Fig. 3From: Seasonally adjusted laboratory reference intervals to improve the performance of machine learning models for classification of cardiovascular diseases(a) AUROC performance metric values for the version 2 neural net model. Dots indicate the median AUROC values and associated lines show the corresponding 95% CIs. The symbol (*) indicates that the version 2 model performed significantly better than the version 1 model (with respect to accelerated bootstrap gains 95% CIs) and the symbol (•) indicates that the models performed the same statistically; no symbol indicates the model performed significantly worse (with respect to accelerated bootstrap gains 95% CIs). (b) Heatmap of net AUROC gains (version 2—version 1) across all available ICD-10 chapter IX codes for the four ML models assessed in this study. Clustering along the Y-axis was performed using hclust algorithm in R. The scale was bounded between + 2% and -2% to enable easier viewing, although net gains can be higher or lower, as listed in supplementary Table 4. Non-significant gains/ losses (95% CI accelerated bootstrap gains didn’t include 0) were changed to “0” before performing the clustering and are therefore represented as white. All corresponding values are also included in supplementary Table 4Back to article page