Heatmap-Based and Regression Analysis of Antibiotic Co-Resistance Networks Among Gram-Positive Bacterial Isolates in a Tertiary Care Hospital

Description

Antimicrobial resistance (AMR) among Gram-positive bacteria is an escalating global health concern, with resistant pathogens such as Staphylococcus aureus and Enterococcus faecium listed among the World Health Organization’s high-priority threats. In 2019 alone, AMR was linked to nearly 5 million deaths globally, with methicillin-resistant S. aureus (MRSA) among the leading causes. This burden falls disproportionately on low- and middle-income countries, where gaps in healthcare infrastructure and infection control amplify the impact of resistant infections. Once considered a greater problem in Gram-negative organisms, resistance in Gram-positive bacteria now demands equivalent surveillance attention and novel analytical strategies for intervention. Conventional approaches to AMR surveillance typically assess resistance to individual antibiotics in isolation, summarizing the proportion of non-susceptible isolates per agent. However, this one-drug-at-a-time method fundamentally overlooks how resistances co-occur within a single organism; for instance, resistance to erythromycin commonly accompanies clindamycin resistance, and MRSA characteristically demonstrates resistance across multiple β-lactam classes. This fragmented framework fails to capture the multidrug-resistant (MDR) phenotype comprehensively and may mislead empirical treatment decisions. Co-resistance, defined as the concurrent resistance to multiple antibiotic classes within a single organism, is frequently driven by co-localization of resistance determinants on mobile genetic elements such as plasmids, or by co-selection under sustained antibiotic pressure, and identifying which resistance phenotypes cluster together enables clinicians to more accurately predict treatment challenges and supports antibiotic stewardship programs by discouraging empirical selection of agents likely to fail concurrently. To uncover these co-resistance networks, statistical and visualization tools, particularly heatmaps and pairwise phi-correlation matrices, have demonstrated considerable utility, allowing detection of clusters of co-resistant antibiotics and providing a visually intuitive representation of associations that would remain obscured in standard antibiogram reporting. This study is a retrospective observational study of Gram-positive bacterial isolates recovered from adult patients at An-Najah National University Hospital (NNUH), a tertiary care center in Palestine, between January 2021 and May 2025. The study population included all Gram-positive bacterial isolates obtained from adult participants (≥18 years) processed at the Microbiology Laboratory of NNUH during the study period, from clinical sources including blood, urine, respiratory specimens, wound swabs, tissue specimens, and other body fluid specimens, with eligible organisms comprising Staphylococcus aureus (including MRSA), Staphylococcus epidermidis, Enterococcus faecalis, Enterococcus faecium (including vancomycin-resistant strains), and other coagulase-negative staphylococci. Records were excluded if they involved Gram-negative organisms, Gram-positive species outside the predefined scope, patients under 18 years, specimens collected outside the designated study period, records lacking sufficient AST data (>10% missing required variables), polymicrobial cultures, follow-up cultures from the same infectious episode, patients with documented systemic antibiotic use more than three months prior to specimen collection, or duplicate isolates from the same patient, with only one non-duplicate index isolate retained per participant. AST was performed in accordance with CLSI guidelines, and data were retrieved from the hospital’s electronic laboratory information system; extracted variables included patient age, sex, specimen type, ward location, bacterial species identification, antimicrobial susceptibility results, comorbid conditions, prior antibiotic exposure, invasive procedure history, and immunosuppressant use. Resistance profiles were analyzed using pairwise phi-correlation matrices and visualized through hierarchical clustering heatmaps, MDR was defined as non-susceptibility to at least one agent in three or more antimicrobial classes, and univariable and multivariable logistic regression analyses were performed using R version 4.5.1 to identify predictors of MDR. This study aims to characterize the patterns of antibiotic co-resistance among Gram-positive bacterial isolates at NNUH through heatmap-based visualization and phi-correlation network analysis, and to identify independent clinical and demographic predictors of multidrug resistance using multivariable logistic regression. The findings are intended to inform empirical prescribing decisions, strengthen local antibiotic stewardship initiatives, and contribute evidence on Gram-positive resistance dynamics from a tertiary care setting in Palestine, a region where AMR surveillance data remain limited.