Streptococcus pneumoniae Serotypes and Mortality in Adults and Adolescents in South Africa: Analysis of National Surveillance Data, 2003 - 2008

Background An association between pneumococcal serotypes and mortality has been suggested. We aimed to investigate this among individuals aged ≥15 years with invasive pneumococcal disease (IPD) in South Africa. Methods IPD cases were identified through national laboratory-based surveillance at 25 sites, pre-pneumococcal conjugate vaccine (PCV) introduction, from 2003–2008. We assessed the association between the 20 commonest serotypes and in-hospital mortality using logistic regression with serotype 4 (the third commonest serotype with intermediate case-fatality ratio (CFR)) as referent. Results Among 3953 IPD cases, CFR was 55% (641/1166) for meningitis and 23% (576/2484) for bacteremia (p<0.001). Serotype 19F had the highest CFR (48%, 100/207), followed by serotype 23F (39%, 99/252) and serotype 1 (38%, 246/651). On multivariable analysis, factors independently associated with mortality included serotype 1 (OR 1.9, 95%CI 1.1–3.5) and 19F (OR 2.9, 95%CI 1.4–6.1) vs. serotype 4; increasing age (25–44 years, OR 1.8, 95%CI 1.0–3.0; 45–64 years, OR 3.6, 95%CI 2.0–6.4; ≥65 years, OR 5.2, 95%CI 1.9–14.1; vs. 15–24 years); meningitis (OR 4.1, 95%CI 3.0–5.6) vs. bacteremic pneumonia; and HIV infection (OR1.7, 95%CI 1.0–2.8). On stratified multivariate analysis, serotype 19F was associated with increased mortality amongst bacteremic pneumococcal pneumonia cases, while no serotype was associated with increased mortality in meningitis cases. Conclusion Mortality was increased in HIV-infected individuals, which may be reduced by increased antiretroviral therapy availability. Serotypes associated with increased mortality are included in the 10-and-13-valent PCV and may become less common in adults due to indirect effects following routine infant immunization.


Introduction
Streptococcus pneumoniae is a common cause of pneumonia, meningitis and septicemia and is associated with substantial morbidity and mortality worldwide [1,2]. Since the introduction of pneumococcal conjugate vaccines (PCVs), a decline in the incidence of vaccine-serotype invasive pneumococcal disease (IPD) has been observed in vaccinated children and unvaccinated adults (through herd effect) in areas where the vaccine is widely used [3]. A significant burden of IPD, however, exists worldwide due to serotypes that are not included in PCV, which has increased in children and adults since the introduction of PCV into childhood immunization programs [4].
Incidence, severity and mortality of IPD are influenced by host-and organism-related factors. The host-related factors include extremes of age, underlying chronic illness, immunosuppression and access to antibiotic treatment [5][6][7][8]. Pathogen-related factors include the polysaccharide capsule which is the serotype determinant [9,10]. Previous studies have shown an association between pneumococcal serotypes and mortality in adults [11,12]. These associations could differ in a middle-income country, such as South Africa, where there is a high HIV prevalence (19% in adults aged 15-49 years in 2012) [13,14] and a greater diversity of serotypes associated with IPD [15]. In addition, few studies have examined the association between serotypes and mortality in the absence of potential effects of pneumococcal vaccination and none have examined the association between serotype and mortality separately in HIV-infected and HIV-uninfected individuals [12].
We aimed to determine the association between pneumococcal serotype and in-hospital mortality among patients aged !15 years with IPD in South Africa during the period 2003-2008, prior to the introduction of PCV in the routine immunization program.

