While the impact of pneumococcal conjugate vaccines (PCVs) on notified (laboratory confirmed) cases of invasive pneumococcal disease (IPD) has been demonstrated in multiple settings, including Australia,1–3 the true burden of IPD that is preventable by vaccination is largely unknown. This is because notified cases represent only the tip of the iceberg; cases need to first seek treatment, then have a blood culture collected, which in turn is positive, with the sensitivity of blood culture for detecting bacteremia estimated at around 31%.4
To obtain a more complete picture of the vaccine-preventable IPD incidence [absolute rate reduction (ARR) in IPD due to vaccination], 2 Finnish vaccine probe studies5 assessed the impact of the 10-valent PCV vaccine (PCV10) on nonnotified but clinically suspected IPD.4,6 One study was nested in a preprogram cluster randomized trial and the other compared cohorts, matched by age and season, before and after introduction of a universal 2 + 1 infant vaccination program. Using linked IPD notification and hospitalization records, they estimated that the ARR postvaccination for nonnotified but clinically suspected IPD or unspecified sepsis (based on hospital discharge diagnosis codes) was 2–3-fold greater than for notified IPD. In contrast, a study in England using different methods but similar hospital discharge diagnosis codes found a more than 2-fold increase in unspecified sepsis coded hospitalizations (unlinked to notification records) between the pre-PCV7 and post-PCV13 eras in children <2 years of age.7 Given such divergent findings and the variation between countries in diagnostic coding and criteria for hospital admission and obtaining blood cultures, it is important to see if the Finnish results are confirmed in other settings.
In Australia, IPD notification and hospitalization records have been linked for a population-based cohort of children born before and after the introduction of universal PCV7 and PCV13 vaccination programs.8,9 Using a similar vaccine probe design to the Finnish investigators, we aimed to estimate the absolute and relative impact of the introduction of PCV7 and PCV13 on the incidence of nonnotified but clinically suspected IPD and unspecified sepsis, and compare the results to those obtained for PCV10 in Finland.
We conducted a retrospective population-based study comparing rates of IPD-related outcomes in cohorts born before and after introduction of PCV7 and PCV13 vaccination programs in New South Wales (NSW) and Western Australia (WA). NSW and WA represent 42% of Australia’s population and include approximately 125,000 births annually.10
PCV7 and PCV13 Vaccine Programs in Australia
Australia introduced a publically funded PCV7 vaccination program for children at increased risk of IPD (including Aboriginal and Torres Strait Islander infants) in 2001. From January 2005, a 3 + 0 PCV7 schedule (primary course of three doses at 2, 4 and 6 months, with no booster dose unless children were in particular risk groups) was funded for all children, with a catch-up program for children <2 years of age.11 By the end of 2005, coverage with 3 doses by 12 months of age had increased to above 90%.12 In 2011, PCV7 was replaced with PCV13 and a supplementary PCV13 dose was also funded for children 12–35 months of age who had completed the primary PCV7 course.11
Assembly of the Study Data
A detailed description of the study datasets and linkage process are provided elsewhere.8,9 In brief, birth records for all singleton live births between 2002 and 2012 in NSW and WA having both a perinatal and birth registration record (97.5% of live births in the perinatal data) were probabilistically linked (using name, date of birth, residential address and sex) to the National Death Index and state-based infectious disease notification and hospitalization datasets, available until December 2013. Hospitalization data cover all inpatient separations (discharges, transfers and deaths) in each state and includes a primary diagnosis code (first-listed diagnosis), and up to 50 (NSW) or 20 (WA) secondary diagnosis codes [coded using the International Classification of Diseases version 10-Australian Modification (ICD-10-AM) coding system]. Notification databases include mandatory reports of laboratory-confirmed IPD cases (culture or detection by nucleic acid testing of Streptococcus pneumoniae from a normally sterile site), which have been notifiable in both states since 2001.
