Antibiotic susceptibility patterns of clinical isolates of salmonella species producing extended spectrum beta lactamases as predictor of multidrug resistance in a tertiary hospital, Southeastern Nigeria

Abstract

The global rise of multidrug-resistant (MDR) and extended-spectrum β-lactamase–producing (ESBL) Salmonella undermines treatment efficacy and threatens public health, particularly in low-resource settings. In Nigeria, data on resistance mechanisms in clinical isolates remain sparse. This study evaluates whether resistance genes and antibiogram profiles can reliably predict MDR and ESBL phenotypes to enhance early detection and surveillance. This cross-sectional, laboratory-based study was conducted from January-December 2024 and analyzed 265 clinical samples (241 faecal, 24 blood) from patients with suspected enteric fever at a Nigerian tertiary hospital. Sixty-five Salmonella isolates were identified via convenience sampling using standard microbiological methods and tested for antibiotic susceptibility using the Kirby–Bauer disk diffusion method, per CLSI guidelines. ESBL production was screened by Double Disc Synergy Test, and PCR assays were performed to detect blaTEM, blaSHV, tetA, qnrA/B, and sul1 genes. MDR was defined as resistance to ≥ 3 antibiotic classes. Statistical analyses included chi-square tests, logistic regression (α = 0.05), and machine learning models: Classification and Regression Trees (CART), and Random Forest. SHAP (Shapley Additive Explanations) was used for interpretability. Salmonella was isolated in 65 of 265 samples (26.9%), all from fecal specimens. Resistance was highest to amoxicillin/clavulanic acid (98.5%), tetracycline (96.9%), and sulfamethoxazole/trimethoprim (90.8%) while imipenem and polymyxin B remained effective with 96.9% and 95.4% susceptibility rate respectively. ESBL production was confirmed in 18 isolates (27.7%), while 28 (43.1%) met MDR criteria. The MDR rate was 89.2%, with a mean multiple antibiotic resistance index (MARI) of 0.52. BlaTEM (77.8%), tetA (72.2%), and sul1 (61.1%) were the most prevalent resistance genes. ESBL status was strongly associated with MDR (aOR:4.6; 95%CI:1.5–14.3; p < 0.01). CTX resistance, blaTEM, tetA, and sul1 demonstrated the strongest predictive power for MDR, with respective AUCs of 0.91, 0.88, 0.82, and 0.78. These markers consistently ranked highest across multiple predictive modelling approaches, with SHAP analysis confirming their dominant contribution to MDR classification. Resistance genes and antibiogram markers—particularly blaTEM and CTX resistance—predict MDR and ESBL status reliably. Leveraging these markers through machine learning—combined with SHAP-based interpretability—enables early, accurate detection and supports targeted antimicrobial interventions in resource-limited settings. Not applicable.