Table 4. Performance measures for prediction of extended-spectrum β-lactamase-producing Enterobacterales infection, University Hospital Basel, Switzerland, January 2010–December 2016 (n = 376).

Performance measure	Prediction of infection/colonisation with ESBL-PE on hospital admission	Prediction ESBL-PE in patients with bacteraemia
	Score by Tumbarello et al. [11]	Decision tree by Goodman et al. [12]a	Prediction model derived using stepwise variable selectionb	Known history of ESBL-PE colonisation/infectionc
Sensitivity	67.0%	33.0%	74.5%	34.0%
Specificity	52.8%	97.2%	67.7%	97.2%
Positive predictive valued	32.1%	79.5%	43.5%	80.0%
Negative predictive valued	82.8%	81.3%	88.8%	81.5%
Youden-Index	0.2	0.3	0.4	0.3
Hosmer–Lemeshow statistic	1.96 (p = 0.855)	4.43 (p = 0.816)	1.61 (p = 0.657)	NA
Area under the curve (AUC)	0.627e	0.651f	0.759 (0.710g)	0.656 (0.598g)

ESBL-PE: extended-spectrum β lactamase-producing Enterobacterales; NA: Not applicable.

^a For calculations of the Hosmer–Lemeshow statistic and the area under the curve (AUC), logistic regression analyses including the classifying variable, i.e. ESBL-prediction positive or negative, into the regression model was performed.

^b Including history of ESBL-PE colonisation or infection, admission from another healthcare facility, and antibiotic therapy with β-lactams or fluoroquinolones lasting > 48 h during the 3 months preceding admission.

^c Selected by lasso regression and recursive partitioning algorithms (decision tree statistics).

^d Both the positive and the negative predictive values are influenced by the study design and thus have to be interpreted with caution.

^e Ranging from 0.83 (derivation cohort) to 0.92 (validation cohort).

^f Reported as 0.77 in the original publication (for the final decision tree and following cross-validation).

^g k-fold cross-validation.