A hospital-based cross-sectional study was conducted from July through to September 2025 at the MCHH in the Kumasi Metropolis among patients referred for blood culture and susceptibility testing upon suspicion.

The study was carried out in MCHH in the Kumasi Metropolis among patients referred for blood culture and susceptibility testing coming to the laboratory unit of the hospital. The MCHH is centrally located in the Subin Sub-Metro within the Kumasi Metropolis, established in 1910. The facility provides a wide range of services, including Outpatient Department (OPD) care, admissions, antenatal care (ANC), deliveries, postnatal care (PNC), and laboratory services. Emergency services are also available, along with specialised clinics such as mental health, eye, and ear, nose, and throat (ENT). Additional diagnostic and treatment services include ultrasound scanning, management of sexually transmitted infections (STIs), prevention of mother-to-child transmission (PMTCT), antiretroviral therapy (ART), and physiotherapy. Public health services, family planning, cervical and breast cancer screening, as well as comprehensive theatre services are also offered. The hospital also serves as a referral centre for patients who need culture and susceptibility testing due to its well-resourced microbiology laboratory. According to the 2021 Population and Housing Census, the Kumasi Metropolis has a total population of 443, 981, representing 8.2% of the regional population. Of this figure, males account for 213, 662 (48.1%), while females number 230, 319 (51.9%) (Ghana Statistical Service, 2021).

Patients referred to the laboratory upon suspicion of blood stream infection were recruited for this study after informed consent and assent were obtained.

Patients with suspected bloodstream infections who were already receiving any form of antibiotic treatment were excluded from the study. In addition, suspected patients who declined to provide consent or assent were also excluded.

On average, about 150 patients visit the hospital each month for culture and susceptibility testing. Since sample collection took place over a three-month period, the total population of suspected patients was estimated to be 450. Using the Raosoft online calculator (Rapsoft, 2004), with 5% margin of error, 95% confidence interval, and 50% response distribution, the minimum sample size calculated was 208. However, 229 samples were collected at the end of study.

In this study, a convenience sampling technique was used to recruit participants. Patients who met the inclusion criteria were selected based on their availability and willingness to participate. This method was chosen because it allowed for easy access to participants within the study facility and was practical given the time and resource constraints of the study.

After written informed consent, and parental consent and the child’s assent (for those below 18 years) were obtained, 1.5mL and 8-10mL of blood samples were aseptically collected from paediatric and adult patients, respectively, into paediatric and adult blood culture bottles. The inoculated blood culture bottles were incubated in the BD BACTEC automated blood processor (Model: Fx40) for the initial twenty-four hours. After 24 hours, culture bottles that showed no indication of growth were incubated further for six days. For culture bottles that indicated growth after 24 hours, the samples were sub-cultured on Blood agar, MacConkey agar, and Chocolate agar and incubated for 24–48 hours. Blood agar and MacConkey agar plates were incubated aerobically at 37°C, while the inoculated Chocolate agar plates were incubated in a microaerophilic environment at 37°C.

Gram staining was performed on plates with visible growth to determine the appropriate blood culture vial for identification and antibiotic panels for susceptibility testing. For isolation and susceptibility testing Phoenix BD BACTEC Machine (Model: M50) was used. Quality control was done using in-house generated quality control organisms. Data were collected using a data collection sheet that included demographic characteristics such as age, sex, and ward. Results of bacterial identification and antimicrobial susceptibility testing were also documented.

Extended-spectrum β-lactamase (ESBL) and carbapenemase production among Gram-negative isolates were determined using the BD Phoenix M50 automated system, which applies Clinical and Laboratory Standards Institute (CLSI)-defined algorithms that compare antimicrobial Minimum Inhibitory Concentrations (MIC) patterns in the presence and absence of β-lactamase inhibitors (BD Phoenix). For ESBL detection, the Phoenix system evaluates reductions in MICs of cefotaxime, ceftriaxone, ceftazidime, and aztreonam when combined with clavulanic acid, and isolates flagged as ESBL-positive were reported in accordance with CLSI 2024 criteria (Lewis II et al., 2024). Carbapenemase production was similarly analysed using Phoenix-based confirmatory reactions that detect characteristic MIC elevations to ertapenem, imipenem, and meropenem indicative of serine carbapenemases or metallo-β-lactamases. For isolates showing carbapenem non-susceptibility, results were interpreted using CLSI 2024 breakpoints; no modified Hodge test or Carba NP test was required due to the platform’s validated detection algorithms. AST interpretive criteria strictly followed CLSI 2024 guidelines (Lewis II et al., 2024).

