ABSTRACT:
The increasing prevalence of multidrug-resistant (MDR) K. pneumoniae leads to an alarming global health challenge, notably in hospital-acquired infections among immunocompromised patients. Conventional models, known as antibiotic susceptibility testing (AST), need up to 48 hours and often delay effective medication, underscoring the urgent need for rapid diagnostic alternatives. This review systematically examines the applications of machine learning (ML) in predicting antibiotic resistance (ABR) in K. pneumoniae, focusing on their clinical utility in detecting resistance to critical antibiotics such as carbapenems and polymyxins. An extensive literature search across six databases—Scopus, PubMed, ProQuest, IEEE Xplore, ASME Digital Collection, and the Cochrane Library— identified 19 relevant studies, revealing that ML techniques—particularly random forests, logistic regression, and support vector machines (SVMs)—achieve high predictive accuracy, with area under the curve (AUC) values frequently exceeding 0.9. The integration of whole-genome sequencing (WGS)-based data with clinical metadata, including patient demographics, infection histories, and treatment records, further enhances predictive performance and supports more precise treatment decisions. Despite these advancements, challenges remain, including limited external validation, data standardization issues, and a lack of transparency in ML decision-making, which inhibits clinical adoption. Addressing these barriers through broader validation studies and expanding ML applications to additional antibiotic classes beyond mentioned antibiotics will be essential for optimizing clinical implementation. In conclusion, ML-based diagnostic tools have the potential to reshape ABR detection, enabling faster, data-driven treatment strategies that improve patient outcomes and support global antibiotic stewardship efforts.