Pneumococcal conjugate vaccines and hospitalization of children for pneumonia: a time-series analysis, South Africa, 2006–2014

Abstract Objective To assess the impact of immunization with pneumococcal conjugate vaccines on all-cause pneumonia hospitalizations among children in Soweto, South Africa. Methods We used data collected at the Chris Hani Baragwanath Hospital in Soweto between 2006 and 2014 – i.e. before and after April 2009, when a pneumococcal conjugate vaccine was first included in South Africa’s routine immunization programme. Using a Bayesian generalized seasonal autoregressive moving-average model and the data collected in 2006–2008, we estimated the numbers of children that would have been hospitalized for pneumonia between 2010 and 2014 if no pneumococcal conjugate vaccines had been used. These estimates were then compared with the corresponding numbers of hospitalizations observed. Findings Between 2006 and 2014, 26 778 children younger than five years – including 3388 known to be infected with human immunodeficiency virus (HIV) – were admitted to the study hospital for pneumonia. We estimated that, for the children known to be infected with HIV and for the other children, pneumococcal conjugate vaccines reduced the numbers of hospitalizations for pneumonia in 2014 by 33% (50% credible interval, CrI: 6 to 52) and 39% (50% CrI: 24 to 50), respectively. In the study hospital in 2012–2014, as a result of immunizations with these vaccines, there were an estimated 3100 fewer pneumonia hospitalizations of children younger than five years. Conclusion In our study hospital, following the introduction of pneumococcal conjugate vaccines into the national immunization programme, there were significant reductions in pneumonia hospitalizations among children.


Introduction
Among children younger than five years, Streptococcus pneumoniae is the leading cause of bacterial pneumonia and such pneumonia caused an estimated 411 000 deaths in 2010 and 335 000 deaths in 2015 globally. 1 Although Africa has only 23% of the world's children younger than five years, it accounts for approximately 43% of the deaths in this age group attributed to bacterial pneumonia. 2 By June 2016, as part of the fight against bacterial pneumonia, pneumococcal conjugate vaccines were included in many national immunization programmes, including those of 24 low-income and 33 lower-middle-income countries. 3 The public health impact of seven-, 10-or 13-valent pneumococcal conjugate vaccines against all-cause pneumonia has been investigated in several high-income countries and some middle-income Latin American countries. In these studies, the temporal reduction seen in all-cause pneumonia following childhood immunization with a pneumococcal conjugate vaccine has varied from 0 to 77%. [4][5][6][7][8][9] In Malawi, the impact of a 13-valent vaccine (PCV13) on clinically diagnosed severe or very severe pneumonia has recently been evaluated. In this investigation, however, the study period began after introduction of the vaccine, ran for only 2.5 years and covered a time when only about 50% of the study population were estimated to be receiving all of the scheduled doses of the vaccine. 10 In addition, there was no attempt to see if impact of the immunizations in Malawi was affected by human immunodeficiency virus (HIV) status. In the present study, we used data collected at a single hospital in the South African township of Soweto over an eight-year period centred on 2009 -the year a pneumococcal conjugate vaccine was first included in South Africa's routine immunization programme. We used a time-series analysis and a Bayesian model to investigate the apparent impact of infant immunizations with pneumococcal conjugate vaccines on hospitalizations for pneumonia among HIV-uninfected and HIV-infected children younger than five years.