A comprehensive panel of antibiotics was selected based on CLSI guidelines (2024) and the routine therapeutic agents used for managing bloodstream infections in Ghanaian hospitals. The panel included β-lactams, carbapenems, aminoglycosides, fluoroquinolones, glycopeptides, lipopeptides, oxazolidinones, and other clinically relevant classes. Antibiotic choices were tailored to the Gram reaction and expected resistance mechanisms of the isolates. For Gram-negative bacteria, agents such as cephalosporins, β-lactam/β-lactamase inhibitor combinations, carbapenems, aminoglycosides, and fluoroquinolones were tested. For Gram-positive isolates, β-lactams, macrolides, lincosamides, glycopeptides, and oxazolidinones were included.

Intrinsic resistance was taken into account when interpreting results. For instance, Pseudomonas aeruginosa and Acinetobacter baumannii are intrinsically resistant to agents such as amoxicillin–clavulanate, cefazolin, and cefuroxime, and these were interpreted accordingly. The inclusion of mupirocin for Staphylococcus aureus was intended to assess resistance patterns relevant to MRSA surveillance rather than systemic therapy. Similarly, ertapenem was included to assess carbapenem susceptibility among Enterobacterales, with interpretation for non-fermenters limited to descriptive reporting.

Quality control was performed using in-house reference isolates with known susceptibility profiles, validated periodically against ATCC standard strains to ensure assay reliability.

Data were cleaned, coded, and validated for completeness using Microsoft Excel 365 before being exported into Stata version 15.0 (StataCorp, College Station, TX, USA) for statistical analysis. Descriptive statistics were computed to summarise the prevalence of bacterial isolates, with proportions and 95% confidence intervals presented. For calculating the overall percentages of ESBL- and carbapenemase-producing isolates, the total number of Gram-negative organisms identified during the study period was used as the denominator. Associations between sociodemographic/clinical characteristics and bacterial infection were analysed using a modified Poisson regression with robust variance to derive adjusted Prevalence Ratios (aPR) and 95% confidence intervals. This method was selected over standard logistic regression to provide a direct measure of risk association, as the high prevalence outcome (>60%) makes odds ratios less interpretable. Predictors of multidrug resistance (MDR), extensively drug-resistance (XDR), pan drug-resistance (PDR), and overall antibiotic resistance (ABR) were assessed using Firth’s penalised-likelihood logistic regression, reporting adjusted Odds Ratios (aOR) with 95% confidence intervals. This approach was implemented to address potential small-sample bias, particularly for the rarer XDR and PDR outcomes. The assumption of independence was upheld as only one isolate per patient was analysed. The discriminative performance of each resistance model was evaluated using Receiver Operating Characteristic (ROC) curves. Statistical significance was set at p < 0.05.

Ethical clearance was obtained from the Research Ethics Committee of the University of Health and Allied Sciences (REC-UHAS) with reference number UHAS-REC A.8[64] 24-25. An administrative approval was also obtained from the study facility to conduct the study. Written informed consent was also obtained from all participants aged 18 years and above who agreed to take part in the study. For participants below the legal age of 18 years, both informed parental consent and the child’s assent were obtained. All data were kept anonymous, and no personal or identifiable details were collected as part of the study. No third parties had access to any of the information provided. Confidentiality of all participants’ information and study data was strictly maintained in accordance with REC-UHAS guidelines.

The study analysed samples from 229 participants to determine the prevalence of bacterial infections across sociodemographic and clinical factors. Overall, 61.6% had bacterial infections, with the highest prevalence observed among children aged 6–10 years (67.4%) and those over 10 years (65.0%), though age differences were not statistically significant (p>0.05). Infection rates were similar between males (62.4%) and females (60.7%), showing no significant sex association (p=0.795). Notably, duration of hospital stay strongly predicted infection: children admitted for 5–7 days had almost three times higher risk of infection compared to those admitted for 2–4 days (cPR=2.94, 95% CI: 1.98–4.37; aPR=2.86, 95% CI: 1.92–4.27; p<0.001). Although infections were slightly more common in the paediatric ward (63.2%) than in the emergency ward (48.0%), this difference was not statistically significant (p=0.201). (Table 1).

