Truace
Subscribe

1D CNN prediction of inpatient red blood cell transfusion volume

Source article: Convolutional neural network for red blood cell transfusion prediction: A machine learning based study

Background Subjective, inconsistent clinician judgement on red blood cell (RBC) transfusion demand causes unnecessary blood wastage, increased perioperative safety risks, and poorly balanced hospital blood inventory. To tackle these persistent clinical limitations, we developed a workflow-adapted clinical decision support tool: a one-dimensional convolutional neural network (1D CNN) trained on routinely acquired haematological and clinical variables to stratify inpatient RBC transfusion volume categories. Method…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1784· cite & verify
Convolutional neural network for red blood cell transfusion prediction: A machine learning based study
"Cobra Gold 20: US, Royal Thai sailors conduct whole blood transfusion training" by Armed Services Blood Program is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.
Trace impact readingNegative state
P 71The P score combines the specificity and measured human impact of the grounded problem claim with the strength of this Trace’s cited sources.

Both sides are scored from claims and sources, not community votes.

G 63The G score combines the specificity and measured human impact of the grounded gain claim with the strength of this Trace’s cited sources.

In brief

Researchers at Beijing Friendship Hospital built a one-dimensional convolutional neural network to sort inpatients into expected red blood cell transfusion volume categories. It used routinely collected haematological and clinical data from a retrospective cohort of 7,250 patients and was tested on a held-out 30% of the data.

The tool targets inconsistent clinician judgement, which the authors link to blood wastage, perioperative safety risks and unbalanced hospital blood inventory. Its usefulness is uncertain because it was built at one centre and missed rarer, larger transfusions, so it has not been shown to work across other hospitals or high-demand cases.

Main points

  1. The cohort covered 7,250 patients treated between April 2024 and June 2025.
  2. The model used 19 standardised pre-transfusion variables to predict categories of 1, 2, 3, 4 or 5 or more units.
  3. Overall test accuracy was 0.7314, and the 2-unit class made up 68.03% of cases.
  4. Minority classes, including 3 units and 5 or more units, had low prevalence and weaker performance.
  5. The authors call for external validation, more minority-class data and indication-specific sub-models.

The gain

In a retrospective single-hospital cohort, the model forecast routine transfusion volume categories from 19 pre-transfusion variables, correctly identifying the dominant 2-unit class.

The problem

The model under-detected high-volume transfusion events, with weak performance in the minority classes of 3 units and 5 or more units, which limits clinical reliability for these cases.

The rundown

The study frames subjective transfusion demand estimation as a persistent clinical problem. The authors designed the model as a workflow-adapted decision support tool that relies on variables already collected in routine care.

Methods included optimisation with the Adam algorithm and evaluation on a 30% hold-out test set. The authors report that the model converged without overfitting and also characterised baseline clinical features to describe transfusion contexts and indications.

The authors conclude the model reliably forecasts routine transfusion needs in a single-centre setting. They state that external validation across diverse centres, targeted enrichment of minority classes and indication-specific sub-models are mandatory for better generalisability and clinical utility.

What this doesn’t fix

The model was developed and tested in a single-centre retrospective cohort and under-detected high-volume transfusions, so generalisability and usefulness for rare, large transfusions remain unproven.

Sources

  1. Peer-reviewedTransfusion Medicine2026-10-07

The debate