Pneumococcal conjugate vaccine coverage and indirect protection against invasive pneumococcal disease and pneumonia hospitalisations in Australia: a retrospective record linkage cohort study

Background: There is limited empiric evidence on the coverage of pneumococcal conjugate vaccines (PCV) required to generate substantial indirect protection. We investigate the association between population PCV coverage and indirect protection against invasive pneumococcal disease (IPD) and pneumonia hospitalisations among under-vaccinated Australian children. Methods: Birth and vaccination records, IPD notifications and hospitalisations were individually linked for children aged < five years, born between 2001 and 2012 in two Australian states (New South Wales and Western Australia; 1.37 million children). Using Poisson regression models, we examined the association between PCV coverage, in small geographical units, and the incidence of (1) 7-valent PCV (PCV7)-type IPD, (2) all-cause pneumonia and (3) pneumococcal and lobar pneumonia hospitalisation in under-vaccinated children. Under-vaccinated children received < two doses of PCV at < 12 months of age and no doses at [≥] 12 months of age. Potential confounding variables were selected for adjustment a priori with the assistance of a directed acyclic graph. The main limitations of this study include the potential for differential loss to follow-up, geographical misclassification of children (based on addressed at birth only) and unmeasured confounders. Findings There were strong inverse associations between PCV coverage and the incidence of PCV7-type IPD (adjusted incidence rate ratio [aIRR] 0.967, 95% CI 0.958-0.975, p-value <0.001), and pneumonia hospitalisations (all-cause pneumonia: aIRR 0.991 95% CI 0.990-0.994, p-value<0.001) among under-vaccinated children. Subgroup analyses for children < four months old, urban, rural and Indigenous populations showed similar trends, although effects were smaller for rural and Indigenous populations. Fifty-percent coverage of PCV7 among children < five years of age prevented up to 72.5% (95% CI 51.6-84.4) of PCV7-type IPD among under-vaccinated children, while 90% coverage prevented 95.2% (95% CI 89.4-97.8). Conclusions In this study we observed substantial indirect protection at low PCV coverage, challenging assumptions high vaccine coverage is required.


Introduction
Infections due to Streptococcus pneumoniae, the pneumococcus, are a leading cause of morbidity and mortality 32 among children globally [1]. Pneumococcal conjugate vaccines (PCVs) have been successful in reducing 33 pneumococcal disease through the direct protection of vaccinated individuals and indirect protection of both 34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint Indigenous children), experience significantly higher rates of morbidity across a range of infectious diseases 23 including pneumococcal disease, likely related to higher levels of social disadvantage [11,12]. Pre- PCV 24 introduction, invasive pneumococcal disease rates in Indigenous children from Central Australia, were some of 25 the highest in the world [13]. 26 In June 2001, 7-valent PCV (PCV7) was introduced for high risk children, including Indigenous children and 27 children with specified medical conditions, at ages two, four and six months. Children who were medically at 28 risk received a fourth dose of PCV7 at 12 months and a dose of 23-valent pneumococcal polysaccharide 29 vaccine (PPV23) at 5 years of age, while Indigenous children in high-incidence jurisdictions received PPV23 at 30 18-24 months of age. A three-dose PCV program (2, 4 and 6 months) became universal in January 2005, with 31 catch-up doses for children aged less than two years of age. In July 2011, PCV13 replaced PCV7 with a catch-32 up program funding a supplemental dose of PCV13 for children 12-35 months of age [14]. For Indigenous 33 children in high-incidence jurisdictions, a fourth dose of PCV13 replaced the PPV23 vaccine at 18-23 months. 34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06. 29.21259668 doi: medRxiv preprint Using the same large linked immunisation and hospitalisation dataset, indirect protection following PCV 1 introduction previously in Australia has been reported previously at the State level for children less than two 2 years of age, register with all-cause pneumonia declining among unvaccinated Indigenous (12% reduction; 95% 3 CI 3-25) and non-Indigenous children (45% reduction; 95% CI 41-49) [15]. 4 In this study, we extended this analysis by examining the association between PCV coverage in small 5 geographic units within two states for both incidence of invasive pneumococcal disease (IPD) and hospitalised 6 pneumonia among under-vaccinated children, in periods when coverage was changing rapidly. Our study 7 provides a unique opportunity to evaluate indirect protection against pneumococcal disease in the absence of 8 the widespread use of the booster dose making this analysis more applicable to many low-income countries 9 where, at the time of the study, the 3+0 schedule is primarily used [16]. 10