Study setting
In April 2009, a seven-valent vaccine (PCV7; Prevnar®; Wyeth Vaccines, New York, USA) became the first pneumococcal conjugate vaccine to be introduced into the South African public immunization programme. 11 Only infants beginning their routine childhood vaccinations were offered this vaccine, at six, 14 and 40 weeks of age, and there was no catch-up campaign among older children. In May 2011, however, the seven-valent vaccine was replaced with PCV13 (Prevnar13®; Pfizer Vaccines, Pearl River, USA) and a limited catch-up campaign, for children under 30 months of age, was launched. Although there are limitations in the use of administrative data to estimate vaccine coverage, such data indicate that national coverage with a third dose of pneumococcal conjugate vaccine among children aged nine months of age was 10.4% in 2009, 64.3% in 2010, 89.8% in 2011 and 99.0% in 2012. 12 Objective To assess the impact of immunization with pneumococcal conjugate vaccines on all-cause pneumonia hospitalizations among children in Soweto, South Africa. Methods We used data collected at the Chris Hani Baragwanath Hospital in Soweto between 2006 and 2014 -i.e. before and after April 2009, when a pneumococcal conjugate vaccine was first included in South Africa's routine immunization programme. Using a Bayesian generalized seasonal autoregressive moving-average model and the data collected in 2006-2008, we estimated the numbers of children that would have been hospitalized for pneumonia between 2010 and 2014 if no pneumococcal conjugate vaccines had been used. These estimates were then compared with the corresponding numbers of hospitalizations observed. Findings Between 2006 and 2014, 26 778 children younger than five years -including 3388 known to be infected with human immunodeficiency virus (HIV) -were admitted to the study hospital for pneumonia. We estimated that, for the children known to be infected with HIV and for the other children, pneumococcal conjugate vaccines reduced the numbers of hospitalizations for pneumonia in 2014 by 33% (50% credible interval, CrI: 6 to 52) and 39% (50% CrI: 24 to 50), respectively. In the study hospital in 2012-2014, as a result of immunizations with these vaccines, there were an estimated 3100 fewer pneumonia hospitalizations of children younger than five years. Conclusion In our study hospital, following the introduction of pneumococcal conjugate vaccines into the national immunization programme, there were significant reductions in pneumonia hospitalizations among children.

Study facility
The data we analysed came from the Chris Hani Baragwanath Academic Hospital, which was the only public hospital in the South African township of Soweto during our study period. Soweto is a low-income area of the city of Johannesburg, in Guateng province, where unemployment among adults is 53% and 40% of households have a daily income below two United States dollars. 13,14 Further detail of Soweto's socioeconomic status and health-care infrastructure is available from the corresponding author.

Data source and management
At the study hospital, we established an electronic database covering every patient younger than 14 years who was admitted to a medical ward between 1 January 2006 and 31 December 2014.

HIV status
Children aged at least 18 months were considered to be HIV-infected and HIV-uninfected if they had been found positive and negative, respectively, when tested using an enzyme-linked immunosorbent assay (ELISA) or polymerase chain reaction (PCR) assay. Children aged at least nine months but younger than 18 months were considered to be HIV-infected if found positive in one PCR assay or two ELISAs and HIVuninfected if found negative in one ELISA or PCR assay. Younger children were considered HIV-infected and HIVuninfected if found positive and negative, respectively, in one PCR assay. If a child was hospitalized more than once, an HIV test performed during any one of the hospitalizations was used to assign HIV status to the child -unless the child's status changed from HIV-uninfected to HIV-infected.
To estimate the age-specific numbers of HIV cases in the whole of Soweto, we used an age pyramid for the township, provided by the Gauteng Department of Health and Social Development, 16 and estimates of the age-specific prevalences of HIV infection in Gauteng -provided by the Actuarial Society of South Africa's AIDS and Demographic model. 17

