Caracterização molecular e predição de resistência antimicrobiana em isolados Gram-negativos de pacientes hospitalizados em Governador Valadares, Brasil

Abstract

Context: In hospital settings, antibiotic-resistant bacterial strains pose a serious threat to patients and visitors, increasing hospital readmission rates, broad-spectrum antibiotic use, and mortality. The main mechanisms of resistance involve the expression of β-lactamases, changes in membrane permeability, and efflux pumps. Among these, β-lactamases have been increasingly detected since the 1980s and remain of great clinical importance. This study collected Gram-negative bacterial isolates over one year and investigated the presence of β- lactamase-encoding genes to build a predictive model of antibiotic resistance. Results: We collected 330 Gram-negative isolates, predominantly E. coli, K. pneumoniae, P. aeruginosa, and A. baumannii. Among them, 176 showed resistance to third- or fourth- generation cephalosporins, carbapenems, monobactams, or amoxicillin/clavulanic acid. The most common resistant profiles were to ceftazidime (84%), ceftriaxone (81%), amoxicillin/clavulanic acid (80%), cefepime (73%), and aztreonam (72%). The most frequently detected genes were blaCTX and blaTEM. K. pneumoniae was most often associated with these genes, representing 40.6% of blaKPC, 53.8% of blaCTX, 68.4% of blaSHV, and 45.4% of blaTEM. Phylogenetic analysis revealed high sequence similarity among isolates carrying the same gene, suggesting horizontal gene transfer. blaKPC was significantly associated with resistance to cefepime, ceftazidime, ertapenem, imipenem, and meropenem; blaCTX with aztreonam and ceftriaxone; and blaSHV with amoxicillin/clavulanic acid sensitivite. blaTEM showed no significant association. Tests with five machine learning algorithms indicated that Random Forest yielded the best performance, allowing the creation of a predictive model of resistance based on the presence or absence of these genes. Conclusion: K. pneumoniae was the primary species linked to resistance, with blaKPC being the most clinically impactful gene. Evidence of clonal persistence and horizontal gene transfer was observed. The machine learning model successfully predicted the resistance phenotype, suggesting that further refinement may enhance its utility for predicting antibiotic resistance in Gram-negative hospital isolates.