Analysis of the Trends of Methicillin-Resistant Staphylococcus aureus in Gauteng Public Hospitals from 2009 to 2018

ABSTRACT Most investigations into the distribution of methicillin resistant Staphylococcus aureus (MRSA) have focused exclusively on bloodborne infections within individual health care institutions for shorter time periods. This has limited the analysis of a community-spread pathogen to snapshots within the hospital domain. Therefore, in this study we determined the demographic and geographical patterns of MRSA infections and their fluctuation in 10 years within all public hospitals in Gauteng, South Africa. A retrospective analysis of S. aureus samples was done by deduplicating samples in two groups. The sample groups were placed into subsets with respect to demographic and geographical fields and compared across the studied period. Logistic regression was utilized to determine odds ratios for resistant infections in univariate and multivariable configurations. A total of 66,071 unique infectious events were identified from the 148,065 samples received over a 10-year period, out of which 14,356 were identified as bacteremia. MRSA bacteremia rates in Gauteng peaked in 2015 and have since decreased. Within Gauteng, metropolitan areas have the greatest burden of MRSA with children under 5 years of age and males being most affected. Medical wards have the highest S. aureus bacteremia rates, while intensive care units have the highest MRSA bacteremia rates. Patient age, admitting ward, and geographical district are the most important associated factors of resistance. MRSA acquisition rates have shown tremendous growth since 2009 but have since spiked and subsequently decreased. This may be due to the initiation of the National Guidelines on Antimicrobial Stewardship and Infectious Disease Surveillance. Further studies to determine the trajectory of infections are required to support these claims. IMPORTANCE S. aureus is the leading cause of a variety of devastating clinical conditions, including infective endocarditis, bacteremia, and pleuropulmonary infections. It is an important pathogen responsible for substantial morbidity and mortality. MRSA is a variant of interest originally responsible for difficult to treat hospital-acquired infections that has since achieved community spread throughout the world. Most investigations into the distribution of MRSA have focused exclusively on bloodborne infections within individual health care institutions for shorter periods. This has limited the analysis of a community-spread pathogen to snapshots within the hospital domain. This study sought to determine the demographic and geographical patterns of MRSA infections as well as how these have fluctuated over time within all public hospitals. This will also help in understanding the epidemiology and resistance trends of S. aureus, which will help clinicians to understand the clinical prospective and policy makers to design guidelines and strategies for treating such infections.

Demographics. Data analysis in this study revealed that males contract S. aureus infections at a greater rate than females, irrespective of the site of origin. The cohort properties representing age, sex, and ethnicity of the S. aureus infections are presented in Table 1.
General isolates had broadly similar age distributions between classes under a two-sided Kolmogorov-Smirnov test (D = 0.165; P = 0.073), whereas the distributions for bacteremia were significantly different (D = 0.289; P , 0.001). Adults bore the majority of sensitive bacteremia cases, 28.88% (n = 2,816), while most resistant bacteremia events occurred in patients under 5 years, 58.40% (n = 1,585) (Fig. 2). Among the patients under age five, the newborn group had the highest proportion of resistant bacteremia cases, 39.50% (n = 1,072). Age group categories had a moderate correlation with the type of infection for general isolates, V = 0.20 (95% CI, 0.19 to 0.21; P , 0.001). This correlation significantly increased for blood isolates, V = 0.33 (95% CI, 0.31 to 0.35; P , 0.001).
Geographical locations. Geographically, among the 11 different provinces of South Africa, Gauteng is the most densely populated province and has been identified with the highest number of monthly infections per year from 2009 to 2018 in Gauteng (Fig. 3).
Within the Gauteng Province, infections were substantially more common in the metropolitan municipalities (Johannesburg, Tshwane, and Ekurhuleni), with more than half the MRSA cases occurring in Johannesburg alone (MRSA isolates derived from any source [MRSAI], 54.5%; MRSA isolate subset from blood cultures [MRSAB], 55.2%) (Fig. 4). The cohort properties representing district and municipality locations of the S. aureus infections are presented in Table 2.
