top of page

Daily Intervention Lists for Covid-19 Therapy Using Points-Based Predictive Model

Yatir Ben-Shlomo,
Data Scientist

Yatir.jpg

Background

Once Covid-19 Therapy became available (Regeneron’s REGEN-COV™ in September 2021, and later Pfizer’s PAXLOVID™ and Merck’s Molnupiravir™) The Israeli Ministry of Health purchased an initial stock, and declared the drug, as “a national resource in short supply […] Patients will be determined based on a model that will weigh personal risk by using criteria such as age, illnesses and medical background, and vaccination status” (Haaertz).
 

The Challenge

Each HMO was allocated daily quotas for treatment, and there was an urgent need to prioritize the daily intervention lists using risk prediction models in a way that will combine fairness, explainability and high predictivity.

Solution

For this task, we leveraged a points-based grading system that was derived from the raw coefficients of a logistic regression model built for predicting severe Covid-19 for infected patients.

Examples of variables used in the predictive model are age, chronic conditions which are known as Covid-19 risk factors, hospitalization history, and Covid-19 vaccination history.

Each coefficient was linearly transformed into integer risk points.
The final risk score is the sum of all risk points associated with all model variables.

The Operationalization of the intervention lists is as follows

  • The list of new Covid-19 Positives is retrieved from the appropriate database every morning, along with all relevant personal medical information required for risk scores calculation.

  • The list of patients with score higher than a pre-specified threshold is then established and sorted by score.

  • Each prediction is accompanied by the full list of contributing factors in a way that allows verification of this score, or its recalculation, in rare cases of misdiagnosis

  • If the Covid-19 therapy is limited by contra indications, such as lack of a recent Kidney function lab test (GFR), these variables are also retrieved and applied as exclusion criteria.

  • The list is then split by district, with each sub-list automatically sent to the appropriate district-level “war room” for further distribution to the regional clinics.

  • Patients will then receive a phone call from their primary physician or their regional clinic offering the therapy.

  • Once patients consent is achieved – the therapy is then shipped into the patient home by district level dedicated teams.
     

This process runs automatically every morning, 6 days a week since September 2021, and counting.

bottom of page