Explainable haemoglobin deferral predictions using machine learning models: Interpretation and consequences for the...

Area of Science:

• Biomedical Science
• Data Science
• Public Health

Background:

• Accurate prediction of hemoglobin (Hb) deferral is crucial for blood banks to optimize efficiency and donor retention.
• Complex predictive models are necessary for accuracy but require explainability for practical implementation.
• Estimating the impact of prediction models on blood supply is essential to prevent shortages before deployment.

Purpose of the Study:

• To develop an explainable model for predicting hemoglobin deferral in whole-blood donors.
• To assess the model's impact on blood supply by simulating earlier or later donation invitations.
• To reduce deferral rates and enhance blood bank operational efficiency.

Main Methods:

• Utilized donation visit data (Oct 2017-Dec 2021) including donor demographics, donation history, and ferritin levels.
• Employed Support Vector Machines for prediction and SHapley Additive exPlanations for model interpretability.
• Assessed model performance using precision and recall, and estimated blood supply impact.

Main Results:

• A novel, explainable model for predicting Hb deferral was developed.
• Application of the model could lead to 64% of non-deferred donors being invited earlier and 80% of deferred donors later.
• Predicted a 60% reduction in deferral rates, with specific impacts on female and male donors.

Conclusions:

• The developed model effectively predicts Hb deferral with high explainability.
• Implementing this model is projected to increase blood bank visits by 15% while significantly decreasing deferral rates.
• This approach offers a pathway to improve blood supply management and donor engagement.