Despite this clustering, district-level geographical information demonstrated an overall weak correlation with infection resistance for both general isolates, V = 0.10 (95% CI, 0.09 to 0.10; P , 0.001) and blood isolates, V = 0.12 (95% CI, 0.11 to 0.13; P , 0.001). This did not improve when using municipalities, as this subdivided the regions with the lowest case numbers and did little to alter the relationship.
Ward distribution. S. aureus bacteremia was substantially more common in medical wards, accounting for 56.69% (95% CI, 55.90 to 57.50%) of positive blood culture isolates, significantly more than all other wards surveyed. This distribution did not differ based on resistance, as seen by the proportion of isolates being similar (t = 0.79, P = 0.426). The cohort properties representing ward distribution of the S. aureus infections are presented in Table 3.
Infection rates. Both MRSA and MSSA infection rates have been observed to change with time. The annual infection rates per 100,000 people are represented in Table 4. Factor analysis. Logistic modeling of isolate resistance revealed significant discrepancies between associated features of general versus blood isolates. The overall factor analysis, including age, sex, district, and ward, is presented in Fig. 6. Age groups were significant associates across all models, irrespective of confounders. The elder groups tended to have lower odds ratios (ORs) for bacteremia compared to general isolates, with newborns as a baseline.  This reflected the greater likelihood of acquiring resistant S. aureus bacteremia infections in earlier years of life. Sex associations demonstrated nonsignificant ORs except in the case of univariate bacteremia, but this resolved when confounders were considered for the multivariable case. District distribution interestingly showed a small but significantly increased propensity for resistant general isolates in Ekurhuleni compared to Johannesburg as the baseline. This occurred despite the district having a resistant bacteremia OR lower than that for Johannesburg. Tshwane, in comparison, exhibited an OR at least equivalent to the Johannesburg OR for resistant bacteremia but demonstrated a small but significant reduction in resistant general isolates. Ward type estimates were quite different between sample types. Considering confounders, Medical, Surgical, OB-GYN, and ICU wards, there were significantly positive associations of general isolate resistance, while Casualty and Oncology units were significantly negative associations. Pediatric units were significantly associated with general isolate resistance in the univariate case but dissipated in the multivariable variant.
For bacteremia, surgical and ICU wards were the only significant positive associations of resistance, while Casualty and Oncology units showed significant negative associations

DISCUSSION
Staphylococcus aureus is one of the successful human pathogens that can cause a wide range of infections owing to its extensive virulence factors (1). Although S. aureus is one of the well-studied pathogens, there are visible knowledge gaps in understanding this pathogen's epidemiology. Given this, we conducted one of the most extensive studies to investigate S. aureus infections in South Africa. This study is the first to examine infections from all levels of health care, including clinics and secondary and tertiary health care centers. All the collected and deduplicated S. aureus samples were considered general samples irrespective of acquisition source (blood, urine, swabs, etc.); about 22% were samples derived from blood cultures. This split into an overarching group and a bacteremia subgroup enabled broader analysis than is typically achieved through exclusive investigation of clinically significant cases. It can highlight all known resistant cases and the associated distribution across Gauteng. This further enabled an understanding of changing resistance patterns that may have preempted changes seen with only blood isolates. This may also have provided greater insight into the patient population susceptible to invasive resistant infections compared to the larger group, which may harbor MRSA without systemic involvement.
