The retrospective study was conducted in Korle-Bu Teaching Hospital (KBTH), a 2000-bed tertiary teaching hospital with about 200 admissions per day 15 . The hospital covers all medical specialties and provides referral healthcare services to an estimated population of 24 million Ghanaians. The central outpatient department records about 29,757 patient turnout per month 15 . The bacteriology unit of the Microbiology Department of KBTH processes over 40,000 clinical cultures annually.

From January 2010 through December 2013, we studied all clinical bloodstream isolates of salmonellae collected at the bacteriology unit of the Central Microbiology Laboratory in KBTH. Two sampling approaches were used. First, all Salmonella isolates recovered from blood cultures of patients visiting the microbiology laboratory of KBTH were characterized and analysed. Second, we reviewed laboratory records of all patients with Salmonella BSI for microbiological data. Salmonella species were selected based on the following criteria: (i) confirmed as the causative agent of the infection for which blood cultures were performed, and (ii) identified as first isolate per patient within study period. Multiple isolates per patient were considered only if antibiogram was different than previous, and their bacteraemia episodes were more than 3 months separate.

To provide accurate information, patients and isolates data were abstracted in the following two steps, (i) manual work through of laboratory records, and (ii) physician-assisted medical reconciliation of data. Data were retrospectively reviewed and compared. When bacteraemic episodes with Salmonella were identified, we categorized those with typhoid and non-typhoid isolates. Data were retrospectively reviewed and compared. Subsequently, univariate and multivariate analysis were conducted to compare patients with Salmonella BSI caused by typhoidal strains to their counterparts with non-typhodal isolates using patients’ data from laboratory records as independent predictor variables. Similar analysis were also conducted to compare a second case group comprising patients with BSI caused by MDR Salmonella to those with non-MDR Salmonella BSI. Each patient was included only once for each outcome. Patients’ groups were compared regarding the following: demographics (age, gender), sickle cell disease, month and year of infection as well as patients’ hospital and assigned department of care. We also collected data on variables related to infections: types of Salmonella species and antibiogram.

During the study period, and as part of hospital routing care, 23,708 patients submitted blood cultures for bacteriological investigations. Speceimens were processed with BACTEC 9240 blood culture system (Becton Dickinson, NJ, USA) according to manufacturer’s instructions. A total of 2,768 blood cultures were positive for various infections. Subcultures were made on blood, chocolate and MacConkey agar; and were incubated aerobically at 37°C for 20 hours. Typical lactose non-fermenting colonies were speciated with standard bacteriologic reactions and BBL Crystal Enteric/Nonfermenter identification system® (Becton Dickinson, NJ, USA). Salmonellae were confirmed with slide agglutination using specific antisera (Himedia Laboratories, India).

The susceptibilities of isolates to the antimicrobial agents ampicillin (10 μg), cefuroxime (30 μg), cefotaxime (30 μg), chloramphenicol, (30 μg), cotrimoxazole (25 μg), gentamicin (30 μg), amikacin (30 μg), ciprofloxacin (5 μg), and levofloxacin (5 μg) (HiMedia Laboratories, India) were determined with Kirby-Bauer disc diffusion method in accordance with the Clinical and Laboratory Standards Institute (CLSI) guidelines 16 . The reference strain E. coli ATCC 25922 was included as quality control in the susceptibility assays. According to the international standard definitions for acquired resistance, and relative to the panel of antibiotics tested, multidrug resistant (MDR) phenotype was defined as in vitro non-susceptibility to ≥1 agent in ≥3 antimicrobial categories 17 : penicillins, cephalosporins, beta-lactamase inhibitor combinations, fluoroquinolones, aminoglycosides, chloramphenicol, folate pathway inhibitors, tetracyclines, macrolides and glycopeptides.

