Prediction models for hemoglobin deferral in whole blood donors

Each year, a relevant proportion of the invited blood donors is eventually deferred from donation because of low hemoglobin (Hb) levels. Deferrals are meant to protect donors from developing iron deficiency anemia after a blood donation, however, they may increase the risk of donor lapse, even though the donor may actually meet the Hb criterion at the time of the next donation invitation. Early estimation of the risk of Hb deferral on the next visit to the blood collection center could be helpful in the management of the blood donation program. This thesis presents studies on prediction models for Hb deferral in whole blood donors. After developing a first prediction model in a sample of Dutch whole blood donors, we developed sex-specific models in a large cohort consisting of 220,946 Dutch whole blood donors. Strong predictors of Hb deferral were Hb level measured at the previous visit, age, seasonality, difference in Hb levels between the previous two visits, time since the previous visit, deferral at the previous visit, and the total number of whole blood donations in the past two years, for both men and women. The discriminative ability of the models was good: the concordance (c)-statistic was 0.89 for men and 0.84 for women. Subsequently, the prediction models were externally validated in a cohort of Irish whole blood donors. Validation demonstrated underestimation of predicted risks and lower c-statistics for men and women compared to the Dutch cohort. Updating the models for the Irish donors resulted in different predictor effects. This improved mainly the model calibration; the improvement in discrimination was small. Increased zinc protoporphyrin (ZPP) levels can indicate iron deficiency, and may be predictive for Hb deferral. Therefore, the added value of ZPP levels to the prediction models was investigated. Addition of ZPP into the models improved discrimination, particularly in women. The added value of ZPP was confirmed by measures of clinical usefulness. The prediction models described above are logistic regression models predicting the risk of Hb deferral. Furthermore, the logistic models consider only the previous Hb level plus a change in Hb level as information on Hb level history. In the last part of this thesis three different linear regression models predicting continuous values of Hb level were developed. The models differed in type of regression model and predictors for Hb level history. Results showed that the history of Hb levels can easily be incorporated in a simple linear prediction model as the mean of all previous Hb levels. A mixed effect model that included all previous Hb levels separately did not outperform the simple linear model. The performance of the linear regression models was similar to the performance of the logistic regression models. Results from studies presented in this thesis show that with a limited number of easy-to-measure characteristics the risk of Hb deferral in whole blood donors can be reliably predicted.