Before and After Vaccination Study Cohorts
The preuniversal program reference cohort was limited to children born from January 2002 to December 2004 with follow-up until December 2004, as complete IPD notification data were only available from 2002 and catch-up vaccination programs began in 2005. The 2 PCV7 eligible cohorts (PCV7-early and PCV7-late; Table 1) were matched for season and age/follow-up time to this 2002–2004 reference cohort, such that all children in the PCV7 evaluation cohorts were 0 to <36 months of age. The PCV13-eligible cohort was limited to children born from May 2011 (the first cohort to be eligible for the first dose of PCV13 at 2 months) to December 2012 (last year of births available for analysis) with follow-up ending in December 2013 (end of study follow-up). This cohort was matched for season and age/follow-up time to 2 reference cohorts; one born before universal vaccination and one eligible for the PCV7 universal program, such that all children in the PCV13 evaluation cohorts were 0 to <32 months of age (Table 1).
Three IPD-related outcomes were investigated:
- Notified IPD: The first notification per child of (laboratory confirmed) IPD.
- Nonnotified but clinically suspected IPD or unspecified sepsis: The first occurrence per child of a hospitalization coded with an ICD-10-AM code (as primary or secondary diagnosis) considered compatible with either IPD or unspecified sepsis (Table 2) that was not linked to an IPD notification. That is, the child either did not have an IPD notification or the date of the first hospitalization with the code(s) of interest was more than 90 days before the onset date of the notified IPD record. The ICD-10-AM codes included were the same as the ICD-10 codes used in the Finnish studies4,6 except we excluded B95.5 (unspecified Streptococcus as the cause of diseases classified elsewhere) and B95.3 (S. pneumoniae as the cause of diseases classified elsewhere) as these codes were predominantly related to infections in nonsterile sites such as urinary tract infection, acute bronchiolitis, or acute upper respiratory tract infection in our data.
- Nonnotified but clinically suspected IPD: This was a subgroup of the above outcome where the codes were restricted to those compatible with nonnotified IPD only: A40.3/G00.1/M00.1.
As an indication of the contribution each specific ICD-code made to the study outcomes, the distribution of the first-mentioned ICD-code per person for: (1) IPD-compatible and (2) unspecified sepsis without an IPD-compatible code, by whether or not they were linked to an IPD notification record was described for the largest prevaccination cohort (pre-PCV7), and the season- and age-matched PCV7-late cohort.
Incidence rates for each IPD-related outcome were calculated using person-time-at-risk as the denominator and were reported by calendar year for <1 and 1 to <3 year olds. The person-time for each child started at birth and was censored on the date of their first IPD-related outcome, death, or at the end the follow-up period for each cohort, which ever came first. Incidence rate ratios (and corresponding 95% CIs) comparing the reference groups and age- and season-matched vaccine-eligible cohorts were obtained using the Poisson regression models. Relative rate reductions (RRRs) were calculated as (1-incidence rate ratio) × 100%, and ARRs as the rate in each vaccine-eligible cohort minus the rate in the age- and season-matched reference group per 100,000 person-years. Rate comparisons and their 95% CIs were obtained using the stir command in STATA v14. Proportions were compared using Fisher exact test and P values <0.05 were considered statistically significant.
Ethical and Data Custodian Approvals
Approval was obtained from the relevant data custodians and the Department of Health WA Human Research Ethics Committee, the NSW Population and Health Services Research Ethics Committee, The WA Aboriginal Health Ethics Committee, the NSW Aboriginal Health and Medical Research Council Ethics Committee, The Australian Government Department of Health and Ageing Departmental Ethics Committee and the Australian Institute of Health and Welfare.
Description of ICD-coded and -notified Cases for Pre-PCV7 and PCV7 Late Cohorts
Of the 544 IPD notifications occurring in the pre-PCV7 and PCV7 late cohort combined, 511 (94%) had at least one linked hospitalization within 90 days of the recorded onset date, with almost all (482, 90%) occurring within 7 days of the recorded onset date. Of the 511 cases with a linked hospitalization, 247 (48%) were assigned either an IPD-compatible or unspecified sepsis code, with most (197, 80%) having an IPD-compatible code (Table 2). Of the 264 notifications with a linked hospitalization that did not have one of the codes of interest, by far the most common code was pneumonia due to S. pneumoniae (ICD code: J13; 22%). There were no significant differences in these proportions between the pre-PCV7 and PCV7 late cohorts (data not shown).