Data from laboratory investigations were captured into Microsoft ACCESS to generate a database, and exported into Statistical Package for Social Sciences (SPSS, Version 20.0) for data editing and statistical analyses. Continuous data were compared using student’s t-test, analyses of variance (ANOVA) or Mann–Whitney U test, (respectively for normalized and non-parametric distributions), with point estimates of statistical significance indicated by with 2 tailed P-values <0.05. Categorical data were compared across study parameters using Chi-square with Maracuilo’s post hoc tests for multiple comparisons, or the Fisher’s exact test. Correlations were assessed with Pearson coefficient (r) or Spearman’s rho (r_s) or their coefficient of determination (r² or r_s ²) where appropriate. Univariate comparisons were computed with Chi-square tests and unadjusted Odds ratios (OR) at 95% confidence interval (CI). From univariate analyses, variable with a P-value <0.05 were analysed in a multivariate logistic regression models to identify independent risk factors. Predictive accuracy of the models was evaluated by Hosmer and Lemeshow goodness-of-fit test with P-value >0.05 suggesting that the model predicts accurately on average. The area under the ROC (Receiver Operating Characteristic) Curve > 0.7 was used to analyse the discriminatory capability of bactereamia with typhoidal salmonella or MDR salmonellae versus their respective controls.

Ethical approval was not required as the study was regarded as part of routine surveillance measures for infection control by the Ethical and Protocol Review Committee of University of Ghana Medical School, College of Health Sciences. Considering the retrospective nature of the study, we could not obtain patients consent for use of clinical data. Nevertheless, on receipt of isolates and patients’ data, we de-identified all patients to ensure anonymity. We also allotted arbitrary numbers to all isolates assigned to the study.

During the study period, 23,708 blood cultures were submitted for bacteriological investigations. A total of 2,768 blood cultures were positive for various infections. From these, 181 (6.5%) non-duplicate salmonellae were recovered in 2010 (n = 37/749, 4.9%), 2011 (n = 53/701, 7.5%), 2012 (n = 63/796, 7.9%), and 2013 (n = 28/432, 6.5%). Thus 181 patients were included in this study (98 males and 83 females), with a mean age of 9.5 years (interquartile range, 6 years). The prevalence of TS and NTS were respectively 2.4% [n = 66/2768; 95% confidence interval (CI), 1.93-3.13] and 4.10% (n = 115/2768; 95% CI, 3.10-4.51); with the total number of TS being significantly lower than that of NTS (n = 66/181, 36.5% versus 115/181, 63.5%; P-value <0.002). Overall, we did not observe correlation between salmonellae prevalence and study years [r_s(2) = 0.4, P-value = 0.617]. Table 1 shows the distribution of Salmonella species across study years. Salmonella Enteritidis (n = 93/181, 51.1%), and S. Typhi (n = 48/181, 26.5%) were the predominant serotype. Others were S. Typhumurium (n = 22/181, 11.5%) and S. Paratyphi (n = 18/181, 9.9%). Marascuilo’s post hoc test was conducted to compare the proportions of salmonellae recovered across the 4 study years. Over the study period the yearly populations of each salmonella serotype had similar proportion of defects except for S. Enteritidis for which more isolates were recovered in 2013 compared to 2010 (P-value = 0.018), and in 2011 compared to 2013 (P-value = 0.019).

_aCrx res, cefuroxime resistant; _bP-value determined with Marascuilo’s comparisons post hoc tests for multiple proportions; ^c P, P-value for significant pairwise comparisons between study years;

*Differing subscripts within a column indicate significant difference at P-value < 0.05; SpearmansRanked Correlation between prevalence of Salmonella isolates and study years, r_s(2) = +0.4; P-value = 0.617.

As a crude measure to determine beta-lactamase producing phenotypes, we examined the resistance pattern of the salmonellae to cefuroxime (30 μg) (Table 1). The 4-year prevalence of cefuroxime resistance was similar among TS and NTS (n = 11/66, 16.7% versus n = 21/115, 18.2% respectively; P-value = 0.786). Marascuilo’s comparisons showed homogeneity in the proportion of cefuroxime resistant strains across the four yearly populations of typhoidal salmonellae (P-value = 0.1115). Conversely, although we recorded significant difference (P-value = 0.0161) in the proportion of cefuroxime resistance for the yearly populations of non-typhoidal salmonellae, no difference between any two particular years was significant (marascuilo’s post hoc > 0.05 for all pairwise comparisons). For TS and NTS, the Spearman’s ranked test showed no correlations between cefuroxime prevalence and study years [TS, r_s(2) = −0.6, P-value = 0.438; NTS, r_s(2) = −0.8; P-value = 0.231].

