COVID-19 Disease Management

Method


By using a dataset of COVID19 patients’ Electronic Health Records (EHRs) and applying machine learning techniques we aim to detect, monitor, and predict disease progression and remission. These capabilities will assist in prognosis and treatment of patients, and will also enable institutes to better plan their resource allocation (such as ICUs beds, respiration machines and relevant staff). Overall, our objective is to contribute to the global efforts in understanding and treating the COVID-19 disease.

 

Goal


Develop predictors and indicators for COVID-19 disease trajectory, progression and remission, and risk-factors that will enable better disease management.

Data Acquisition and Representation -
The raw EHR data (acquired from TrinetX Ltd. , an integrator of health data from hundreds of health institutes around the world), is a heterogeneous multivariate time series, with many fields of interest reported at various rates, and it is highly sparse by nature. Utilizing supervised and unsupervised deep learning methods, we are developing models for data embedding and representation, that extract useful information and insights.

 
Severity index and score -
Due to the information scattering, sparsity, and lack of historical data and well-formed practices during a pandemic outbreak labels representing the severity of the disease at any given time needs to be developed. The development of measurables scores and indices (lables) to indicate severity of disease based on patients’ measurements, is the subject of an on-going effort.

 
Patient’s “fingerprint” -
The pandemic taught us that different individuals experience a different illness trajectory. We are aiming at finding patient’s “fingerprint” – a unique indicator based on the patient’s medical history (e.g. pre-existing medical conditions, medications, and treatments) that may influence the severity at which this individual will suffer from COVID-19 once infected. Such a fingerprint may be a useful predictor also for other infectious diseases.

 
Predictive models –
Develop machine learning models to detect and predict the severity of the disease and the disease trajectory over time based on data from patients’ EHR.

 

 

 

Resources:

"How to evaluate Electronic Health Records representations for Machine Learning models?" | Written by Keren Ofek Granov

    • Gil Tomer, PhD

      Project Leader

    • Shai Fine, PhD

    • Yarden Rachamim

    • Itamar Efrati

    • Keren Ofek