cskyan/ehr-dream-challenges repository overview

⁠Example model for COVID-19 DREAM Challenge: Question 1

⁠Overview

Here we describe how to build and run locally an example model provided for Challenge Question 1 of the COVID-19 DREAM Challenge⁠. The goal of this continuous benchmarking project is to develop models that take as input the electronic health records (EHRs) of a patient and outputs the probability of this patient to be tested positive for COVID-19.

⁠Description of the model

This example model takes 13 features that include age, clinical symptoms and vital signs. The feature selection refers to the research conducted by Feng et al.⁠ and Giuseppe et al.⁠. Here we use these features listed in the table below to build a simple rule-based model that generates a probability of a patient being COVID-19 positive. First, we generate a risk score for each patient based on the hypothesis listed in the column Risk score +1 if. The probability for a patient to be COVID-19 positive is then given by Risk score / num_features.

⁠Dockerize the model

1. Start by cloning this repository.

2. Move to this example folder

3. Build the Docker image that will contain the move with the following command:

   docker build -t awesome-covid19-q1-model:v1 .
   

   Copy

⁠Run the model locally on synthetic EHR data

1. Go to the page of the synthetic dataset⁠ provided by the COVID-19 DREAM Challenge. This page provides useful information about the format and content of the synthetic data.

2. Download the file synthetic_data.tar.gz⁠ to the location of this example folder (only available to registered participants).

3. Extract the content of the archive

   $ tar xvf synthetic_data.tar.gz
   x synthetic_data/
   x synthetic_data/procedure_occurrence.csv
   x synthetic_data/location.csv
   x synthetic_data/visit_occurrence.csv
   x synthetic_data/condition_era.csv
   x synthetic_data/device_exposure.csv
   x synthetic_data/drug_era.csv
   x synthetic_data/observation.csv
   x synthetic_data/goldstandard.csv
   x synthetic_data/drug_exposure.csv
   x synthetic_data/condition_occurrence.csv
   x synthetic_data/person.csv
   x synthetic_data/measurement.csv
   x synthetic_data/observation_period.csv
   

   Copy

4. Create an output and scratch (optional) folders.

   mkdir output scratch
   

   Copy

5. Run the dockerized model to generate predictions for the patients included in the synthetic dataset.

   docker run \
       -v $(pwd)/synthetic_data:/data:ro \
       -v $(pwd)/output:/output:rw \
       -v $(pwd)/scratch:/scratch:rw \
       awesome-covid19-q1-model:v1 bash /app/infer.sh
   

   Copy

6. The predictions generated are saved to output/predictions.csv. The column person_id includes the ID of the patient and the column test-positive the probabily for the patient to be COVID-19 positive.

   $ cat output/predictions.csv
   person_id,score
   0,0.6153846153846154
   1,0.5384615384615384
   2,0.5384615384615384
   3,0.3076923076923077
   ...
   

   Copy

⁠Submit this model to the COVID-19 DREAM Challenge

This model meets the requirements for models to be submitted to Question 1 of the COVID-19 DREAM Challenge. Please see this page⁠ for instructions on how to submit this model.