Premium
Prediction of hemoglobin levels in whole blood donors: how to model donation history
Author(s) -
Baart A. Mireille,
Vergouwe Yvonne,
Atsma Femke,
Moons Karel G.M.,
Kort Wim L.A.M.
Publication year - 2014
Publication title -
transfusion
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.045
H-Index - 132
eISSN - 1537-2995
pISSN - 0041-1132
DOI - 10.1111/trf.12430
Subject(s) - concordance , hemoglobin , mixed model , linear regression , statistics , random effects model , medicine , statistic , linear model , mathematics , demography , meta analysis , sociology
Background Recently, prediction models for hemoglobin ( Hb ) deferral risk have been developed. These models consider the previous Hb level plus change in Hb . Here, we investigated if the performance of models could be improved by considering more information on Hb level history. Study Design and Methods Data of 166,497 D utch whole blood donors with sequential Hb measurements during 2 years (760,444 in total) were used to develop and internally validate three different regression models: two simple linear models with Hb level history included as 1) Hb at the previous visit plus change in Hb or 2) mean of all previous Hb levels and one mixed‐effect model including measurements of all previous Hb levels. Results Thirteen percent of men and 21% of women were deferred because of a low Hb level at least once in 2 years. The simple linear models and the mixed‐effect model performed similar, if an estimate of the random intercept of the mixed‐effect model was used for individual donors to calculate the predicted Hb level. In men, the concordance (c)‐statistic ranged from 0.87 to 0.89 and the R 2 ranged from 0.42 to 0.45. In women, the c‐statistic ranged from 0.81 to 0.84. Values of R 2 in women were higher for the simple linear models than for the mixed‐effect model, 0.37 and 0.40 versus 0.30, respectively. Conclusion Previous Hb levels could be summarized with one predictor as the mean value of all previous Hb levels. This predictor can be used in an easy‐to‐use simple linear regression model.