Identifying determinants of persistent MRSA bacteremia using mathematical modeling
Fig 6
Key determinants to distinguish resolving, persistent and relapsing bacteremia.
(A) Line plots of the number of total SA over time (0–180 days) for resolving bacteremia (RB) without relapse, persistent bacteremia (PB) without and with relapse bacteremia. (B) Violin plots and box plots of the parameter distributions. Dotted lines indicate the range of randomized parameter values. (C) Weights calculated by QPFS methods are shown. The ranked features by QPFS, cP and gP were used to build a multinomial logistic regression model. For the comparison, all parameter and all parameter except for cP and gP were applied to the classification model. The overall accuracy and area under ROC curve for each class are graphed. In the classification, an equal number of datasets in each class were used by reducing the number of data in major classes, RB without relapse and PB with relapse, to the same number of minor class, 940 of PB without relapse. (D) Two-dimensional scatter plot for cP and gP with probability density of each value.