Figure 1 shows the monthly distribution of TS and NTS over a four-year study period. There was seasonality for TS, but not NTS, with rainfall pattern. The peak seasons for TS bactereamia were observed in March though July and from September through November; and coincided with the two rainy seasons in Ghana — April to July, and September to November (Figure 1). For instance, the highest TS prevalence peaks were observed in March (2.2%) and May (2.0%) for 2010; and in May (2.3%) for 2011. April (2.2%), May (2.75%) and August (2.2%) recorded the highest bactereamic episodes in 2012; whereas Sepetember (1.6%) and November (1.6%) accounted for the highest prevalence in 2013. Altogether, non-typhoidal salmonellae (NTS) appeared to be more fairly distributed across the months and over the years. For the particular years under study, we sought to examine the correlation between months and the number of salmonellae isolated. The number of TS recovered in 2013, unllike the other years, showed some significant degree of covariation with months; and the direction of covariation was positive. More so, 51.9% of the variance in the number of TS was coupled with advancing months (r_s = 0.799; r_s ² = 0.519; P-value = 0.001). On the contrary, no significant trend in correlation was observed between the number of non-typhoidal Salmonella (NTS) and months of isolation for any particular year.

The in vitro susceptibility results of salmonellae to potentially useful antimicrobial agents are shown in Table 2. Overall, ampicillin, chloramphenicol and tetracycline were the least effective antibiotics with over 70% of TS and NTS resistant to each antimicrobial. For TS, no resistance was found to cefotaxime and the ciprofloxacin; whereas gentamicin, amoxicillin/clavulanate and amikacin were respectively ineffective against 4.1% (n = 1/24), 8.3% (n = 1/12) and 9.1% (n = 1/11) of the isolates. Similarly, NTS were mostly susceptible to the ciprofloxacin (n = 134/135, 99.2%), amikacin (n = 70/71, 98.6%), cefotaxime (n = 93/99, 93.9%), and gentamicin (n = 84/93, 90.3%). The percentage resistance between TS and NTS were similar for all tested antibiotics except cotrimoxazole — with a smaller percentage of TS than NTS being resistant to this antimicrobial (TS, 15.5%, n = 27/174 versus NTS, 62.1%, n = 54/87; P-value = 0.002). Chloramphenicol was the only antibiotic to which salmonellae resistance appears to be increasing; we noted a positive correlation between time over 4 years and percentage of NTS resistant to this antimicrobial (r_s = +1, P-value = 0.0001).

^aTS, typhoidal salmonellae; ^bNTS, non-typhoidal salmonellae; ^cr_s, Spearman’s Correlaltion Coefficient, measures the overall significant increase or decrease in antibiotic resistance level over the 4-year period; ^dP, P-values at <0.05 are significant.

The number of salmonellae tested for susceptibility differ for each antibiotic. Results are thus recorded as proportions (number of resistant strains per total isolates tested for a particular antibiotic) with percentages in parenthesis.

Eighty-one (44.8%) salmonellae expressed multidrug resistant phenotypes. The multidrug resistant strains (MDRs) were significantly more in NTS (n = 59/115, 51.3%) compared to TS (n = 22/66, 33.3%); with cefuroxime resistant phenotypes more abundant in TS (n = 9/22, 40.9%) than in NTS (n = 18/59, 30.5%). Some MDRs (TS, n = 4/22, 18.2%; NTS, n = 11/59, 18.6%) remained resistant to all except one class of the antimicrobials tested. These were susceptible to gentamicin (TS, n = 3/4, 75%; NTS, n = 9/11, 81.8%), amikacin (TS, n = 4/4, 100%; NTS, n = 11/11, 100%), the ciprofloxacin (TS, n = 4/4, 100%; NTS, n = 6/6, 100%), and cefotaxime (TS, n = 4/4, 100%; n = 6/11, 54.5%).

