NCT06594484

Machine Learning Versus Traditional Scores in Predicting Erythrocyte Need

Comparative Analysis of Machine Learning Versus Conventional Models for Predicting Erythrocyte Need in Cardiovascular Surgery

• Status: Completed
• Phase: —
• Study type: Observational
• Enrollment: 430 (actual)
• Sponsor: Kocaeli City Hospital · Other Government
• Sex: All
• Age: —
• Healthy volunteers: Not accepted

Summary

In this study, we compared perioperative bleeding prediction scores with our machine learning-based prediction system in predicting the need for erythrocyte suspension during cardiovascular surgery.

Detailed description

The success of ML algorithms in predicting perioperative blood product use in CABG remains an under-tested topic. Unnecessary preparation of blood products or not being able to supply them when necessary is critical for both patient safety and the effective use of hospital resources \[8\]. Bleeding amounts and blood product use strategies can vary with institute protocols. Scoring systems that determine the general framework may not perform well due to local factors. ML algorithms can be created locally according to previous patient data of each clinic and can improve themselves with learning mechanisms, suggesting significant potential in this field. In the current study, a new estimation system created with the ML algorithm was compared with the known estimation systems. Comparing the ML algorithm with 6 different classical scoring systems is important in terms of demonstrating the potential of this technology. The aim of this study is to investigate whether the model created with ML in predicting perioperative blood product consumption in cardiovascular surgeries is superior to predictive scoring systems that have proven themselves in the literature. Secondary aim is to compare the predictive value of using more than one scoring system in combination.

Conditions

• Erythrocyte Transfusion
• Machine Learning

Interventions

Type	Name	Description
OTHER	Ml Based Algorithm 1	The values in the ML algorithm were selected according to logistic regression analysis and the values used in the other six scores tested. The success rate of the constructed networks correct predictions was considered as the success rate of the algorithm. The usefulness of the test was determined through AUROC analysis. Two algorithms were tested in our study. In the first algorithm (ML1), the dependent variable was erythrocyte suspension (ES) consumption, and the independent variables included patients; demographic data, laboratory data, and operational data
OTHER	Ml Based Algroithm 2	is an Ml algorithm created by combining commonly used bleeding scores
OTHER	Bleeding Scores	ACTION CRUSCADE TRACK WILL-BLEED PAPWORTH TRUST ACTAPORT skores used to predict ES need

Timeline

• Start date: 2024-02-22
• Primary completion: 2024-05-22
• Completion: 2024-05-30
• First posted: 2024-09-19
• Last updated: 2024-09-19

Locations

1 site across 1 country: Turkey (Türkiye)