FIGURE 2.

The effect sizes and importance of variables in haemoglobin and deferral prediction. On the left panel, the regression coefficients of the DLMM when predicting haemoglobin on variables of the combined eProgesa and Biobank data. The dots and the lines denote the posterior means and the 95% highest posterior density intervals (HPDIs), respectively, for each variable and sex. In order to make the regression coefficients comparable, we left the binary variables as they are but scaled other variables by 2  SD. Hence, the units of the regression coefficients are two times the standard deviation. On the right, the importance of variables, when predicting deferral using a random forest model on eProgesa data, are marked with dots. With the random forest model, we did not train separate models for the male and female subsets but instead used sex as a predictor. Note that for both DLMM and random forest (RF) models the previous haemoglobin was clearly the most important variable. The difference of effect size between sexes seems to be mostly small, the age being a notable exception