Surveillance for IPD
From 2003, national, active, population-based, laboratory-based surveillance for IPD was conducted through the GERMS-SA programme [16]. Reports of laboratory-confirmed IPD, including demographic details as well as date of specimen, and source of isolate together with isolates were sent to the National Institute for Communicable Diseases (NICD) in Johannesburg from >130 laboratories nationally. We conducted annual laboratory audits of all publicsector laboratories in 8 provinces annually using Disa Ã Lab Laboratory Information Management System. We included cases identified by audit. At 25 sentinel enhanced surveillance hospitals in all nine provinces we collected additional information from patient interviews including HIV serological status, admission date, discharge diagnosis and outcome.
Cases were defined as patients !15 years of age with S. pneumoniae cultured from normally sterile site specimens (e.g., cerebrospinal fluid [CSF], blood, pleural, peritoneal or joint fluid) from January 2003 through December 2008. We excluded repeat isolates from the same individual within 21 days of the initial positive culture. We defined specimen source using a hierarchical definition, as follows: CSF specimen regardless of other specimens; blood specimen regardless of other specimens (excluding CSF); and other e.g., pleural fluid, joint fluid etc. without CSF or blood. We only obtained information on clinical syndrome for enhanced surveillance sites and we defined clinical syndrome using a hierarchical definition, as follows: meningitis if a clinical diagnosis of meningitis was noted in the medical records or the pneumococcus was isolated from CSF; bacteremic pneumonia if the clinical diagnosis of pneumonia was noted in the medical records and the pneumococcus was isolated from blood culture; other (including any diagnoses not including the preceding two, including bacteremia without any localising site and localized pneumonia with pneumococcus isolated from other sterile sites only). We defined predisposing conditions other than HIV infection as follows: asplenia, sickle cell anemia; chronic illnesses including lung, liver, renal, cardiac disease and diabetes; other immunocompromising conditions including malignancy, primary immunodeficiency, immunotherapy and organ transplant; as well as other Advisory Committee on Immunization Practices [17] risk factors including head injury with CSF leak, alcohol and smoking. Disease severity was measured using the Pitt bacteremia score, which is a composite severity score including patient temperature, presence of hypotension, receipt of mechanical ventilation, cardiac arrest and mental status [18,19]. We defined acute severe illness as a Pitt bacteremia score of !4 at the time of specimen collection. Provinces were grouped into 3 categories based on poverty rates (low, intermediate and high) determined by the findings of the living conditions survey conducted by Statistics South Africa in 2008-2009 [20].

Serotyping and susceptibility testing
The Quellung reaction using specific antisera (Statens Serum Institut, Copenhagen, Denmark) was used for pneumococcal serotyping. We separately distinguished serotypes 6A, B, C and D for the whole study period, although serotypes 6C and 6D were identified from <1% of cases [21]. Isolates were screened for penicillin resistance using oxacillin disc diffusion (Mast Diagnostics, Merseyside, United Kingdom) [22] and minimum inhibitory concentrations (MICs) of the potentially resistant isolates were determined using agar dilution or Etest 1 (AB-Biodisk, Solna, Sweden). We interpreted results as non-susceptible (intermediately resistant and resistant) or susceptible using 2008 Clinical and Laboratory Standards Institute (CLSI) definitions [22]. We considered isolates with MICs !0.12mg/L to be non-susceptible to penicillin at using the oral penicillin breakpoint. We defined multi-drug resistance as non-susceptibility to any three or more different antibiotic classes, according to the 2009 definitions of the CLSI. [22] For determination of appropriate antibiotic prescription, data on the antibiotics used for patient management was compared with the recommended antibiotic guidelines to determine whether the appropriate antibiotic was administered to the patient [23].
with one of the 20 most common serotypes, to allow sufficient numbers in each group for comparison. We a priori included the serotypes in PCV-13 i.e 1, 3, 4, 5, 6A, 6B, 7F, 9V, 14, 18C,  19A, 19F and 23F. In addition we included serotypes 8, 12F, 16, 9N, 22F, 25 and 13. Mortality was defined as death in hospital and within 30 days of the IPD episode, to avoid inclusion of deaths unrelated to IPD. To explore possible bias introduced by including only patients presenting to enhanced surveillance sites we compared the characteristics of patients presenting to enhanced and non-enhanced sites using logistic regression.
To assess differences between serotypes we evaluated the association of age (age group 15-24 years, 25-44 years, 45-64 years, !65 years), clinical syndrome (meningitis, bacteremic pneumonia and other IPD), HIV co-infection and in-hospital mortality for each serotype compared to serotype 4. We chose serotype 4 as the referent as it was the third most common serotype (289/3953, 7%) identified and we wanted to use a serotype with an intermediate CFR (29%) as referent to allow for identification of serotypes with particularly high or low CFR. We used univariate multinomial regression models, generating a separate estimate of effect for each predictor on each outcome relative to the base level. The effect measures are the ratios of two relative risks (relative risk ratios) with each relative risk describing the probability of the outcome in the category of interest relative to the baseline category [24].
To determine factors associated with in-hospital mortality, we used logistic regression. We assessed all variables that were significant at p<0.2 on univariate analysis, and eliminated from the model non-significant factors (p!0.05) with stepwise backward selection. Patients with missing data for included variables were excluded from the model. Serotype was retained in all multivariate models a priori because it was the main variable of interest and clinical syndrome because it is an important potential confounder. For the analysis of factors associated with mortality in HIV-infected individuals CD4 + T cell count was not included in the multivariable model because of a large amount (>40%) of missing data, however, CD4+ T cell count was not associated with serotype. All analyses were done using Stata Version 13 (StataCorp Limited). Two-sided p values <0.05 were considered significant. We performed stratified analyses for meningitis and bacteremic pneumonia separately and HIV-infected and-uninfected individuals separately, to explore whether the association between serotype and mortality varied within these subgroups. For each analysis, we used all available case information. Variables were binary (yes/no), defined as the presence or absence of the attribute excluding missing data, or categorical variables in multiple levels. We performed a sensitivity analysis restricting the study population to individuals aged !18 years.

