Skip to main content
Fig. 5 | BMC Medical Research Methodology

Fig. 5

From: Assessment of predictive performance in incomplete data by combining internal validation and multiple imputation

Fig. 5

Data-driven simulation. Boxplots showing distribution of performance estimates across the 250 simulations of missing values into the complete-observations data set derived from the MONICA/KORA subcohort. a, AUC (10 years), performance of a model comprising the 15 inflammation-related markers; b, ΔAUC (10 years), added performance of the inflammation-related markers on top of the baseline covariates. Note that the variability across the 250 simulations reflects variability in imposing missing values as well as resampling variability, but not population variability as is part of the variability in Fig. 3. BS, bootstrap; CVK, K-fold CV; CVKrep, repeated K-fold CV MI, multiple imputation; MI(-y), multiple imputation without including the outcome; No val., no validation (i.e., apparent performance); OBB, out-of-bag estimate; opt.corr., ordinary -corrected estimate; SS, subsampling; Val, validation

Back to article page