This project is for the paper "Identification of the risk of sepsis in cancer patients using digital health care records" in BMC Medical Informatics and Decision Making.
- Authors of source code: yangdonghun3@kisti.re.kr, jmkim@kisti.re.kr
- Current version of the project: ver. 0.1
- Objective: Sepsis is diagnosed in millions of people every year, resulting in high mortality rate. Although sepsis patients present multimorbid conditions, including cancers, sepsis predictions have mainly focused on patients with severe injuries. Here, we present a machine learning-based approach to identify sepsis risk in cancer patients using electronic health records (EHRs).
- Materials and Methods: We utilized anonymized EHRs from the Samsung Medical Center in Korea, including 8,580 cancer patient records in longitudinal manner (2014~2019). To build a prediction model based on physical status that would differ between sepsis and nonsepsis patients, we analyzed 2,462 laboratory (lab) test results and 2,266 medication prescriptions using a graph network and statistical analysis. Based on the results, the model was trained with lab tests and medication relationships.
- Results: Sepsis patients showed differential medication trajectories and physical status. For example, in the network-based analysis, narcotic analgesics were prescribed more often in the sepsis group, along with other drugs. Likewise, 35 types of lab test, including albumin, globulin, and prothrombin time, showed significantly different distributions between sepsis and nonsepsis patients (p value<0.0001). Our model outperformed the model trained using only common EHRs, showing improved accuracy, AUROC, and F1-scores up to 11.9%, 11.3%, and 13.6%, respectively. (Accuracy: 0.692, AUROC: 0.753, and F1-score: 0.602 for the random forest-based model).
- Discussion and Conclusion: We elucidated that lab tests and medication relationships can be used as efficient features for predicting sepsis in cancer patients. Consequently, identification of sepsis risk in cancer patients using EHRs and machine learning is feasible.
- Python 3.7
- Pytorch 1.5 (GPU version is recomended)
- scipy
- scikit-learn
- shap
- numpy
- pandas
- matplotlib
This project was supported by the Korea Institute of Science and Technology Information (KISTI) (K-21-L02-C10, K-20-L02-C10-S01). HP and JK were also supported by the Ministry of Science and ICT (N-21-NM-CA08-S01). This research was also supported by the Program of the National Research Foundation (NRF) funded by the Korean government (MSIT) (2021M3H9A203052011). The computational analysis was supported by the National Supercomputing Center, including the resources and technology.
- Use of source codes are free for academic researchers. However, the users of source codes from the private sector will need to contact to the developers of the project.
We present the source codes as an example of our research project to help readers understand. We don't provide data preprocessing code. For preprocessing, such as generation drug relationship features, please refer to our previous work. Due to personal information issues, we provide dummy data instead of the EHR information of cancer patients used in our experiments for testing the proposed sepsis prediction model.
We provide training and testing codes for two machin learning models and three deep learning models for sepsis prediction.
- Machine learning models: Logistic Regression and Random Forest
- Deep learning models: ANN, ResNet10, and RNN-LSTM
- Each training code (Sepsis_prediction_xxx.ipynb) includes our setup and experimental results.
- The trained models, provided in the "trained_model" directory, can be tested on dummy data using each testing code (Sepsis_prediction_xxx_Test_using_dummy_data.ipynb).
- In the SHAP_for_RF.ipynb file, important features, that mainly contributed to the proposed sepsis prediction model (Random Forest-based), are shown.
- Won Chul Cha. wc.cha@samsung.com
- Hyojung Paik. hyojungpaik@gmail.com