Background
Streptococcus pneumoniae is a common gram-positive bacterial pathogen with more than 90 identified serotypes that may reside in the nasopharynx of healthy individuals. It can cause diseases such as otitis media and pneumonia if it spreads to adjacent organs. If the bacteria enter the bloodstream, they cause invasive pneumococcal disease (IPD) manifesting as bacteraemic pneumonia, meningitis or other clinical presentations. Pneumococci are transmitted by direct contact with respiratory secretions from both healthy carriers and patients [
1].
The burden of disease is not evenly distributed, with those individuals with certain chronic conditions (immunosupression, cochlear implants, asthma, diabetes, alcoholism, chronic diseases of the lungs, heart, liver, kidneys) at a much greater risk of infection [
2].
Many developed countries have implemented a pneumococcal vaccination programme targeting different sectors of the population to protect against a number of the most prevalent serotypes. In England, the 23-valent polysaccharide vaccine (PPV23) was first recommended for those in clinical risk groups in 1992, then extended to elderly cohorts in a step-wise manner, eventually being offered to all individuals aged 65+ years in 2003. Uptake for 2014–2015 was 69.8% of eligible elderly individuals [
3]. The seven-valent conjugate vaccine (PCV7) was first offered to all infants in September 2006 as a 2 + 1 schedule (at 2, 4 and 12 months), with a catch-up campaign for children up to 2 years old. PCV7 was replaced in April 2010 by the 13-valent conjugate vaccine (PCV13) on the same schedule but without a catch-up campaign.
Analyses in England have demonstrated the effectiveness [
4] and population impact of the sequential PCV7 and PCV13 vaccination programme in infants through the reduction of vaccine-type carriage [
5,
6] and IPD caused by the serotypes targeted by the vaccines [
7,
8]. The impact of this programme has also been demonstrated in terms of indirect protection against IPD provided to unvaccinated cohorts, though serotype replacement with non-vaccine types has reduced the overall impact on IPD [
7,
8].
We present an analysis of the estimated direct and indirect impact of the PCV programme on hospital-diagnosed pneumonia, sepsis and otitis media on all ages using Hospital Episodes Statistics (HES) data from 2004–2005 to 2014–2015. We compare the disease trends for these outcomes to trends for five control conditions unlikely to be affected by the PCV vaccination to infer both direct and indirect impact of the PCV programme.
Discussion
Our national study shows that the PCV programme in England has been associated with significant reductions in hospital admissions across a range of non-specific disease endpoints in the age groups targeted for vaccination. The endpoints chosen were ones in which the pneumococcus is likely to have a causative role, although the percentage contribution of pneumococcal infection to these syndromes, and the serotype distribution, cannot be directly determined.
This is the first study to investigate the impact of the PCV programme on hospital admissions for a wide range of pneumococcal-specific and syndromic disease endpoints across all ages in England. In addition, we sought to improve the interpretation of our impact analyses by comparing with changes in control conditions to allow for biases generated by secular trends in admission practice. We also used as a benchmark changes in overall IPD incidence, which given the specificity of this outcome is likely to represent a maximal impact when compared with less specific disease endpoints such as all-cause pneumonia.
The reduction in admissions for pneumonia of unspecified organism in children under 5 years of age (which comprises more than 90% of all pneumonia admissions in this age group) is similar to that in studies in Sweden [
12,
13], Uruguay [
14], Scotland [
15] and southern Israel [
16], where reductions ranging from 19 to 32% following the implementation of sequential PCV7/PCV13 programmes have been reported. The consistency of these observations in different settings suggests a causal association with PCV7/PCV13 use. The reductions observed in pneumonia admissions in vaccine-eligible children post-licensure are greater than suggested by the pre-licensure trial of PCV7 in the USA in which clinically diagnosed pneumonias were only reduced by 4.3% [
17]. Unlike the US trial, our study was restricted to pneumonias requiring hospital admission. As the contribution of the pneumococcus to hospital-admitted pneumonias in children is likely to be higher than in those not admitted, the percentage reduction in inpatient pneumonia activity is likely to be higher than 4.3%. The greater than expected benefit of sequential PCV7/PCV13 programmes may also reflect the contribution of the additional serotypes in PCV13 to pneumonia — especially serotypes 1 and 19A. The development of the indirect effect will also increase the reduction observed in the vaccinated age groups.