Ethics Procedures
The national surveillance programme includes national laboratory-based surveillance where demographic data and isolates are sent to NICD routinely without requiring patient consent as part of the NICD's national public health surveillance responsibility. In addition, at sentinel enhanced surveillance sites, additional data is obtained from patient interviews, only from individuals who provide written informed consent. For participants under the age of 18, written informed consent was obtained from parents or legal guardians. The surveillance protocol (laboratory-based and enhanced surveillance) was approved by the University of the Witwatersrand Human Research Ethics Committee (Medical) (M0801117) and other relevant institutional ethics committees.
Comparing patients at enhanced to non-enhanced surveillance sites, on multivariable analysis, patients from enhanced sites were more likely to come from an intermediate poverty level province (odds ratio [OR] 1.3, 95% confidence interval [CI] 1.2-1.5) and less likely to come from a high poverty level province (OR 0.5, 95% CI 0.4-0.6)(vs low poverty province), more likely to be of non-black race ( Univariate multinomial analysis of the 20 commonest serotypes by age group, syndrome, in-hospital outcome and HIV co-infection Analyzing the 20 most common serotypes, using multinomial regression, comparing the age distribution of all other serotypes to the age distribution of serotype 4 (25-44 years as the referent group), serotype 3 was significantly more likely to be identified from patients 45-64 years of age (relative risk ratio [RRR] 1.9, 95% CI 1.2-2.8), serotype 1 was significantly more likely to be isolated from patients 15-24 years of age (RRR 1.8. 95% CI 1.2-2.8) and serotype 19F was significantly more likely to be isolated from patients 45-64 years of age (RRR 1.6 95% CI 1.1-2. 6) (Fig 2).

Factors associated with in-hospital mortality stratified by clinical syndrome
Among patients with bacteremic pneumonia, on multivariable analysis, compared to serotype 4, infection with serotype 14 (OR 2.2, 95% CI 1.1-4.7) and 19F (OR 3.5, 95% CI 1.6-8.0) was associated with increased in-hospital mortality (Table 3). Additional factors associated with in-

