Predicting Blood Donations in a Tertiary Care Center Using Time Series Forecasting

Area of Science:

• Transfusion Medicine
• Health Informatics
• Operations Research

Background:

• Platelet stock management relies on whole blood donations (WBD) for pooled platelet production.
• Uncertainty in WBD rates necessitates backup strategies like apheresis collections.
• Accurate WBD prediction is vital for preemptive adjustments in platelet production.

Purpose of the Study:

• To develop and evaluate predictive models for daily WBD at a tertiary care center.
• To improve the management of platelet inventory by anticipating donation fluctuations.

Main Methods:

• Collected 62,248 WBD counts over three years.
• Utilized six prediction models (standalone and ensemble) to forecast daily WBD.
• Evaluated model performance using Mean Absolute Error (MAE).
• Identified Autoregressive Integrated Moving Average (ARIMA) with a 365-day frequency as a strong trend-modeling approach.

Main Results:

• Forecasting models achieved a MAE of approximately 8.6 donations.
• The predictive models demonstrated a 34% improvement over simpler methods (mean/median).
• ARIMA models effectively captured donation trend lines with a 365-day frequency, requiring at least two years of data.

Conclusions:

• Predictive modeling offers a significant improvement for managing platelet inventory.
• Accurate forecasting of WBD enables proactive adjustments to platelet production.
• ARIMA models show promise for long-term donation trend analysis in transfusion medicine.