Skip to main content

Table 2 Evaluation of model predictive performance

From: Development of a bedside tool to predict the probability of drug-resistant pathogens among hospitalized adult patients with gram-negative infections

Performance Metrica Hospital Community Hospital Community Hospital Community
AUC for training data 0.90 0.79 0.87 0.71 0.94 0.84
AUC for test data 0.92 0.79 0.87 0.70 0.94 0.83
Correct prediction among top 10% scored subjects in test data, n/N (%) 145/3496 (4.15) 165/3607 (4.57) 627/3411 (18.38) 786/3624 (21.69) 405/3586 (11.29) 388/3628 (10.69)
Lift of top 10% scored subjects in test datab 7.8 3.8 6.4 2.6 7.9 5.0
  1. 3GC-R third-generation cephalosporin-resistant Enterobacteriaceae, AUC area under the receiver operating characteristic curve, CRE carbapenem-resistant Enterobacteriaceae, LASSO least absolute shrinkage and selection operator, MDRP multidrug-resistant Pseudomonas aeruginosa
  2. aThe method of predictor selection was LASSO logistic regression that minimized the cross-validation misclassification error; LASSO C value was 0.1
  3. bLift defined as probability of a positive case given a top 10% score divided by the probability of a positive case in overall sample. This ratio evaluates how much a top score enriches for selecting positive cases compared with random sampling in the absence of a model. A higher lift indicates a stronger association between the predicted score and the outcome