Discussion
We have described extremely high in-hospital mortality in individuals aged !15 years with IPD in South Africa, particularly for meningitis, where CFRs exceed 50%. Individual serotypes (1, 14 and 19F) were associated with increased mortality, even after controlling for confounders, suggesting pneumococcal capsule characteristics may be related to virulence. Other important factors associated with increased in-hospital mortality were increasing age and underlying HIV infection. All of the serotypes associated with increased mortality are included in PCV-10 and -13 and use of these vaccines in infants has reduced the burden of these serotypes in children as well as among adults through a herd effect [25,26].
As has been described in other studies, serotypes causing IPD varied by age group, HIV status and clinical syndrome, all important factors potentially associated with mortality [27,28]. For this reason, it is important to control for potential confounders when assessing the association between serotype and mortality. We found serotypes 1, 14 and 19F to be significantly associated with mortality, and particularly when presenting with bacteremia. Previous studies have found serotype 19F to be associated with increased CFR [11,12,29]. Serotype 14 was also associated with increased mortality in the current study, which has not been previously described. Serotype 1 was the commonest serotype in our study population and we found it to be associated with increased CFR. Some studies have found serotype 1 to be associated with decreased risk of death [11,12], although increased CFR as found in our study has been observed previously [30]. Previous studies have suggested that serotypes with increased relative risks of death had a higher carriage prevalence and low invasiveness [11]. Serotype 1 has been described as among the more invasive serotypes and serotype 19F among the less invasive serotypes [12,29,31,32].
Differences in the association between serotype and mortality in our study compared to previous studies could be partly as a result of the high HIV-prevalence (89% HIV prevalence amongst included IPD cases) in the population studied. Interestingly, when stratified by HIVinfection status, different serotypes were found to be associated with increased mortality in HIV-infected (serotype 19F) and HIV-uninfected (serotypes 12F, 16 and 9N) individuals. Serotype 1 was not found to be independently associated with increased mortality in either the HIV-infected or -uninfected group, although CFRs were elevated for this serotype in both HIV-infected and-uninfected individuals. Other factors associated with mortality also     differed between the two groups. In the HIV-infected subgroup, meningitis was associated with almost five times increased odds of death, but mortality was not significantly elevated in patients with meningitis in the HIV-uninfected group. The overall CFR was extremely high (55%) in patients with meningitis and elevated compared to individuals with bacteremic pneumonia. A study from Malawi, a country with a high population HIV prevalence, found 65% mortality in patients with meningitis compared to 20% for bacteremia [33]. In HIVinfected individuals consumption of antibiotics in the two months preceding hospitalization was predictive of elevated mortality; this likely reflected the severity of underlying HIV-related illness, rather than infection with antibiotic non-susceptible organisms, because strain susceptibility was not related to mortality on multivariable analysis. Residence in intermediate and high poverty level provinces was also associated with increased mortality in HIV infected individuals. This is possibly related to different specimen-taking practices [34], where in lessresourced provinces only the sickest individuals have specimens collected for culture; or due to worse access to care or quality of care, although we do not have data to support this hypothesis.
We found HIV-infected individuals to have an overall two times greater odds ratio of inhospital death compared to HIV-uninfected individuals. When stratified by syndrome, this association was only seen in patients with meningitis (who experience six times increased odds of death) but not in patients with bacteremic pneumonia. This difference in mortality association by clinical syndrome is similar to what was found in a study of IPD in patients aged <15 years from South Africa in the pre-PCV era [35]. Published data on the association between HIV infection and mortality vary, with some studies describing increased mortality in HIVinfected individuals and others not finding this association [36][37][38][39]. These varying findings could result from different populations studied including differing clinical syndromes included, different proportions of patients receiving ART as well as limited statistical power to detect an association in some studies.
Studies of the association between serotype and mortality have varied in the patient populations included, with some studies including only patients with bacteremic pneumonia and others including all IPD [11,29,40]. When stratified by clinical syndrome, amongst patients with bacteremic pneumococcal pneumonia, serotype 19F remained associated with increased mortality, while no serotype was associated with increased mortality in patients with meningitis. This suggests that the association between serotype and mortality may vary depending on the clinical syndrome evaluated.
Other factors associated with increased mortality included increasing age group, which has been described in several studies, most likely as a result of immunosenescence [41][42][43][44]. Notably, the numbers of cases aged !65 years was low, possibly as a result of reduced health seeking in this age group as well as the relatively younger population than many high income countries and the relatively low life expectancy in South Africa [45,46]. In contrast to previous studies, we did not find underlying illnesses (other than HIV) to be associated with increased in-hospital mortality. This could reflect poor diagnosis and documentation of underlying illness data in our setting or reflect the relatively smaller contribution of underlying illness to in-hospital outcome in this predominantly HIV-infected population [5][6][7].
Our study had some limitations. Only patients admitted to enhanced surveillance sites with viable isolates were included. Patients at enhanced sites differed from those presenting to nonenhanced sites with regard to a number of characteristics and this could potentially have introduced bias. Because of the nature of laboratory-based surveillance, only patients who presented to a healthcare facility and who had specimens taken could be included; these individuals may not be fully representative, especially in the poorer provinces. We examined a large number of serotypes individually, which may have reduced our power to detect significant associations, however we did have data on almost 4000 individuals and >1000 deaths. The ability to detect a significant association with mortality for different serotypes could have been affected by available numbers of cases allowing insufficient power to detect statistically significant associations in some cases. Some previous studies did not assess severity association for individual serotypes but rather grouped them, limiting the ability to compare with our findings [40]. We elected not to group serotypes for the mortality analysis as several different grouping have been used in the literature, and all groupings are somewhat arbitrary and could introduce spurious associations [11,29,40]. We included individuals aged !15 years in our study, although a cut-off of !18 years is more standard for adults. We did this because in South Africa 15 years is the age at which patients are generally admitted to adult rather than pediatric wards in South Africa. Importantly, individuals aged 15-17 years only made up 1% of the study population and study conclusions remained unchanged on sensitivity analysis excluding this group of individuals. CD4+ T cell count data was only available for 56% of HIV-infected patients and we therefore did not include it in the analysis of factors associated with mortality; CD4+ T cell count, however, was not associated with serotype and it is therefore unlikely to be a confounder of the association between serotype and mortality.
In conclusion, we have found that individual serotypes, included in PCV-13, were associated with increasing mortality in the pre-PCV era in South Africa, even controlling for potential confounding. HIV infection was also an important risk factor for death in our study, especially among patients with meningitis and older individuals. Studies of the association of pneumococcal serotype and mortality, following widespread PCV introduction and possible serotype replacement, may be informative.