Search form

Medisch Labo Medina

Contact details
Traineeship proposition

Abstract advanced bachelor of bioinformatics 2019-2020: Development of a predictive tool for ESBL and CPE outcome based on Vitek2® results

Routine antibiotic susceptibility testing (AST) and resistance mechanism detection is performed by a Vitek2® analysis. Organisms suspected by Vitek2® analysis for Extended Spectrum Beta-Lactamase (ESBL) or Carbapenemase Producing Enterobacteriaceae (CPE) receive a comment-mark by the system. Vitek2® gives a mark for ESBL when the sensitivity of cefotaxime and ceftazidime is decreased (Minimum Inhibitory Concentration ((MIC)-value > 1mg/l) and upon detection of ESBL phenotypes. A mark for CPE will be given when the MIC-value of Meropenem is ≥ 0.5μg/ml; or the MIC-value of Ertapenem is ≥ 1μg/ml; or phenotype Carbapenemase is detected. Each ESBL- or CPE-suspected organism is confirmed by a disk diffusion antibiogram, which is a time-consuming job. There are a lot of possible ESBL samples and a few possible CPE samples. The results of the disk diffusion antibiogram are stored in large MS-Excel files, one for ESBL and one for CPE.
To reduce the amount of organisms needed to be confirmed, and thus reduce the workload in the lab, an additional rule was implemented several years ago. This rule considered an organism ESBL positive when the gentamicin MIC-value by Vitek2® analysis was ≥ 8 μg/ml and ciprofloxacin MIC-value was ≥ 4 μg/ml. For lower values, an extra confirmation test was still necessary. However, this rule was never validated and lacked reliable correctness. Therefore, the rule isn’t used anymore resulting again into an increased amount of extra confirmation tests to be performed.
The objective of this traineeship is to create a user friendly web interface where the results of the disk diffusion antibiogram can be entered easily into a database and more importantly, where this database could be used for the identification and validation of new rules. These rules should predict the outcome with sufficient reliability and reduce the workload considerably.
All the data were entered in a MySQL database from where the relevant data could be queried. We were able to identify and test a new and reliable rule based on the already available test information via statistical analysis (different barplots, trees, …) using R, i.e.: if Vitek2® asks to perform a confirmation test and the MIC-value of cefotaxime ≥ 64 μg/ml, the organism is ESBL positive.
This new rule was tested for all organisms and each organism separately. The most reliable results were found for Escherichia coli and for Klebsiella pneumoniae, i.e.:
• While the previous rule includes only 75 (72 true positives (TP) and 3 false positive (FP)) samples with a correctness of 96% for Escherichia coli, the new rule, based on a statistically more relevant set of 1222 (1189 TP and 33 false positive FP) samples, resulted into a correctness of 97.30%;
• For Klebsiella pneumoniae, the previous rule contained only 16 (15 TP and 1 FP) samples with a correctness of 93.75%, where the new rule contains 197 (192 TP and 5 FP) samples and a correctness of 97.46%;
• The other organisms didn’t have enough samples available to draw reliable conclusions.

For Escherichia coli and Klebsiella pneumoniae we conclude that our new rule reduces the need for additional confirmation tests with approximately 60% and this with a higher reliability compared to the rule previously implemented in the lab. Unfortunately, we weren’t able to find a predictive rule for CPE which is in accordance with our earlier expectations.


9200 Dendermonde


Traineeship supervisor
Evilien Vekens
Via Map