Nonnotified IPD and Unspecified Sepsis-coded Hospitalizations
Most (≥82%) cases with an IPD-compatible code were notified (Table 2). Therefore, of all cases of nonnotified IPD-compatible or unspecified sepsis, only 2.8% (n = 39) were coded as IPD-compatible. Sepsis due to S. pneumoniae was the most common IPD-compatible code (≥76%) in both notified and nonnotified IPD-compatible cases, with similar proportions in each cohort.
In contrast to cases with IPD-compatible codes, only 5% or fewer cases with a hospitalization coded as unspecified sepsis (without an IPD-compatible code) were notified (Table 2). In addition, the proportional distribution differed between notified and nonnotified cases and by cohort. A49.9 (bacterial infection, unspecified) and A41.9 (sepsis, unspecified organism) were the most common nonnotified unspecified sepsis codes. The only nonnotified unspecified sepsis code to show a decline in absolute numbers between the pre-PCV7 and PCV7 late cohorts was A49.9 (bacterial infection, unspecified).
Between 2002 and 2013, rates declined across all IPD-related outcomes examined, with the greatest rate reductions occurring between the preuniversal period and following introduction of the universal PCV7 program in 2005 (Fig. 1). Annual rates of nonnotified IPD or unspecified sepsis were at least 1.8 times higher in <1 year olds than in 1 to <3 year olds. In contrast, rates of notified IPD in <1 and 1 to <3 year olds were generally similar. Within each age group, rates tended to be higher for nonnotified IPD or unspecified sepsis compared with notified IPD, particularly for <1 year olds where annual rates were at least 2.4 times higher. Given most IPD-coded hospitalizations were notified, rates for nonnotified IPD were much lower, with fewer than 5 cases per year in either <1 year olds or 1 to <3 year olds in 2005–2013.
Impact of Vaccination
There were significant RRRs and ARRs in all outcomes following successive implementation of universal childhood PCV7 and PCV13 vaccination programs (Tables 3 and 4).
Nonnotified IPD Compared With Notified IPD
The RRRs for nonnotified IPD were similar to those for notified IPD for all comparisons. However, the ARRs were much smaller for nonnotified IPD due to the low baseline rates.
Nonnotified IPD or Unspecified Sepsis Compared With Notified IPD
As expected, the RRRs for the less specific outcome of nonnotified IPD or unspecified sepsis were smaller than the reductions for notified IPD, although still considerable at between 21% and 36% compared with pre-PCV7 cohorts. Point estimates were not significantly different, although generally lower, when the outcome was restricted to just nonnotified unspecified sepsis (Tables, Supplemental Digital Content 1 and 2, http://links.lww.com/INF/D449). Compared with the pre-PCV7 era, the estimated total PCV7 late- and PCV13-preventable incidences (the sum of the ARRs for notified IPD and nonnotified IPD or unspecified sepsis) were 90.3/100,000 and 123.3/100,000, respectively, which are 1.5 and 1.8 times higher than the ARRs for notified IPD alone.
To our knowledge, this is the first study to demonstrate the impact of PCV7 and PCV13 on nonnotified IPD and unspecified sepsis. In keeping with findings from Finnish studies,4,6 but in contrast to findings from an English study,7 we found significant relative and ARRs comparing prevaccination and postvaccination cohorts and an added impact of the change from PCV7 to PCV13. As expected, compared with prevaccination cohorts the RRRs for the more sensitive outcome of nonnotified IPD or sepsis were lower (21%–36%) than for the more specific outcomes of nonnotified and notified IPD (71%–91%). However, the higher incidence of nonnotified IPD or unspecified sepsis meant that the ARRs were substantial (31–54/100,000 person-years). When taken into account, they increased the total preventable incidence of IPD 1.5 to 1.8-fold above that estimated for notified IPD alone. While there are similarities, there are also a number of differences between our findings and those from Finland.6
While the most common nonnotified sepsis code in both our study and in Finland was “sepsis, unspecified organism” (A41.9), our pre-PCV7 cohort had a broader distribution of codes included than the Finnish reference cohorts.6 Furthermore, in contrast to the Finnish observational study,6 which found reductions in all ICD codes except A49.9 “bacterial infection, unspecified,” we only saw a decline in this code and no other unspecified sepsis codes between the pre-PCV7 and PCV7 late cohorts. This suggests differences in coding practices and highlights the need to make comparisons using broad diagnostic categories to account for these variations.
