Machine-learning models to predict iron recovery after blood donation: a model development and external validation...

Area of Science:

• Biomedical Informatics
• Hematology
• Machine Learning

Background:

• Iron deficiency is a concern for blood donors, potentially leading to deferrals.
• Machine learning models can predict iron biomarkers post-donation to manage deficiency.
• International validation of such predictive models is lacking.

Purpose of the Study:

• To develop and externally validate machine learning models.
• To predict returning blood donors' hemoglobin and ferritin levels.
• To assess model generalizability across diverse international settings.

Main Methods:

• Developed models using retrospective blood donation data (US-based RISE study).
• Externally validated models on international cohorts (USA, South Africa, Netherlands).
• Used common donor data: donation history, demographics, baseline iron biomarkers.

Main Results:

• Models showed consistent performance in predicting hemoglobin across datasets (RMSPE ≤ 8%).
• Predicting ferritin was more accurate with baseline ferritin data (RISE RMSPE 14.9%).
• External validation demonstrated strong generalizability for hemoglobin prediction.

Conclusions:

• Machine learning models for predicting hemoglobin and ferritin are effective.
• These models generalize well across diverse international blood donation settings.
• Individualized management of iron deficiency can be enabled, supporting blood supply.