In Figure 2, we examined the age-specific incidence of TS and NTS across study patients using locally weighted scatter plot smoothing (LOWESS) fit lines. Two observations are noteworthy. First, the results show similarities in the incidence of TS and NTS with age. Second, the occurrence of Salmonella bacteraemia show contrasting patterns for patients at extremes of ages. Specifically, TS and NTS bactereamias increased with age and were highest in children below 5 years, but declined gradually over ages 5 through 32, remaining fairly constant among the over 33 to 72-year-olds. A Pearson Product–moment Correlation was run to determine the relationship between incidence of Salmonella bactereamia and age. Whereas a negative correlation was observed between incidence of TS and age, only 21.5% of the incidence variance was accounted for by increases in patients’ age (r = −0.464; r² = 0.2153; P-value = 0.0297). In contrast, no demonstrable measure of covariation was recorded between NTS and age (r = −0.1795; r² = 0.0322; P-value = 0.2613). Overall, the highest number of culture positive cases for TS and NTS was observed in children 1 year of age (TS, n = 7/66, 10.6%; NTS, n = 20/115, 17.4%), and the least from amongst adults >33 years.

In Table 3, patients with typhoidal bactreamia were matched to those with non-typhoidal bactreamia to identify variables that increase the likelihood of developing TS BSI. Similar analysis were performed to determine factors associated with increased risk of Salmonella BSI by MDR salmonellae. Notable findings from univariate comparisons were that patients with typhoidal salmonellae were less likely to be children under 1 year old [Odds ratio (OR), 0.39; 95% confidence interval (CI), 0.18-0.84; P-value, 0.014]; and were less often infected with MDR strains (OR, 0.47; 95% CI, 0.25-0.89; P-value, 0.001). The MDR salmonellae were more frequently cefuroxime resistant (OR, 8.35; 95% CI, 3.02-23.05; P-value, 0.001). Patients with typhoidal bactereamia had been more often presumptively diagnosed with sepsis (OR, 1.92; 95% CI, 0.94-3.96; P-value, 0.072) or enteric fever (OR, 3.79; 95% CI, 1.18-12.09; P-value, 0.021). The MDR salmonellae were more often recovered from patients presumptively diagnosed with enteric fever (OR, 3.99; 95% CI, 1.05-15.05; P-value, 0.029). Non-MDRs were more frequently observed among patients suspected of sepsis (n = 39/64, 60.9% versus n = 23/67, 34.3%; P-vaue, 0.002). Table 4 shows risk factors from multivariate analysis for Salmonella BSI infection with typhoidal strains compared to non-typhoidal isolates and with MDRs compared to non-MDRs. In the end, among patients with Salmonella BSI, those < 1 year old had reduced risk of non-typhoidal infections [Odds ratio, 0.51; 95% confidence interval (95% CI), 0.16-0.92, P-value 0.021]. Similarly, patients with cefuroxime resistant strans were at increased risk of having MDR Salmonella BSI (OR, 8.97; 95% CI, 3.62-24.15; P-value, 0.001).

^aCI, confidence interval; ^bBSI, blood stream infections; MDR, multidrg resistant strains; ^cSD, standard deviation; ^dRecorded for 64 TS and 110 NTS, and 75 MDR and 96 non-MDRs; ^eJan-Mar, January to March; Sep,september; Oct-Dec, October to December; ^fRecorded for 47 TS and 85 NTS, and 67 MDRs and 64 non-MDRs; PUO, pyrexia of unknown origin; BEMP, bone infections, endocarditis, meningitis, pneumonia; ^gRecorded for 42 TS and 87 NTS, and 63 MDRs and 66 non-MDRs; NICU, neonatal intensive care unit; PML, Princess-Marie Louis Children hospital; ^hKBTH, Korle-Bu Teaching Hospital, recorded for 42 TS and 87 NTS, and 63 MDRs and 66 non-MDRs; ⁱCrx, cefuroxime, tested for 63 TS and 104 NTS, and 80 MDRs and 87 non-MDRs.

*P-vaue with Fisher’s exact test.

TS, typhoidal Salmonella; NTS,non-typhoidal salmonella; MDRs, multidrug resistant strains; Crx, cefuroxime; CI, confidence interval, BSI, blood stream infections.