Public repository containing research code for the circEWS project accompanying the manuscript Early prediction of circulatory failure in the intensive care unit using machine learning

When using code from this repository, please consider citing

Hyland, S.L. et al. Early prediction of circulatory failure in the intensive care unit using machine learning. Nat Med (2020). https://doi.org/10.1038/s41591-020-0789-4.

The code is organized in several sub-directories, which contain the following content:

• binarize Binarize time-grid data to only keep measurement patterns.

• calibration Calibration analysis of continuous risk scores of circEWS.

• circews Classes and utility functions.

• circulatory_status Annotation of time series with status of stability or stages of circulatory failure.

• dimensionality_reduction Merging of raw HIRID variables corresponding to identical clinical concepts into meta-variables.

• evaluation Evaluation of alarm system performance.

• external_validation Code for external validation on the MIMIC data-set.

• features
  Contains code for generation of non-shapelet features from imputed data.

• finetuning
  Interpolation of MIMIC/HIRID based models to fine-tune circEWS towards the MIMIC database.

• imputation
  Code concerned with transforming HIRID data to a fixed time grid, making it suitable for feature generation and fitting of machine learning models.

• labels
  Code for creating labels where positive labels correspond to time points where it is desirable to raise an alarm, located in the 8 hours prior to circulatory failure events.

• learning
  Supervised learning scripts for learning a continuous risk score for predicting circulatory failure.

• lstm
  LSTM model implementation.

• pipeline_diagnostics
  Diagnostic code for tracking PIDs in different pipeline stages, and others.

• preprocessing
  Code for preprocessing the HIRID data, including artifact deletion strategies and others.

• shapelet_features
  Code concerned with discovering and applying shapelet features on the HIRID data indicative of future circulatory failure.

• splits
  Code concerned with splitting PIDs for cluster processing and generating data splits for the experimental design.

• visualization
  Code concerned with visualizing patient stays, using data from different pipeline stages.