Statistics
We used the Gauteng Department of Health and Social Development's data on the Soweto population (available from the corresponding author) to convert the records of pneumonia admissions at the study hospital to age-and HIV-status-specific annual and monthly incidences of such hospitalizations per 1000 individuals of the same age group and HIV status. To assess the impact of pneumococcal conjugate vaccines on these incidences, we divided our study period into a prevaccine era (2006-2008), a PCV7 era (2010-2011) and a PCV13-era (2012-2014).
By fitting Bayesian generalized seasonal autoregressive integrated moving-average models to the data from the prevaccine era, we estimated what the incidences of all-cause pneumonia hospitalizations might have been, in the PCV7 and PCV13 eras, if there had been no immunizations with any pneumococcal conjugate vaccines. 18 We created separate models for each HIV status and each of four age groups: under three months; at least three months but under 12 months; at least 12 months but under two years; and at least two years but under five years. Each model included covariates to adjust for the proportion of HIV-infected children on antiretroviral therapy (ART), the influenza season and the number of bronchiolitis-associated admissions and a variable indicating the transition period between PCV7 use and PCV13 use (available from the corresponding author).
For a given period, we calculated the percentage reduction in all-cause pneumonia hospitalizations that prob-ably resulted from immunizations with a pneumococcal conjugate vaccine as: where ŷ t is the estimated number of hospitalizations for month t that would have occurred in the absence of immunizations with a pneumococcal conjugate vaccine and y t is the observed number of all-cause pneumonia hospitalizations for month t. For our analyses, the prevaccine, PCV7 and PCV13 eras were represented by months 1-36, 49-72 and 73-108, respectively.
The number of hospitalizations prevented was estimated as: where T* represents the months when the 25th percentile of the percentage reduction was greater than zero.
For the percentage reductions and numbers of hospitalizations prevented, we report medians and the middle 50% of the simulated distributions -i.e. the 50% credible intervals (CrI). We considered a percentage reduction to be statistically significant if the lower limit of the 50% CrI exceeded zero.
In a sensitivity analysis, we investigated the effect on our main findings of assuming that, in each study year, the prevalence of HIV infection in the children with unknown HIV infection status was that recorded in tests of children of the same age group (available from the corresponding author). In addition, we performed a trend analysis to assess if, over our study period, there had been any temporal changes in bronchiolitis hospitalizations -i.e. changes due to environmental, social or other factors not related to the introduction of pneumococcal conjugate vaccines that may have had similar impacts on pneumonia hospitalizations (available from the corresponding author).

Ethics
This observational study was approved by the Human Research Ethics Committee of the University of the Witwatersrand, which deemed informed consent unnecessary.

Without confirmed HIV infection
Among children known to be uninfected with HIV or with unknown HIV status, the observed annual incidences of pneumonia hospitalizations varied little in the pre-PCV era but spiked in 2009 -i.e. the year the first pneumococcal conjugate vaccine was introduced into South Africa's national immunization programme -and then declined (Table 1). Overall, pneumonia hospitalization incidences in such children -per 1000 children of the same age group -were significantly lower in 2010 than in 2009 and also significantly lower in 2012 than in 2011 ( Table 1). The pneumonia hospitalization incidences in the two youngest age groups -which always had higher incidences than the elder age groups -were also significantly lower in 2014 than in 2013 (Table 1; Fig. 1).
When we compared the observed data with the results of our modelling, we found that -among the children known to be uninfected with HIV or with unknown HIV status -the observed number of pneumonia hospitalizations in 2010 was 9% (50% CrI: 6 to 12) lower than expected if there had been no immunizations with PCV7. This reduction represents two admissions averted per 1000 children younger than five years who did not have confirmed HIV infection ( Table 2; Fig. 2). No such reduction was observed in 2011 (Table 2). However, in each year following the transition to PCV13 in 2011, the immunizations appear to have brought a significant reduction in observed pneu-monia hospitalizations. These reductions ranged from 27% (50% CrI: 13 to 39) in 2013 to 39% (50% CrI: 24 to 50) in 2014 ( Table 2; Fig. 2). According to our models, in the PCV13 era, immunizations with PCV13 -and/or previous immunizations with PCV7 -averted seven to nine pneumonia hospitalizations per 1000 children younger than five years who did not have confirmed HIV infection (Fig. 2) -i.e. the immunizations averted a total of 2892 (50% CrI: 1522 to 4484) such hospitalizations between 2012 and 2014.
In 2010, according to our models, children known to be uninfected with HIV or with unknown HIV status in all five of the age groups we investigated had significantly lower incidences of pneumonia hospitalization than might have been expected had there been no immunizations. In that year, in each age group, such immunizations appear to have averted one to four admissions per 1000 children in the same age group who did not have confirmed HIV infection (Table 2; Fig. 2). Between 2011 and 2014, however, there were no corresponding significant reductions in two of the age groups -i.e. among children younger than three months or aged at least one year but younger two years. Significant reductions were observed in both 2012 (30%; 50% CrI: 9 to 46) and 2014 (39%; 50% CrI: 10 to 57) among the children who were aged at least three months but under 12 months and also in 2012, 2013 and 2014 (45−59%) for the eldest age group we considered (Table 2). In 2012 and 2014, the immunizations appear to have averted 12 (50% CrI: 3 to 24) and 13 (50% CrI: 2 to 28) pneumonia hospitalizations of children aged at least three months but under 12 months per 1000 children of the same age group who did not have confirmed HIV infection, respectively (Fig. 2). Among the children we investigated who were aged at least two years, the number of pneumonia hospitalizations averted during the whole PCV13 era was four to six per 1000 children of the same age group who did not have confirmed HIV infection (Fig. 2).
When we allocated the children with unknown HIV status to the HIVuninfected or HIV-infected groups -according to the corresponding prevalences of HIV infection recorded among tested children -we obtained a similar point estimate for the overall reduction in the incidence of pneumonia hospitalization among HIV-uninfected children younger than five years between 2010 and 2014: 41% (50% CrI: 26 to 53). When stratified by age group, the estimated reductions  The sum of those confirmed to be infected and a proportion -equivalent to the percentage of children in the same year who, when tested for HIV, were found positive -of the hospitalized children with unknown HIV status. c The sum of those confirmed to be uninfected and a proportion -equivalent to the percentage of children in the same year who, when tested for HIV, were found negative -of the hospitalized children with unknown HIV status.
(. . .continued) were also generally similar to those observed in the primary analysis (Table 2).

