EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

1189. Optimizing red blood cell matching policies to reduce alloimmunization in regularly transfused patients

Invited abstract in session MB-15: Machine learning and analytics in healthcare, stream OR in Health Services (ORAHS).

Monday, 10:30-12:00
Room: 18 (building: 116)

Authors (first author is the speaker)

1. Merel L. Wemelsfelder
Business Analytics, University of Amsterdam
2. Folarin Oyebolu
MRC Biostatistics Unit, University of Cambridge
3. Dick den Hertog
University of Amsterdam
4. William J. Astle
MRC Biostatistics Unit, University of Cambridge
5. Mart P. Janssen
Transfusion Technology Assessment, Sanquin Research

Abstract

We propose a method for optimizing daily red blood cell (RBC) matching policies to minimize long-term alloimmunization rates in regularly transfused patients. When antibodies are formed against an antigen, a risk that comes with RBC transfusions, the patient can never be mismatched on that antigen again. Clinical guidelines recommend extended preventive matching for patients at high risk of alloimmunization, such as sickle cell disease patients, who need a transfusion every six weeks. They must be matched not only for major antigens A, B, and D, but for an additional 5 to 12 minor antigens as well. We use linear optimization for daily RBC matching, including parameters for weighting elements of the objective function. We train machine learning models to predict the long-term outcomes of the matching strategy, then use these models to optimize parameter values. A wide range of potentially optimal parameter values is found, likely a result of the stochastic nature of real-world scenarios, combined with the interplay of factors influencing long-term outcomes. We implemented a selection of these for daily matching, and observed an increased performance for several long-term outcomes with respect to the baseline setting without tuned parameters. The methodology developed in this paper in the context of RBC matching, could also be used in applications where we optimize long-term performance via a parameterized short-term optimization model.

Keywords

Status: accepted


Back to the list of papers