Figure 1
AUC variable importance with respect to predicting the outcome “severe versus mild” for young and elderly patients calculated using random forests. The larger the importance value of a covariate is, the greater the improvement in prediction performance by including this covariate in prognosis. Complement C3, C4, LYM, the ratio of CRP versus ALB, and UA seem to influence the prognosis of the development of severity mostly in young patients. The bars show the medians of the 20 importance values calculated using the 20 imputed data sets from the multiple imputation. The error bars illustrate the variabilities of the importance values: The lower/upper ends show the first/third quartiles of the 20 importance values, that is, 25% percent of the importance values lie below/above these values. To make the raw importance values comparable between young and elderly patients, both for the young and for the elderly patients, the raw importance values were divided by the means of all importance values with positive sign.
