Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/dokato/c2s2
https://github.com/dokato/c2s2
Last synced: 18 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/dokato/c2s2
- Owner: dokato
- Created: 2019-08-08T09:24:24.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2020-09-11T08:43:46.000Z (over 4 years ago)
- Last Synced: 2024-10-16T12:39:40.571Z (2 months ago)
- Language: Python
- Size: 98.6 KB
- Stars: 3
- Watchers: 4
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Medical Cohort Selection
Python scripts with a competition solution prepared for "2018 N2C2 Shared Tasks".
Data to train the models is available upon request at https://n2c2.dbmi.hms.harvard.edu/.
## Setup
You might need the following to run these scripts:
```
- python 2.7
- numpy
- sklearn
- xgboost
- conreader (optional)
```
You might also need the following data structure:
```
clitri/
test_gold/
models/
runsequence.sh
README.md
testoutput/
output/
preproc/
```
where `clitri/` is the folder with scripts for training, testing models; `test_gold/` - is folder with annotation of the test files; `models/` is the folder where you keep trained ML models and vectorisers; `output/` output of training data (eg. for crossvalidation); `testoutput/` output of prediction on test set.
**!!!** You may change all paths to the files in `clitri/utils.py`.
## Running the preprocessing (feature extraction)
If you followed the directory structure from above, you should be able to navigate into the `preproc/` folder and simply call:
```
python preprocessing.py
```
You might want to also change `in_dir` variable from `preprocessing.py` script to the path containing the raw data:
```
# --- input files
in_dir = './00_input/'
```
All the lexicons used for data cleaning and filtering are available in `preproc/lexicon/` catalogue.
## Running the classification script
This script is intended to be used via command line:
```shell
$ ./runsequence.sh
```
It performs model trainign with `clitri/classifiers.py` script, model prediction with `clitri/discovry.py` and evaluation with
`track1_eval.py`.
## Output for Track 1: Cohort Selection for Clinical Trials
Our best score:
```
------------ met ------------- ------ not met ------- -- overall ---
Prec. Rec. Speci. F(b=1) Prec. Rec. F(b=1) F(b=1) AUC
Abdominal 0.6486 0.8000 0.7679 0.7164 0.8776 0.7679 0.8190 0.7677 0.7839
Advanced-cad 0.8302 0.9778 0.7805 0.8980 0.9697 0.7805 0.8649 0.8814 0.8791
Alcohol-abuse 0.2222 0.6667 0.9157 0.3333 0.9870 0.9157 0.9500 0.6417 0.7912
Asp-for-mi 0.8767 0.9412 0.5000 0.9078 0.6923 0.5000 0.5806 0.7442 0.7206
Creatinine 0.8000 0.8333 0.9194 0.8163 0.9344 0.9194 0.9268 0.8716 0.8763
Dietsupp-2mos 0.7885 0.9318 0.7381 0.8542 0.9118 0.7381 0.8158 0.8350 0.8350
Drug-abuse 0.4000 0.6667 0.9639 0.5000 0.9877 0.9639 0.9756 0.7378 0.8153
English 0.9125 1.0000 0.4615 0.9542 1.0000 0.4615 0.6316 0.7929 0.7308
Hba1c 1.0000 0.8286 1.0000 0.9062 0.8947 1.0000 0.9444 0.9253 0.9143
Keto-1yr 0.0000 0.0000 1.0000 0.0000 1.0000 1.0000 1.0000 0.5000 0.5000
Major-diabetes 0.8500 0.7907 0.8605 0.8193 0.8043 0.8605 0.8315 0.8254 0.8256
Makes-decisions 0.9762 0.9880 0.3333 0.9820 0.5000 0.3333 0.4000 0.6910 0.6606
Mi-6mos 0.3333 0.5000 0.8974 0.4000 0.9459 0.8974 0.9211 0.6605 0.6987
------------------------------ ---------------------- --------------
Overall (micro) 0.8397 0.9129 0.8786 0.8747 0.9354 0.8786 0.9061 0.8904 0.8957
Overall (macro) 0.6645 0.7634 0.7799 0.6991 0.8850 0.7799 0.8201 0.7596 0.7716
```
The official ranking measure is Overall (micro) F(b=1) (0.8904).
### Criterion
In this study - based on a textual hisotry of patients - we wanted to predict if they meet or not the following medical criteria:
- Abdominal
- Advanced-cad
- Alcohol-abuse
- Asp-for-mi
- Creatinine
- Dietsupp-2mos
- Drug-abuse
- English
- Hba1c
- Keto-1yr
- Major-diabetes
- Makes-decisions
- Mi-6mos
### Details
For details of this approach, please refer to this article: https://medinform.jmir.org/2019/4/e15980/.
If you use lexicons, or part of our code, please cite the above-mentioned article as:
```
Spasic I, Krzeminski D, Corcoran P, Balinsky A
Cohort Selection for Clinical Trials From Longitudinal Patient Records: Text Mining Approach
JMIR Med Inform 2019;7(4):e15980
URL: https://medinform.jmir.org/2019/4/e15980
DOI: 10.2196/15980
PMID: 31674914
```