11
Study design 12 Our study design was a retrospective population-based cohort that used a subset of children under five years of 13 age from a dataset of all children born in New South Wales (NSW) and Western Australia (WA) between 2001 14 and 2012, a combined birth cohort of approximately 1.37 million. [17]. Birth records were probabilistically 15 linked (using name, date of birth, residential address and sex) to health data including vaccination register , IPD 16 notification data and hospitalisation data [17]. 17 Linked data sources 18 Vaccination status was obtained from the Australian Childhood Immunisation Register (ACIR), which includes 19 all children enrolled in the publicly funded healthcare system and comprises ~99% of children in Australia by 20 age 12 months [18]. IPD cases, defined as isolation of S. pneumoniae by culture or detection of nucleic acid 21 from a normally sterile, are notified as part of state-based passive surveillance systems [18]. Hospitalisation 22 data covered all inpatient separations (discharges, transfers and deaths) and included primary diagnosis, and up 23 to 50 (NSW) or 20 (WA) secondary diagnoses (coded using the Australian version of the International 24 Classification of Diseases [ICD-AM] coding system). IPD data was available from January 2001 onwards, 25 while hospitalisation data were available from July 2001 onwards. Demographic and health risk factor data 26 were obtained from state perinatal data collection and birth registries (S1 File, Table 1) [18]. See S1 File (Fig 1)  27 for a flow chart of the study cohort and data sources. 28

Study definitions
29 PCV7-type IPD was defined as IPD due to serotypes in PCV7 i.e. serotypes 4, 6B, 9V, 14, 18C, 19F and 23F. 30 PCV13, non-PCV7 type IPD, was defined as IPD due to the additional 6 serotypes in PCV13 i.e. serotypes 1, 3, 31 5, 6A, 7F and 19A. For all-cause pneumonia, we identified all hospitalisations with a pneumonia-related 32 diagnostic code in the principal or additional diagnosis fields. Pneumococcal or lobar pneumonia 33 hospitalisations were restricted to hospitalisations coded as either pneumococcal pneumonia (J13) or lobar 34 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint pneumonia (J18.1) (S1 File, Table 2). Inter-hospital transfers and admissions within 14 days of a previous 1 separation date were merged and classified as a single episode as per previous analyses [19]. All outcomes were 2 evaluated for children under five years of age. 3 We defined a child as vaccinated with PCV if they received an adequate number of doses to develop a 4 protective immune response against vaccine serotypes at least 14 days prior to onset of any study outcomes i.e. 5 two or more PCV doses administered at less than 12 months of age, or at least one PCV dose administered after 6 the age of 12 months [20]. Otherwise cases were classified as under-vaccinated. Given children who have 7 received one dose of PCV at less than 12 months of age may have partial protection, we have also completed 8 sensitivity analyses examining indirect effects among completely unvaccinated children (S1 File, Tables 14 and  9 15). 10 11 For descriptive analyses of PCV coverage over time, we calculated and graphed coverage among children less 12 than five years of age at three-monthly intervals (  was censored at the earliest of the following: death, when the child reached five years of age or at the end of the 30 study period. For the under-vaccinated group, person-time was also censored when the child was considered 31 vaccinated as previously defined. 32

Statistical analyses