With confirmed HIV infection
Among children with known HIV infection, the highest incidence of allcause pneumonia hospitalization in the prevaccine era was recorded among the children aged at least three months but younger than 12 months, followed by those younger than three months ( Table 1). The incidence of pneumonia hospitalization among children with known HIV infection decreased during the prevaccine period, increased in 2009 -in all age groups -and then declined again (Table 1; Fig. 3). In the prevaccine (117 versus 23; P < 0·001), PCV7 (68 versus 23; P < 0·001) and PCV13 eras (40 versus 18; P < 0·001), the incidence of all-cause pneumonia hospitalization per 1000 was 2-to 5-fold higher in the children with known HIV infection than in the other children (available from the corresponding author).
In the children with known HIV infection, the annual incidence of pneumonia hospitalizations declined across the prevaccine era (Table 1; Fig. 3). The incidence of pneumonia hospitalization observed among such children in 2010 was similar to the modelled estimate -i.e. the value that might have been expected if there had been no immunizations with pneumococcal vaccine conjugates. The corresponding incidences observed in 2011 and 2014 were 11% (50% CrI: 6 to 16) and 33% (50% CrI: 6 to 52) lower than the corresponding modelled estimates. During the whole PCV13 era, such reductions represented the equivalent of 14 to 17 pneumonia hospitalizations averted per 1000 children younger than five years with known HIV infection (Fig. 2).
In terms of the post-2009 reductions in the incidence of pneumonia hospitalizations -among children with known HIV infection -that we tentatively attributed to PCV7 and/or PCV13, the five age groups we investigated gave varied results. Among children younger than three months, the observed incidence of pneumonia hospitalizations was significantly lower than the modelled estimate in both 2011 (20%; 50% CrI: 9 to 28) and 2013 (55%; 50% CrI: 27 to 72). Other significant reductions were recorded only in 2011 for the children aged at least three months but under 12 months (15%; 50% CrI: 5 to 22) and in 2013 for the children aged at least two years but less than five years (51%; 50% CrI: 3 to 71; Table 2).
In the sensitivity analysis, where some of the children with unknown HIV status were assumed to be HIV-infected, there was no consistent significant reduction in any age group from 2010 onwards ( Table 2). Although the median of the difference between the overall observed incidence of pneumonia hospitalization in 2013 and the corresponding modelled estimate was 46% (50% CrI: 23 to 62)mainly due to a relatively low incidence among the children aged at least two years but less than five years (Table 2)there was no such significant difference recorded for 2014.

