UserEmb_Explainable

Model Diagram

Runtime Platform

• OS: Ubuntu 20.04
• Python 3.8
• Java JDK 1.8
• CUDA 11
• PyTorch 1.7.1+cu110

Data Acquisite

In this work, we have used two clinical narrative data, diabestes and MIMIC-III. Both datasets require you to take a list of training courses before acquiring the dataset. This will ensure you follow basic ethic rules to use the datasets. The datasets should be used for research purposes only.

• Diabetes
  • This dataset contains a list of candidates for clinical corhort selection. Each user/patient has multiple clinical narratives.
  • Each user/patient suffers diabetes.
  • Download 2018 (Track 1) Clinical Trial Cohort Selection Challenge: https://portal.dbmi.hms.harvard.edu/projects/n2c2-nlp/
• MIMIC-III
  • This dataset contains ICU records from a hospital.
  • Each user per visit will generate a list of clincal records.
  • Download data, MIMIC-III Clinical Database: https://physionet.org/content/mimiciii/1.4/

Prerequisite

• Download and follow instructions in the MetaMap to setup MetaMap.
  • Follow the instructions to download both data (dictionaries) and metamap toolkit.
  • If you have not installed Java, please install and following the JAVA link.
  • Note that our code supports both MetaMap and MetaMap-lite (faster version), but you have to change by yourself in data_builder.py.
• Python MetaMap Interface
  • You will install PyMetaMap toolkit.
  • Basically, the toolkit is a subprocess communicator between your Python script and Java files.
  • But our data_builder.py has its built-in supports, but you have to change codes by yourself using metamap_concepts or metamaplite_concepts.
• Other dependencies
  • Install PyTorch

Data Analysis

Data analysis scripts will be under the folder of data. We probe data via two types of analysis aspects, records and concepts.

Diabetes Analysis	Mimic-III Analysis

Experimental Steps

1. Data Preprocessing
   • Be sure to change data directory before running:
     • python data_builder.py;
   • For the mimic-iii data, you have to change the mimic-iii to your own path;
   • For the diabetes data, first, you need to put both training and testing cases into one directory, in our case, we named the directory as all;
   • Then go to the data directory and run the following lines:
     • python data_analyzer.py;
     • python concept_analyzer.py;
     • The two scripts will generate necessary data stats and concept vocabularies for training models.
2. Baselines
   • All baseline models will be under the directory of ./baselines/;
   • Run python any_baseline_script.py data_name will start to train user embeddings:
     • dp2user: python deep_patient2016.py diabetes or python deep_patient2016.py mimic-iii;
     • word2user: python word2user.py diabetes or python word2user.py mimic-iii;
3. Our approach
   • Run the following script will create entity augmented user representations;
     • We provide a list of running commands in shell scripts;
       • Go to the jobs folder: cd ./jobs;
       • To train CAUE_GRU for diabetes: sh run_gru_diabetes.sh;
       • To train CAUE_GRU for MIMIC-III: sh run_gru_mimic.sh.

Contact

Xiaolei Huang, [email protected]

Citation

@InProceedings{pmlr-v174-huang22a,
  title = {Enriching Unsupervised User Embedding via Medical Concepts},
  author = {Huang, Xiaolei and Dernoncourt, Franck and Dredze, Mark},
  booktitle = {Proceedings of the Conference on Health, Inference, and Learning},
  pages = {63--78},
  year = {2022},
  editor = {Flores, Gerardo and Chen, George H and Pollard, Tom and Ho, Joyce C and Naumann, Tristan},
  volume = {174},
  series = {Proceedings of Machine Learning Research},
  month = {07--08 Apr},
  publisher =    {PMLR},
  pdf = {https://proceedings.mlr.press/v174/huang22a/huang22a.pdf},
  url = {https://proceedings.mlr.press/v174/huang22a.html},
  abstract = {Clinical notes in Electronic Health Records (EHR) present rich documented information of patients to inference phenotype for disease diagnosis and study patient characteristics for cohort selection. Unsupervised user embedding aims to encode patients into fixed-length vectors without human supervisions. Medical concepts extracted from the clinical notes contain rich connections between patients and their clinical categories. However, existing \textit{unsupervised} approaches of user embeddings from clinical notes do not explicitly incorporate medical concepts. In this study, we propose a concept-aware unsupervised user embedding that jointly leverages text documents and medical concepts from two clinical corpora, MIMIC-III and Diabetes. We evaluate user embeddings on both extrinsic and intrinsic tasks, including phenotype classification, in-hospital mortality prediction, patient retrieval, and patient relatedness. Experiments on the two clinical corpora show our approach exceeds unsupervised baselines, and incorporating medical concepts can significantly improve the baseline performance.}
}