The minority of general isolates (13.79%) from this cohort were MRSA, which is concordant with the findings of Oosthuysen et al. (14), where only 15.3% of isolates obtained from any source at Tygerberg Hospital were found to be MRSA. However, many other studies in South Africa and globally have found much higher percentages of MRSA in comparison to MSSA when studying S. aureus isolates (8,15,16). Previous South African studies found the proportion of resistant isolates to be as high as 30 to 46%, which differs greatly from this study's finding of 13.79% and 18.91% for general and blood isolates, respectively (8,9,13). Diekema et al. (7) found that MRSA isolates accounted for 40% of S. aureus samples in an international study carried out across 45 countries. Fluctuations of both MRSA and MSSA have been reported by several studies, and it is evident that infection rates of both MSSA and MRSA generally mirror one another (17). This disparity is likely an aliasing phenomenon due to studies occurring at different phases of MRSA case growth, and this study examining a rather broad period. The MRSA proportion, for example, goes up to 34.05% of blood samples when considering only the peak in 2015, which is closer to the estimates of the studies around this period. This was not replicated for general isolates, as the proportion only increased to 17.21% for the same 2015 period. Other potential causes of this disparity could be because most of the previous studies were carried out only in hospitals (and more specifically, tertiary and academic hospitals). In contrast, the data set considered in this study included all samples from Gauteng Province processed by the NHLS. Rural district hospitals in South Africa may be more likely to empirically treat patients without waiting for blood cultures to be performed, due to poor access and long waiting times. This may explain why this study's figures and those of previous studies differ in terms of MRSA preponderance.  (7), much closer to the local peak described by Perovic et al. (8). Our study, which covered many more infection episodes than other local studies, found that the rise of MRSA infection was, in fact, later than the international peak, between 2015 and 2016, as mentioned above.
When taking a closer look at the demographic data obtained, a male predominance was found in MRSA and MSSA. This is similar to findings at Tygerberg Hospital (in Cape Town, South Africa), where 59.4% of those studied with S. aureus were male (14). Perovic et al. and Fortuin de Schmidt et al. have also shown male predominance in previous studies (8,13).
Although males were more likely to get both MRSA and MSSA found in general and blood isolates, the preponderance of MRSA in infected individuals was not different between sexes.
This study also reports that the highest percentage of MRSA bacteremia (39.5%) was among newborns, whereas newborns accounted for a much smaller proportion of MSSA bacteremia. These results are congruent with previous findings where children under 9 years old accounted for most S. aureus infections (2,8,13). Also, it has been observed that adults account for the highest number of general isolates in both MRSAI and MSSAI, which is supported by the previous findings of Shuping et al., who reported adults accounted for the highest percentage of MSSA infections (9). Notably, older adults did not demonstrate an increased proportion of infection. This contradicts previous findings, which showed that those at extremes of age had higher infection rates (8).
Johannesburg had more cases of both general and blood isolates of MRSA than other geographical locations, similar to previous studies (13). MSSA infections in this area were also higher than in other regions, although previous studies did not investigate this. Many infective episodes in Johannesburg could be because this is a highly population-dense area.
Patients who have not recently undergone surgery and are diagnosed with bacteremia are much more likely to be treated in a medical ward. This accounts for why patients treated in medical wards accounted for the highest preponderance of bacteremia. Fortuin de Schmidt et al. found that 60% of the patients presenting for S. aureus infections had one or more predisposing conditions (13). This may serve as another explanation for why bacteremia events were higher in medical wards. Patients may have been admitted for a different medical condition which could have predisposed them to S. aureus bacteremia. The subtype found most in surgical wards was general isolates of MSSA. This may be due to skin commensals infecting surgical sites. Skin colonization with S. aureus is most commonly associated with MSSA instead of MRSA, which is in keeping with this study's findings. Interestingly, surgical patients accounted for very few of the bacteremia cases. The physiological stress of surgery and surgical procedures would seemingly predispose patients to bacteremia. Prophylactic antibiotics before or during surgery may have minimized the risk of bacteremia. Alternatively, after surgery, patients with S. aureus bacteremia may have also presented with a surgical site infection that could have been sampled and cultured without blood cultures being performed. This would have been logged as a general isolate, not a blood isolate. On the other hand, more cases of MRSA bacteremia were seen in ICU and Pediatric wards than other subtypes of S. aureus. This may have been due to the immune status of these patients being such that it is vulnerable to bacteremia.
Conclusion. Multiple conclusions can be drawn from this research. MRSAB rates were increasing and peaked in Gauteng in 2015. Since then, they have started to decrease and maybe plateau, but further years need to be analyzed to determine if this is the case. The high number of cases in children under 5 years of age emphasizes the need to improve infection prevention and control measures in pediatric populations and pediatric wards in Gauteng hospitals. MRSA infections are still dominating the MSSA cases, and therefore use of antibiotic stewardship principles needs to be strongly recommended. The demographic, geographic, and site information is valuable, as it informs the areas in which changes should be implemented.
