EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2028. Optimal Usage of Packed Red Blood Cells in Preterm Neonates Requiring Frequent Small-Volume Transfusions

Invited abstract in session MD-17: Medical decision making, stream OR in Health Services (ORAHS).

Monday, 14:30-16:00
Room: 40 (building: 116)

Authors (first author is the speaker)

1. Malavika Krishnakumar
Health Sciences Research, Amrita Vishwa Vidyapeetham
2. Merin Mathew
Mathematics, Amrita Vishwa Vidyapeetham
3. Anjali Ajith
Neonatology, Amrita Institute of Medical Sciences
4. Sruthi Suresh
Pediatrics, Amrita Institute of Medical Sciences
5. Perraju Bendapudi
Neonatology, Amrita Institute of Medical Sciences
6. Linda John
Transfusion Medicine, Amrita Institute of Medical Sciences
7. Dhanya A
Transfusion Medicine, Amrita Institute of Medical Sciences
8. Veena Shenoy
Transfusion Medicine, Amrita Institute of Medical Sciences
9. Georg Gutjahr
Department of Health Science Research, Amrita Institute of Medical Sciences and Research Center, Kochi, Kerala, India

Abstract

Preterm neonates with low birth weight often require frequent small-volume blood transfusions. Instead of addressing each transfusion event individually, reserving blood bags in advance may prove to be economical and minimize donor exposure. However, it might introduce storage challenges and could lead to the wastage of blood. An optimal strategy for reserving blood bags depends on neonatal clinical characteristics and available blood reserves in the transfusion service.

To model this decision problem under uncertainty, we assume that the known and unknown random variables are structured as a Bayesian network and will use a partially observed Markov decision process. At each day, the state of the process is described by the blood inventory and the neonatal clinical characteristics. Neonatal gestational age, weight, and medical conditions are observed variables. In contrast, the length of stay and the total amount of required blood are unknown at the time of decision-making.

The proposed model can be expressed as a dynamic Bayesian network model in which the decision variables can be optimized by dynamic programming. We explore the sizes of the problem instances that can be solved in this exact way and also propose an efficient approximate dynamic programming method for large instances. We illustrate the methods with one year of data from a neonatal ICU in South India where financial and blood-availability constraints play a crucial role in finding optimal strategies.

Keywords

Status: accepted

Back to the list of papers