To determine the association between PCV coverage and indirect effects against IPD and hospitalised 33 pneumonia (primary analysis), we first calculated the incidence of each disease outcome during the person-time 34 for which children were under-vaccinated up to five years of age. We divided each child's under-vaccinated 35 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint person-time into successive three-month intervals and linked the intervals to PCV coverage in their SLA of 1 residence at that same time period. SLA of residence was based on mother's residence at child's birth. Using 2 Poisson regression models, we examined the association between SLA-level PCV coverage and disease 3 incidence in under-vaccinated children. We used robust variance estimates to account for clustering, 4 accommodating multiple coverage estimates for each individual and checked linearity assumptions by plotting 5 the log of the disease rates against PCV coverage. 6 Analyses relating to PCV7-type IPD used the full study period from January 2001 and December 2012, while 7 analyses relating to PCV13, non-PCV7-type IPD were restricted to from January 2009 onwards -18 months 8 prior to PCV13 introduction. Analyses relating to pneumonia hospitalisations used data from July 2001 9 onwards due to lack of availability of hospitalisation data in NSW prior to this date. 10 Potential confounding variables were selected a priori using a directed acyclic graph (DAG), informed by 11 relevant literature and refined through expert consultation (S1 File, pp 6). These were age group, calendar year, 12 Indigenous status, season, Index of Relative Socio-economic Advantage and Disadvantage (IRSAD) score, 13 Accessibility or Remoteness Index of Australia (ARIA) category, birth weight, gestational age, maternal 14 smoking during pregnancy, number of previous pregnancies, and previous hospitalisation with an ICD code 15 corresponding to the presence of a medical condition increasing risk of IPD. 16 We graphed predicted incidence rates of each outcome among under-vaccinated children under five at each 17 decile of PCV coverage, accounting for the balance of covariates across all the individuals [25]. We also 18 calculated the estimated population preventable fraction of PCV7-type IPD and all-cause pneumonia 19 hospitalisations for each decile of PCV coverage, defined as the proportion of disease among under-vaccinated 20 children estimated to be preventable by increasing vaccine coverage among different age groups [26]. 21 We conducted subgroup analyses, examining PCV7-type IPD and all-cause pneumonia hospitalisations among 22 (1) children too young to be vaccinated (under four months of age); (2) under-vaccinated children in urban and 23 rural Australia; and (3) under-vaccinated Indigenous children. We expected that associations for these 24 subgroups may vary as the relationship between PCV coverage and pneumococcal disease is modulated by the 25 dynamics of pneumococcal transmission, which in turn differs with population density, age group, risk factors 26 and household structure [12,27]. 27 Statistical analyses were performed according to a prospective data analysis plan (see S1 Text, Protocol). is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. 233 person-years of follow-up among children under five years of age, 1,427 cases of IPD were notified: 43.2% 5 due to PCV7 serotypes, 28.4% due to PCV13-non-PCV7 serotypes, 17.7% due to non-PCV serotypes, and 6 10.7% lacked serotype data. Over the same period, 34 757 children experienced at least one episode of 7 pneumonia hospitalisation, of which 799 (2.3%) were coded as pneumococcal or lobar pneumonia. Children 8 were under-vaccinated against any PCV during 1 823 401 person-years of follow-up (31.2% of total person-9 years), with 83% of this time in children who had received no PCV doses. 10 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. 17.1 per 100 000 person-years in the universal PCV13 period (Table 1). Rates of PCV13, non-PCV7-type IPD 5 increased in the universal PCV7 period, before decreasing in the universal PCV13 period (Table 1). In the 6 universal PCV7 period, rates of PCV7-type IPD were higher in under-vaccinated children compared to all 7 children. Comparable trends were observed among different age groups and among both Indigenous and non-8 Indigenous children (S1 File, Table 4). Annual rates of PCV7 and PCV13, non PCV7-type IPD are presented 9 alongside annual PCV coverage estimates in S1 File, Tables 6 and 7. 10   (Table 2). Substantial reductions were observed for both 15 all-cause pneumonia and pneumococcal or lobar pneumonia across different age groups by vaccination period 16 (S1 File, Table 5). 17 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint Among children under five years, coverage of any PCV increased steeply to 40% reflecting a catch-up program 5 up to two years of age, then more gradually over the next four years as vaccinated children aged into older age 6 groups, reaching 86% by December 2012. is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021.