In support of the above comment, the RRRs overall were remarkably similar between the 2 countries for all outcomes measured. This is despite the fact that Australia started with PCV7 then moved to PCV13 (with catch-up programs initially in both instances), while Finland started with PCV10 (with no catch-up program). It also suggests that despite differing health systems and health seeking behaviors and referrals, the combination of codes selected for analysis are measuring similar outcomes. The relative reductions for notified cases are also consistent with vaccine effectiveness estimates we obtained, using the same linked data, against all cause IPD (80% for all-cause IPD)13 and changes in IPD epidemiology observed in surveillance data, which include a 65%–70% decline in IPD due to PCV13-non-PCV7 serotypes, and 74%–82% decline in all-cause IPD following the introduction of universal vaccination.1–3
While in Finland,6 the ARR for nonnotified IPD or unspecified sepsis was 122 per 100,000 person-years, our ARRs for nonnotified IPD or unspecified sepsis were considerably lower at 31–54 per 100,000 person-years. This is in contrast to the higher prevaccination incidence and absolute reduction observed for notified IPD in Australia compared with Finland. This could be explained by the fact that the prevaccination rate of nonnotified IPD or unspecified sepsis was considerably lower in our study and that there was a lower proportion with the more specific codes of nonnotified IPD (23% in Finland vs. <3% in our study).6 However, even if the prevaccination notified and nonnotified IPD or unspecified sepsis rates were combined in our study, they would still be lower than the nonnotified IPD or unspecified sepsis rates for Finland (359 per 100,000 person-years).6 It is unlikely that the lower baseline rates are due to the exclusion of the B95.3 and B95.5 codes from our study, as these made up <2% of the Finnish codes. They could be lower because we only included inpatient records and the diagnosis at discharge (not outpatient diagnoses and working diagnoses in emergency) as well as only the first episode of an IPD-related outcome. However, in both countries, most cases were hospitalized and only 4% of children in our study had a repeat episode recorded. Furthermore, we included diagnoses in any primary and up to 20 (WA) or 50 (NSW) secondary diagnosis fields (the Finnish study used codes listed in the first 3 positions) which should only serve to increase, rather than reduce, the baseline rates. A possible explanation could be that Finnish hospital pediatricians have a higher degree of suspicion of pneumococcal disease and sepsis in febrile children, to the extent that the IPD compatible and unspecified sepsis codes are more likely to be assigned in the absence of laboratory confirmation. Despite the methodologic differences, the estimated total preventable incidence of IPD-like illness is considerably higher in both countries after taking into account the nonnotified cases.
The main strengths of our study are that it was population based and that there was complete capture of inpatient hospitalizations. As in the Finnish studies, we were able to report the additional burden of preventable IPD not captured though notifications, in contrast to the English study which was based on unlinked hospitalization records.4,6,7 The main limitation is the ecologic study design that may be confounded by unmeasured changes over time in ICD coding practices, health systems and health seeking behaviors. However, as described above, the RRRs we observed for notified IPD are consistent with known VE estimates and surveillance data, as well as what was observed in Finland, and we are unaware of any changes to coding practices or surveillance methods during this time. Finally, the size of our cohorts was limited by when surveillance for IPD notifications began, the timing of changes to the vaccination program and the follow-up time available. This is particularly a limitation with the PCV13 analysis, as there was only 20 months of follow-up available post-PCV13 introduction for <1 year olds, and the declines post-PCV13 appear to continue a trend that started pre-PCV13. Therefore, longer-term follow-up of the PCV13 cohort is required before accurate conclusions can be drawn about the additional impact of PCV13 above that seen for PCV7. Despite this limitation, we were able to include ~300,000 person-years of follow-up in each PCV13 cohort (an annual birth cohort that was twice the size of Finland’s) and show a small, but statistically significant, impact of the change from PCV7 to PCV13 on nonnotified IPD or unspecified sepsis, although the confidence intervals are wide.