Figure 1 shows that E. coli was the most frequently isolated organism, accounting for 22.7% of all isolates (95% CI: 16.1–30.5). This was followed by CoNS at 16.3% (95% CI: 10.6–23.5) and S. aureus at 14.9% (95% CI: 9.5–21.9). Klebsiella species were also common, representing 13.5% of isolates (95% CI: 8.3–20.3), while Pseudomonas species comprised 9.2% (95% CI: 5.0–15.2). Less frequent isolates included Enterobacter species and organisms grouped as “others, “ each contributing 6.4% (95% CI: 3.0–11.8), followed by Acinetobacter species (4.3%, 95% CI: 1.6–9.1). Rare isolates such as Serratia species (2.1%), Burkholderia gladioli, Streptococcus species, and Enterococcus species (each 1.4%) were also isolated.

Gram-negative isolates exhibited substantial resistance to cephalosporins, with E. coli showing 45.3% and 52.0% resistance to ceftriaxone and cefepime, respectively, and 54.2% resistance to levofloxacin. In contrast, amikacin demonstrated excellent activity, with only 3.5% of the organisms showing resistance to it. Among Gram-positive bacteria, staphylococcal isolates were particularly problematic, with CoNS showing high resistance to cefoxitin (60.6%), penicillin (50.0%), and oxacillin (60.6%), indicative of methicillin resistance. Notably, resistance to last-resort agents like vancomycin and daptomycin remained low at 9.2% and 4.3%, respectively. The β-lactam/β-lactamase inhibitor combination, piperacillin-tazobactam, retained good efficacy against most Gram-negative pathogens (16.3% overall resistance), whereas resistance to amoxicillin-clavulanate was considerably higher (46.1%). Table 2.

Table 3 shows analysis of antimicrobial resistance among the 141 bacterial isolates, with 81.6% classified as MDR, 8.5% as XDR, and 5.0% as PDR. Gram-negative pathogens demonstrated significant production of ESBL and carbapenemase, particularly in E. coli (56.3% ESBL) and Klebsiella species (36.8% ESBL, 21.1% carbapenemase). Notable pan-drug resistance was observed in Pseudomonas species (23.1%) and Enterobacter species (22.2%). Among Gram-positive isolates, both S. aureus (81.0% MDR) and CoNS (91.3% MDR) exhibited high rates of multidrug resistance, while nearly all isolates (96.5%) demonstrated resistance to at least one antibiotic.

The majority of isolates exhibited MDR (81.6%) and acquired resistance to at least one antibiotic class (ABR, 96.5%), while XDR and PDR categories were less common, occurring in 8.5% and 5.0% of cases, respectively. MDR was consistently high across all age groups, although those >10 years had a lower prevalence (65.4%) compared to infants and younger children (>80%). In contrast, XDR was significantly more frequent among participants >10 years (19.2%; aOR = 7.16, 95% CI: 1.02–15.07, p = 0.047). Table 4. Model performance of predictors associated with various categories of resistance.

Figure 2 shows the predictive performance of models for different resistance categories using ROC curves. The MDR and XDR models demonstrated comparable discriminative ability, each with an AUC of 0.7452, indicating moderate predictive accuracy. The PDR model performed best, achieving an AUC of 0.8513, reflecting strong discriminatory power. In contrast, the ABR model showed the lowest predictive performance with an AUC of 0.7250, suggesting only fair accuracy.

Even though sample collection and processing were performed under aseptic conditions, the high proportion of CoNS raises the possibility of contamination. This is particularly important because only a single blood culture was taken from each patient, limiting the ability to confirm whether the CoNS isolates represented true bacteraemia or skin flora contaminants. This may have led to an overestimation of CoNS as bloodstream pathogens. Additionally, molecular confirmation of isolates was not performed, and the three-month study duration does not account for seasonal variations in bloodstream infection epidemiology or antimicrobial resistance trends.