*Vaccinated defined as two or more PCV doses administered at < 12 months of age, or at least one PCV dose 1 administered after the age of 12 months 2 analysis at different levels of PCV coverage are provided in S1 File, Table 8. 12 For each percentage point increase in PCV coverage among children 0-59 months of age, the adjusted 13 incidence of PCV7-type IPD decreased by 3.3% (95% CI 2•5-4•2, <0.001) among under-vaccinated children 14 under five years of age (S1 File, Table 10). The steepest declines in PCV7-type IPD occurred between 0% and 15 50% coverage among children 0-59 months of age (Fig 4). 16 Additionally, there was a trend towards decreasing incidence of PCV13-type IPD among under-vaccinated 17 children as PCV13 coverage increased, however confidence intervals were wide (adjusted incidence rate ratio 18 [aIRR] 0.989; 95% CI 0.971-1.007; p=0.241). (Fig 4 and S1 File, Table 10). 19 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. Wales and Western Australia, 2001-12. 14 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021.   Similarly, there were inverse associations between PCV coverage and incidence of both all-cause pneumonia 10 and pneumococcal or lobar pneumonia hospitalisations among under-vaccinated children (Fig 6 and S1 File, 11 Table 11). The number of pneumonia hospitalisations and under-vaccinated person-time at the different levels 12 of PCV coverage are provided in S1 File, Table 9. For each percentage point increase in PCV coverage among 13 children less than five years of age, the adjusted incidence of pneumococcal or lobar pneumonia 14 hospitalisations decreased by approximately 1.5% (95% CI 0.7-2.6; p=0.001) (S1 File, Table 11). For each 15 percentage point increase in PCV coverage among children 0-59 months of age, the adjusted incidence of all-16 cause pneumonia hospitalisations decreased by approximately 0.9% (95% CI 0.6-1.0; p<0.001) (S1 File, Table  17 11). 18 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. Indigenous and rural subgroup analyses were wide as there were fewer cases and individuals in these subgroups 7 (S1 File, Table 13). 8 Table 4 shows the preventable fraction of all-cause pneumonia hospitalisations in under-vaccinated children 9 less than five years of age by each decile increase in PCV coverage. We estimate that vaccinating 50% of 10 children less than five years of age will prevent 33.3% (95% CI 27.3-38.8) of all-cause pneumonia 11 hospitalisations in under-vaccinated children less than five years, while vaccinating 90% prevents 51.7% (95% 12 CI 43.7-58.6). 13 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. [21]. By modelling the association, we were able to quantify degree of indirect effects at varying coverage 18 rates, facilitating future comparisons between studies with varying sample sizes and methods. Differences in 19 the relationship between PCV coverage and indirect effects between studies may also be due to differences in 20 the vaccine type studied (i.e. PCV7, PCV10 or PCV13) or context. 21 Importantly, we found similar trends among rural populations and Indigenous populations, although these 22 effects were smaller and had large confidence intervals which likely reflects the smaller sample size available 23 for analysis. These findings are important as they reflect indirect effects in different settings with different 24 patterns of social mixing and therefore patterns of pneumococcal disease transmission. The modelled incidence 25 of PCV7-type disease at zero coverage was lower than expected, given Indigenous children are known to be at 26 high risk of pneumococcal disease. This is likely due to the earlier introduction of PCV for Indigenous children 27 in 2001, which meant under-vaccinated Indigenous children were already benefiting from some degree of 28 indirect effects at the start of our study's observation period. Without sufficient baseline data prior to vaccine 29 introduction, our analyses likely underestimate the full extent of indirect effects among Indigenous children. 30 Given that Indigenous children make up 17.3% of our rural cohort, the earlier introduction of PCV among 31 Indigenous children may also have impacted on our estimates of indirect effects in rural settings. However, the 32 low baseline rates of disease may also be attributable to a more sparsely populated rural population with less 33 social mixing compared to urban settings [32].Previous studies on the direct effect of PCV, found the vaccine 34 to be equally effective among Indigenous children and children from rural areas, suggesting that our results are 35 not due to reduced vaccine effectiveness in these groups. . Evidence of indirect effects among children under 36 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint four months of age, albeit not statistically significant, is also important to note since these children are too 1 young to be fully vaccinated across the time period while vaccine eligibility and coverage changes for other age 2