Overall, this study emphasizes the value and importance of ongoing surveillance of MRSA infections in Gauteng. Surveillance of antimicrobial usage for MRSA and MSSA infections is also necessary. Recommendations for further research include analyzing the trends of antibiotic resistance in this Gauteng population over time and continuing to model the MRSA trends in Gauteng in the years following this study.

MATERIALS AND METHODS
Study design. This study was carried out as a retrospective, descriptive analysis of the data of all S. aureus isolates obtained from patients at all public hospitals in Gauteng that were processed by the South African National Health Laboratory Service (NHLS) from 01 January 2009 to 31 December 2018.
Data collection. The data set collected contained all S. aureus isolates, either MSSA or MRSA, from any biological source received by the NHLS from January 2009 to December 2018 from any facility located within Gauteng, South Africa. Samples with missing or invalid patient details and samples without identification codes were excluded, as were any samples received that were collected outside this period. The data set Sources of bias. The data set was drawn from laboratory samples performed on clinician request. This may have resulted in underestimating the incidence due to unrecognized cases or those treated empirically without laboratory confirmation. Demographic and geographical data were largely transcribed from written records, which may have been incorrect or incomplete and could not be externally verified.
Age groups. Patients were divided into groups based on their ages following Medical Subject Headings (MeSH) definitions (18). Age groups were grouped for trend analysis into neonate, pediatric, adult, and elderly groups, as described in Table 5. The similarity between the distributions of patient ages across isolate classes was assessed using the Kolmogorov-Smirnov test.
Geographical distribution. Five districts and nine municipalities within Gauteng Province were identified and utilized to visualize the geographical distribution of cases via choropleth maps. The wards where isolates were collected were contacted to ascertain the associated medical discipline(s) and classified in a oneto-many fashion. Each ward could belong to multiple disciplines simultaneously, i.e., a pediatric medical ward would be allocated to both medicine and pediatrics. The wards were classified as Medical, Surgical, Pediatric, OB-GYN, ICU, Casualty, and Oncology.
Annual comparisons. Yearly average case figures were derived by determining each year's overall monthly average case rate and the associated 95% confidence intervals through bootstrapping the mean. Yearly comparisons were made using monthly isolate granularity. Incident figures were determined at the district and municipal levels via provincial population estimates derived from the 2011 South African census (19), with yearly changes in population estimated by national population growth figures.
Logistic regression was utilized to determine ORs for variables associated with drug-resistant infections. All variables were prespecified as informed by previous literature findings and included within both univariate and multivariable analyses. A cross-validated Lasso regression model was additionally employed for confirming the utility of the selected variables (20). Categorical variables were one-hot coded for model construction except for ward type. Owing to the multilabel nature of this field (e.g., a medical pediatric ward), the variable was full rank encoded to enable any combination of ward types to be addressed.
Statistical analysis. A chi-square and Cramer's V tests assessed evidence of the correlation between sample class and categorical variables. These provided a measure of significance and quantified the effect size associated with the correlation (21). Statistical significance was considered with an a of ,0.05. The quantification of effect size was necessary given the size of the data set, as standard chi-square analysis is substantially more likely to produce seemingly significant results with larger sample sizes irrespective of the proper relationship between variables (22). Cramer's V values were interpreted as shown in Table 6 (23). Comparisons between proportions were made using the Student's t test or Wilcoxon test, depending on the expected normality of the data to be compared.
Ethical considerations. This study was approved by the Human Research Ethics Committee (Medical) at the University of the Witwatersrand, Johannesburg (protocol number M190837). This study received a waiver of informed consent, as the analysis is retrospective and of minimal risk to participants. The data were collected during routine patient care; therefore, this study was unlikely to affect participants adversely. In addition, the Very Strong data received were anonymized and securely hosted via an authentication-based cloud hosting solution only accessible to research collaborators. The analysis of further anonymized samples provided only aggregate statistical information, with geographical data granularity limited to the subdistrict level.