In conclusion, we have been able to provide a quantitative estimate of the total burden of IPD preventable by PCV7 and PCV13 vaccination programs in Australia. As found in Finland for PCV10, the additional preventable disease burden identified using the sensitive case definition of nonnotified IPD or unspecified sepsis was significant, but relatively smaller in Australia. This highlights the need for country-specific studies to fully understand the burden of pneumococcal disease preventable by vaccination and thus accurately evaluate the cost-effectiveness of PCV vaccination programs.
The authors are grateful to the staff at the Population Health Research Network (PHRN) and participating PHRN data linkage and infrastructure nodes (the WA Data Linkage Branch, the NSW Centre for Health Record Linkage and the Australian Institute for Health and Welfare), the WA and Commonwealth Departments of Health and NSW Ministry of Health, and the data custodians of all datasets involved in this project for their advice and the data. The authors thank the data linkage units, data custodians, Department of Human Services, and study reference groups (Aboriginal Immunisation Reference Group and Infectious Diseases Community Reference Group) for their support and advice. The authors also acknowledge Dr. Lisa McCallum for her help preparing the data for this analysis. ACIR Investigator Team also includes: Snelling TS, de Klerk N, Andrews RM, Blyth CC, Richmond P, Jorm L, Sheppeard V, Effler P, Menzies R, Hull B, Joseph T.
1. Jayasinghe S, Menzies R, Chiu C, et al. Long-term impact of a “3 + 0” schedule for 7- and 13-valent pneumococcal conjugate vaccines on invasive pneumococcal disease
, 2002-2014. Clin Infect Dis. 2017;64:175–183.
2. Lehmann D, Willis J, Moore HC, et al. The changing epidemiology of invasive pneumococcal disease
in aboriginal and non-aboriginal Western Australians from 1997 through 2007 and emergence of nonvaccine serotypes. Clin Infect Dis. 2010;50:1477–1486.
3. Lowbridge C, McIntyre PB, Gilmour R, et al. Long term population impact of seven-valent pneumococcal conjugate vaccine
with a “3 + 0″ schedule-How do “2 + 1″ and “3 + 1″ schedules compare? Vaccine. 2015;33:3234–3241.
4. Palmu AA, Jokinen J, Nieminen H, et al. Vaccine effectiveness of the pneumococcal Haemophilus influenzae protein D conjugate vaccine (PHiD-CV10) against clinically suspected invasive pneumococcal disease
: a cluster-randomised trial. Lancet Respir Med. 2014;2:717–727.
5. Feikin DR, Scott JA, Gessner BD. Use of vaccines as probes to define disease burden. Lancet. 2014;383:1762–1770.
6. Palmu AA, Kilpi TM, Rinta-Kokko H, et al. Pneumococcal conjugate vaccine
and clinically suspected invasive pneumococcal disease
. Pediatrics. 2015;136:e22–e27.
7. Thorrington D, Andrews N, Stowe J, et al. Elucidating the impact of the pneumococcal conjugate vaccine
programme on pneumonia, sepsis and otitis media hospital admissions in England using a composite control. BMC Med. 2018;16:13.
8. Gidding HF, McCallum L, Fathima P, et al. Probabilistic linkage of national immunisation and state-based health records for a cohort of 1.9 million births to evaluate Australia
’s childhood immunisation program. Int J Population Data Sci. 2017;2:1–13.
9. Moore HC, Guiver T, Woollacott A, et al. Establishing a process for conducting cross-jurisdictional record linkage in Australia
. Aust N Z J Public Health. 2016;40:159–164.
11. Australian Government Department of Health. The Australian Immunisation Handbook. 11th ed. 2015.Canberra, Australia
: Commonwealth of Australia
12. Hull B, Deeks S, Menzies R, et al. Immunisation coverage annual report, 2007. Commun Dis Intell Q Rep. 2009;33:170–187.
13. Gidding HF, McCallum L, Fathima P, et al; ACIR linkage Investigator Team. Effectiveness of a 3 + 0 pneumococcal conjugate vaccine
schedule against invasive pneumococcal disease
among a birth cohort of 1.4 million children in Australia
. Vaccine. 2018;36:2650–2656.