Bronchiolitis hospitalizations
In children without known HIV infection from four of our study age groups, the annual incidence of bronchiolitis hospitalization did not change signifi- cantly across our study period (χ 2 test for trend; P > 0·05). The only children to show significant changes over time were those aged at least three months but younger than 12 months (available from the corresponding author). However, the incidences of bronchiolitis hospitalization observed in the PCV7 and PCV13 eras were still higher than the modelled estimates based on the incidences observed in the prevaccine era (available from the corresponding author).

Discussion
The introduction of PCV7 followed by PCV13 into the South African public immunization programme appears to be temporally associated with a reduction in all-cause pneumonia hospitalization of children younger than five years. We used mathematical models to estimate the number of pneumonia hospitalizations that might have been expected in the absence of any immunizations with pneumococcal conjugate vaccines -for those years when such immunizations occurred. These models indicated that, in Soweto in 2014, as a result of the introduction of PCV7 and/or PCV13, there were 39% fewer pneumonia hospitalizations of children who were not known to be HIV-infected. The immunizations appear to have averted an estimated 1116 (50% CrI: 559 to 1798) hospitalizations among Soweto children without confirmed HIV infection in 2014 alone. The corresponding values for children known to be HIV-infected were lower: a 33% (50% CrI: 6 to 52) reduction in the incidence of pneumonia hospitalizations resulting in 68 (50% CrI: 9 to 153) pneumonia hospitalizations being averted in Soweto.
Our findings for children without known HIV infection are consistent with observations from similar studies from Australia -which reported a corresponding reduction of 17 to 29% in children aged at least two years but younger than five years 19 -and the USA -which reported a reduction in all-cause pneumonia hospitalization in children younger than two years of about 39%. 20 However, the hospitalizations investigated in these studies all occurred after the introduction of PCV7 into public immunization programmes. In the present study the reduction in all-cause pneumonia hospitalization observed among Table 2. Hospitalization of South African children for pneumonia Alane Izu et al.
children without known HIV infection was relatively small while PCV7 was being used and only became significant following the implementation of PCV13. We estimated that, as the result of immunizations with PCV7 or, more likely, with PCV13, in Soweto between 2012 and 2014 -i.e. after PCV7 had been replaced with PCV13 -there had been a 32% decline in the incidence of all-cause pneumonia hospitalization among children younger than five years without known HIV infection. This decline exceeds the 7% reduction observed for lower respiratory tract infection in African efficacy trials of a nine-valent PCV9 -which included serotypes 1 and 5 but not the serotypes 3, 6A, 7F and 19A contained in the PCV13 product -and the reductions recorded with other pneumococcal conjugate vaccines elsewhere. 4,[20][21][22][23][24][25] Given the strength of its antibody responses to serotype 19F and its efficacy against the nasopharyngeal acquisition of this serotype, 26 PCV13 might have higher effectiveness against serotype 19F pneumonia than PCV7 or PCV9. A post-hoc analysis of data from the PCV9 efficacy trial in South Africa indicated that the vaccine had caused a 17% (95% CI: 7 to 26) reduction in all-cause clinical pneumonia hospitalization -the definition of which excluded bronchiolitis cases. 27 We used trends in the incidence of bronchiolitis hospitalization to investigate if the changes we observed, in the incidence of pneumonia hospitalization, might have been attributable to differences in infection with respiratory syncytial virus or epidemics of infection with other respiratory viruses during the study period. We also used bronchiolitis hospitalizations as an indirect measure of any temporal changes in hospitalization patterns for lower respiratory tract infection. The incidence of bronchiolitis hospitalization in the PCV13 era was similar to that in the prevaccine PCV except for a small reduction in the incidence among children aged at least three years but younger than 12 months -which was less than the decline observed in the incidence of pneumonia hospitalization in this age group.
It is unclear why the apparent beneficial impact that we observed for PCV7 and/or PCV13 against pneumonia was greater than that observed in PCV9 efficacy trials. However, it seems possible that the difference reflects the indirect protection offered by pneumococcal conjugate vaccines when they are used in general immunization programmes. After most children in a population have received such vaccines, both the vaccinated and unvaccinated have a lower risk of pneumococcal disease as a result of their reduced exposure to circulating vaccine serotypes. In Soweto there is biological evidence of such indirect benefits, at least for PCV13: a 68% reduction in PCV13-serotype colonization among children between 2010 and 2012 and a corresponding 66% reduction in vaccine serotype colonization among unvaccinated adults. 28 The prevalence of PCV13 serotype colonization is likely to have declined further between 2012 and 2014 with the ongoing use of PCV13. 29 The significant reduction we observed Note: The numbers were estimated in models based on the assumption that, if pneumococcal conjugate vaccines had not been used in the public immunization programme, the trends seen in hospitalization before the vaccines were introduced would continue. The plots show the values for the four age groups studied and for all of these groups combined.

