Long-term forecasting of blood donations using time series of donation activity: Findings from seven blood...

To meet the long-term demand for blood products while preserving donor health in the long run, blood establishments must recruit a sufficient number of new donors annually. To determine what will suffice, it is essential to be able to forecast future blood donation volumes as a function of new donors using estimates based on historical donation activity.

Donor (n = 11,629,873) and donation data (n = 64,510,294) extracted from operational information systems of seven blood establishments were transformed into anonymous donation activity data (grouped by blood establishment, sex and blood group) in the form of time series of mean number of donations per donor, indexed by years since first donation. Linear models were fitted to the time series using ordinary least squares.

Out of the various estimated models, the best fit (mean R² over blood establishments 99.95%) for past mean donation activity was achieved when regressing the logarithm of cumulative donation activity on the logarithm of years since first donation and an indicator variable for the year of first donation. In addition to the predictions, comparison of the estimated parameters revealed that there are significant differences between blood establishments, translating to differences in the expected number of donations accumulated over the lifetime of a donor.

Average donation activity can be modelled using a few variables and simple models, with high precision. The estimates can be used to create forecasts of future donation volumes, which in turn can be useful in long-term blood donation management.