Related Experiment Video
Updated: Jun 13, 2025
13:14
Generation of Discriminative Human Monoclonal Antibodies from Rare Antigen-specific B Cells Circulating in Blood
Published on: February 6, 2018
10.4K
A Machine Learning Method for Allocating Scarce COVID-19 Monoclonal Antibodies.
Mengli Xiao1, Kyle C Molina2, Neil R Aggarwal3
1Department of Biostatistics and Informatics, Colorado School of Public Health, University of Colorado Anschutz Medical Campus, Aurora.
JAMA Health Forum
|September 13, 2024
Summary
A machine learning policy learning tree (PLT) method optimized COVID-19 treatment allocation, reducing expected hospitalizations by 1.6% compared to observed methods. This approach offers a more effective strategy for distributing scarce therapeutics during public health crises.
Area of Science:
- Health Policy
- Medical Informatics
- Machine Learning
Background:
- Effective distribution of limited COVID-19 treatments was a critical policy challenge during the pandemic.
- Limited research has explored machine learning techniques, like policy learning trees (PLTs), for optimizing scarce therapeutic allocation using electronic health record (EHR) data.
Purpose of the Study:
- To evaluate if a machine learning PLT-based scarce resource allocation method can optimize treatment benefits for COVID-19 neutralizing monoclonal antibodies (mAbs) under resource constraints.
Main Methods:
- A retrospective cohort study utilized EHR data from October 1, 2021, to December 11, 2021 (training) and June 1, 2021, to October 1, 2021 (testing).
- Included patients tested positive for SARS-CoV-2 and qualified for COVID-19 mAb therapy per FDA emergency use authorization.
- The primary outcome was the potential reduction in overall expected hospitalizations if a PLT-based allocation system was used, compared to observed allocation.
Main Results:
- The training cohort included 9542 eligible patients (40.5% received mAbs), and the testing cohort had 6248 eligible patients (21.3% received mAbs).
- The PLT model-based allocation resulted in an estimated 1.6% reduction in overall expected hospitalizations in the testing cohort compared to observed allocation.
- The PLT system showed a greater reduction in 28-day hospitalization compared to the Monoclonal Antibody Screening Score.
Conclusions:
- A PLT method, integrated with EHR data platforms, can enhance real-time allocation of scarce treatments.
- Implementing this PLT-based allocation could lead to fewer hospitalizations than standard care.
- The PLT approach demonstrated greater expected reductions in hospitalizations compared to a common point system.
