Machine Learning Model for Predicting Multidrug Resistance in Clinical Klebsiella pneumoniae Isolates

Area of Science:

• Clinical microbiology
• Computational biology
• Infectious disease epidemiology

Background:

• Klebsiella pneumoniae exhibits increasing resistance to carbapenems and broad-spectrum antibiotics.
• Timely antibiotic selection is critical in severe infections like septic shock, but conventional resistance testing is slow.
• Rapid antimicrobial resistance prediction is needed to guide effective empirical therapy.

Purpose of the Study:

• To evaluate machine learning (ML) algorithms for predicting antimicrobial resistance in Klebsiella pneumoniae.
• To compare the performance of ML models against conventional methods for resistance testing.
• To identify the most effective ML algorithm for rapid resistance prediction.

Main Methods:

• Retrospective analysis of 607 Klebsiella pneumoniae isolates (2017-2024) with antibiogram data.
• Integration of patient demographic and clinical information with isolate data.
• Application and evaluation of Decision Tree (DT), Support Vector Classifier (SVC), K-Nearest Neighbors (KNN), and Random Forest (RF) algorithms.

Main Results:

• The Random Forest (RF) model demonstrated superior performance, achieving an average AUC of 0.96.
• RF accurately predicted resistance to 22 antibiotics, including critical ones like Ertapenem (100%), Imipenem (93%), and Meropenem (95%).
• Feature importance metrics highlighted key predictors for resistance.

Conclusions:

• Machine learning models, particularly RF, offer a powerful tool for rapid antibiotic resistance prediction.
• These models can significantly aid in making accurate empirical treatment decisions.
• ML-based prediction supports enhanced antimicrobial stewardship programs.