groups. 3
Our analyses also demonstrated increasing indirect protection against pneumonia hospitalisations with rising 4 PCV coverage. Pneumonia comprises a substantially larger disease burden than IPD and is a crucial 5 determinant of vaccine cost-effectiveness of a PCV program. Our findings are consistent with previous analysis 6 using the same dataset, which found 38% and 28% reductions in all-cause pneumonia hospitalisations among 7 children <2 and 2-4 years of age, respectively, two years post-PCV7 introduction in Australia [33]. Similar 8 relationships have been reported for PCV uptake and pneumonia hospitalisations in Brazil [29]. While our 9 analyses, accounted for individual-level risk factors for pneumonia, it did not account for temporal trends, 10 including changes admission practices or pneumonia epidemiology, which could account for an overall decline 11 in pneumonia across this time period. This may explain the unexpectedly large estimate that over 50% of all-12 cause pneumonia among under-vaccinated children is preventable at 90% PCV coverage among children under 13 five. 14 Our analyses did not identify a clear threshold at which PCV7-type IPD was eliminated -with 5% of PCV7-15 type remaining at 90% coverage. Similarly, a prior study of the long-term impacts of PCV against IPD in 16 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted July 3, 2021. or children who were never registered at birth. SLA of residence was based on the mother's SLA of residence 1 at the time of the child's birth and we do not have updated data on SLA of residence if a family or child moved. 2 We acknowledge that SLAs will not perfectly capture the boundaries of communities which interact socially, in 3 which to measure indirect protection. As with all observational studies, causation cannot be assured. However, 4 we observed a "dose effect" of PCV-for each decile increase in coverage there was a decline in pneumococcal 5 disease and with the exception of parental smoking, the prevalence of measured confounders changed very little 6 over time. Our definition of under-vaccinated included children who received one dose of vaccine under 12 7 months of age who may benefit from some direct protection, however majority of under-vaccinated person-8 time was in children who had no doses (83%) and sensitivity analyses among completely unvaccinated children 9 yielded similar results [37,38]. We consider the risk of misclassification of vaccination status to be low, since 10 vaccination status was determined using an immunisation registry, which has previously been demonstrated to 11 underestimate coverage by less than 5% [39]. Lastly, an important limitation of this study is the potential for 12 loss to follow-up in children who have moved interstate (missing outcome data) or overseas (missing 13 vaccination and outcome data). From 2001 to 2012, the annual interstate out-migration of children less than 14 five years of age ranged from 1.6-2.2% in New South Wales and 1.4-2.1% in Western Australia [40]. Data 15 available from 2004-2012 indicates that overseas out-migration was less common, ranging from 0.9-1.2% [41]. 16 Since our analyses are for children up to five years of age, we estimate that the cumulative unobserved loss to 17 follow-up time to be less than 15% -in line with recommendations for cohort [42]. Furthermore, we do not [6]. Our findings suggest that lower rates of vaccine coverage may still be confer considerable indirect effects 32 and therefore a 1+1 schedule may be suitable despite low coverage PCV coverage. Additionally, we found 33 evidence that these trends are similar across a range of subgroups. However, our findings cannot be directly 34 extrapolated to other settings since degree of indirect protection is highly dependent on local factors influencing 35 pneumococcal transmission [27]. Other studies have demonstrated lower than expected direct and indirect 36 effects following PCV13 introduction in Australia, which used a 3+0 schedule, compared to countries using a 37 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted July 3, 2021. ; https://doi.org/10.1101/2021.06.29.21259668 doi: medRxiv preprint booster dose, such as the UK (2+1 schedule) and US (3+1 schedule) [16,46]. As a result, Australia 1 implemented a switch from the 3+0 to 2+1 schedule in 2018. Further research beyond the timeframes available 2 in this analysis is needed to document the longer-term impacts of PCV13 introduction and the schedule change 3 on indirect effects in Australia. Further research, using consistent and robust methodology, is also required to 4 understand the relationship between PCV coverage and indirect protection in a range of settings, particularly in 5 high-transmission settings. This will assist with understanding the role of PCV coverage as an alternative 6 metric to determine if indirect effects are present in settings with insufficient disease surveillance. 7 CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)