Research
Hospitalization of South African children for pneumonia Alane Izu et al.
in pneumonia hospitalizations among infants younger than three months provides further evidence of the indirect benefits of PCV7 and/or PCV13 to Soweto -since such children should have received no doses or only one dose of pneumococcal conjugate vaccine. In South Africa, immunization with PCV7 and/or PCV13 also seemed to have indirect benefits when tested against invasive pneumococcal disease in adults and infants younger than 10 weeks. 30 The results of the present study indicate that, in Soweto at least, the additional serotypes included in PCV13 make this vaccine more effective against childhood bacterial pneumonia than PCV7. Immunizations with PCV7 only seem to have reduced pneumonia hospitalizations by less than 10%, if at all, in 2010 and 2011 but the reductions seen every year in the PCV13 era were considerably greater. In Israel, the incidence of hospitalization of children younger than five years because of community-acquired alveolar pneumonia was only reduced after PCV7 was replaced with PCV13. 31 Despite the greater apparent impact shown in our study compared with those seen in the PCV9 efficacy trials, we refrain from a detailed comparison of our results, which are based on so-called 2+1 schedule -i.e. of two primary doses plus a booster -with those of the efficacy trials based on a so-called 3+0 schedule -i.e. of three primary doses without a booster. A systematic review has indicated that, when used for routine immunizations against pneumonia in high-income countries, the 2+1 and 3+0 schedules have similar effectiveness. The immunizations we investigated probably offered both direct and indirect protection whereas the PCV9 efficacy trials primarily measured direct protection. 5 The interpretation of time-series analyses on the impact of pneumococcal conjugate vaccines against pneumonia in HIV-infected children is complicated by potential covariates and confounders. After having accounted for temporal changes in ART coverage among children with known HIV infection, we observed a 30% lower incidence of all-cause pneumonia hospitalization in HIV-infected children younger than five years in 2014, compared with the modelled estimates.
We did not, however, demonstrate any significant age-group-specific reductions in such incidence between consecutive years -perhaps because the incidence of pneumonia hospitalization among HIVinfected children was already declining in the prevaccine era because of increased coverage of ART and cotrimoxazole prophylaxis. It also seems possible that HIV-infected children have a less robust immune response to pneumococcal conjugate vaccines and are, in consequence, less protected against pneumonia by such vaccines. 22,32,33 Our findings appear to be the first from an analysis -of the temporal association of pneumonia hospitalizations with the introduction of pneumococcal conjugate vaccines into the public immunization programme of any African country -that includes prevaccine data. Our findings add to the increasing body of evidence indicating that routine immunization with pneumococcal conjugate vaccines prevents a range of disease outcomes when used on different schedules and in areas with high-or low-disease burdens. ■

Research
Hospitalization of South African children for pneumonia Alane Izu et al.