Using Machine Learning Techniques to Predict Antimicrobial Resistance in Stone Disease Patients

We, as a team working on Machine learning Techniques, are very honoured to comment on this UroToday post for our publication in the World Journal of Urology. We stay focused the last couple of years on the use of machine learning techniques in almost all aspects of clinical medicine. Therefore in our hospital, we previously published some data on predicting antimicrobial resistance to aid empirical antibiotic treatment in the setting of the Intensive Care Unit through machine learning algorithms and we expanded our investigations in patients in the Urology Department where especially in stone disease patients is crucial to have more tools to support clinical decision on what antibiotic regimen should be started. We as urologists come across this dilemma many times in our career for stone disease patients in the preoperative setting or for postoperative complications and last but not least in the Emergency Department.

The WEKA-Data Mining Software (version 3.8.3) in Java Workbench was used for analysing our data, creating subgroups for our data for training and testing and we used a feature of WEKA to balance the groups for several variables such as the sensitivity of the species. All data were retrieved from the electronic database of our hospital for these patients and we used MicroScan system (Siemens) for antimicrobial susceptibility testing calculating also the minimum inhibitory concentration (MIC) for each antibiotic. We cross validate our data set for two different versions. The version for knowing the Gram Stain had a Receiver Operator Curve area of 0.768 and F-Measure of 0.708 which is considered fairly good for such algorithms and the version for knowing bacterial species had a ROC area of 0.874 and F-Measure of 0.783 which was surprisingly good.

Of course, our study has its limits too, as it cannot be widely used without training in other institutes and further future studies are needed with bigger data and different populations. We are looking forward to publishing more data using machine learning algorithms in the Urology field, especially in stone disease patients, and further data on evaluating with prospective studies.

Written by: Lazarou Lazaros MD, FEBU, PhDc, Surgeon Urologist, National and Kapodistrian University of Athens

Read the Abstract

Login