
https://github.com/sdv-dev/TGAN https://github.com/sdv-dev/CTGAN
To use CTGAN do a pip install. Also, we will be installing the table_evaluator library which will help us in comparing the results with the original data.
pip install ctgan
pip install tgan
pip install table_evaluator
You can use provided by us conda enviroment.yaml.
conda env create -f environment.yml
MIMIC-IV v1.0 contains deidentified data of 383,220 patients admitted to an intensive care unit (ICU) or the emergency department (ED) between 2008 - 2019. The latest version of MIMIC-IV is v0.4 and only provides public access to the electronic health record data of 50,048 patients admitted to the ICU, which is sourced from the clinical information system MetaVision at the BIDMC.
Researchers seeking to use the database must:
- Become a credentialed user on PhysioNet. This involves completion of a training course in human subjects research.
- Sign the data use agreement (DUA). Adherence to the terms of the DUA is paramount.
- Follow the tutorials for direct cloud access (recommended), or download the data locally.
Data is hard to preprocess - data cleaning procedure need to handle:
(1) Inconsistent units.
(2) Multiple recordings at the same time.
(3) Range of feature values. We use the median of the range as thevalue of the feature.
Predicting the length of stay in ICU can be based on 4 tables from MIMIC-IV database:
patients: Every unique patient in the database (defines subject_id) Columns like: 'subject_id', 'gender', 'anchor_age', 'anchor_year', 'anchor_year_group', 'dod' (count of rows: 382278, count of columns: 6)admissions: Every unique hospitalization for each patient in the database (defines hadm_id) Columns like: 'subject_id', 'hadm_id', 'admittime', 'dischtime', 'deathtime', 'admission_type', 'admission_location', 'discharge_location', 'insurance', 'language', 'marital_status', 'ethnicity', 'edregtime', 'edouttime', 'hospital_expire_flag' (count of rows: 523740, count of columns: 15)icustays: Every unique ICU stay in the database (defines icustay_id) Columns like: 'subject_id', 'hadm_id', 'stay_id', 'first_careunit', 'last_careunit', 'intime', 'outtime', 'los' (count of rows: 76540, count of columns: 8)diagnosis_ICD: Hospital assigned diagnoses, coded using the International Statistical Classification of Diseases and Related Health Problems (ICD) system Columns like: 'subject_id', 'hadm_id', 'seq_num', 'icd_code', 'icd_version' (count of rows: 5280351, count of columns: 5)
More about data preparation and EDA based on chosen features can be found HERE.
The Synthetic Data Vault (SDV) is a Synthetic Data Generation ecosystem of libraries that allows users to easily learn single-table, multi-table and time series datasets to later on generate new Synthetic Data that has the same format and statistical properties as the original dataset.
We tried with CTGAN architecture from the framework.
Package Reference: ctgan
Documentation : https://sdv-dev.github.io/CTGAN/
Github: https://github.com/sdv-dev/CTGAN
We trained the model for 100 epochs only as the discriminator and generator loss becomes quite low after these many epochs.
Then, we generated 100 000 rows of synthetic data and compare its distribiution with real.
Script to run experiments can be found HERE.
We did a feature by feature comparision between the generated data and the actual data. We used python’s table_evaluator library to compare the features.
We call the visual_evaluation method to compare the actual data(data) and the generated data(samples).
As its apparent from the visualizations, the similarity between the original data and the synthetic data is quite high. The results give a lot of confidence as we took a random dataset and applied the default implementation without any tweaks or any data preprocessing.
Notebook to run experiments can be found HERE.
Subsequently, we can use original train data and synthetic train data to predict case Length of Stay. This is regresion task, in which algorithms like Regression Tree, Random Forest, XGBoost, Support Vector Machine and K-Nearest Neighbor, can be used.
We tried with simple neural network:
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_15 (Dense) (None, 100) 4300
_________________________________________________________________
dense_16 (Dense) (None, 50) 5050
_________________________________________________________________
dense_17 (Dense) (None, 50) 2550
_________________________________________________________________
dense_18 (Dense) (None, 50) 2550
_________________________________________________________________
dense_19 (Dense) (None, 1) 51
=================================================================
Total params: 14,501
Trainable params: 14,501
Non-trainable params: 0
_________________________________________________________________
We compared metrics as: Mean Squared Error, Root Mean Squared Error, Mean Absolute Error, Mean Absolute Percentage Error.
| Dataset | RMSE | MSE | MAE |
|---|---|---|---|
| orginal | 4.05 | 16.41 | 2.44 |
| synthetic CTGAN | 7.05 | 49.75 | 5.04 |
Script to run experiments can be found HERE.
The project was developed during the first rotation of the Eye for AI Program at the AI Competence Center of Sahlgrenska University Hospital. Eye for AI initiative is a global program focused on bringing more international talents into the Swedish AI landscape.
Works presented here depends on Puskhar-v's repository.