Reductions in older age groups in pneumonia of unspecified origin could not be demonstrated, with increases in admissions for J18 observed in all age groups from 15 years upwards, continuing the trend observed before the introduction of the PCV7 programme in 2004 [
18]. A similar increase in 65+ year olds post-PCV7/PCV13 implementation has also been reported from Scotland, accompanied by a reduction in the length of stay [
15]. These changes in older age groups may represent an increasing trend towards short stay emergency hospital admissions in an aging population, as suggested by a National Audit Office report [
19]. The increase in pneumonia admissions in our study persisted even after taking into account the upward trend in admissions for the composite control conditions (Additional file
1: Figure S3). However, the ratio of the IRR for J18 compared to the IRR for each individual control condition was closer to 1 when using urinary tract infections as the control in those aged 65 years and over (Fig.
3); admissions for UTIs in the elderly are likely to be subject to similar emergency admission practices as acute respiratory infections. Few other studies have assessed the indirect impact of PCV in older age groups. In the USA, a reduction in all-cause pneumonia admissions in 65+ year olds was reported in a study that encompassed post-PCV7 and PCV13 periods ranging from 7% in 65–74 year olds to 23% in 85+ year olds [
20], but this could not be replicated in a recent analysis from Australia [
21] or in our study. A hospital-based prospective study in England that used urinary antigen detection methods showed a reduction in admissions for non-bacteraemic vaccine-type pneumonia caused by a PCV13 serotype in 65+ year olds similar in magnitude to the reduction in IPD, confirming the indirect protection against non-bacteraemic pneumococcal pneumonia [
22]. However, there was little reduction in overall pneumonia admissions, which may reflect serotype replacement, secular trends in non-vaccine serotypes or changes in the epidemiology of other non-pneumococcal pathogens.
We observed a significant reduction of 80% in pneumococcal pneumonia in children under 2 years of age (Table
2, IRR column) which was similar to the reduction in IPD. Since the diagnosis of pneumococcal pneumonia requires evidence of the causative organism, the correspondence between these two disease outcomes is to be expected. Reductions in pneumococcal pneumonia were also observed in other age groups consistent with the indirect protection observed for IPD [
8]. For pneumococcal sepsis, which includes pneumococcal meningitis and arthritis, significant reductions were observed in children under 15 years, with the largest reduction (66%) in children under 2 years old. The incidence of pneumococcal sepsis when added to pneumococcal pneumonia was considerably lower than the incidence of laboratory-confirmed IPD in all age groups. This is consistent with a previous study in which we linked laboratory-confirmed IPD reports with HES admissions and found that only a minority of laboratory-reported IPD cases had a specific pneumococcal sepsis or pneumonia code in HES [
23], thus highlighting the limitations in documenting pathogen-specific causes of admission in HES.
Our results for the impact on empyema (Fig.
1, Table
2) are comparable to those reported in the USA [
24] and Scotland [
25], where the incidence of empyema increased after the introduction of PCV7 but declined after PCV13 for children aged < 15 years. This suggests that empyema is linked to additional serotypes covered by PCV13 which increased due to serotype replacement post-PCV7 and is consistent with studies showing that the three most prevalent serotypes causing empyema are 3, 1 and 19A [
23], all of which are covered by PCV13 but not PCV7.
Mixed results were obtained for the other non-specific disease endpoints. No reductions for non-specific sepsis were observed with increases both in raw IRRs and IRRs relative to the composite control in all age groups. This suggests that these codes do not contain many cases of occult pneumococcal sepsis, at least not with a predominance attributable to vaccine-type serotypes. With the exception of children less than 2 years old, the increase in incidence was greater for the risk groups than the non-risk group (Additional file
1: Figures S4-S6). The reasons for this difference are unclear, but they reflect our findings in Table
3 showing that the greatest increases in incidence for pneumonia were also in the risk groups. The findings reported from both a Finnish trial of PCV10 [
9] and a post-licensure study [
26] in which large reductions in non-specific sepsis codes were observed in vaccinated children and vaccine-eligible children respectively could not be confirmed in this post-licensure study of a sequential PCV7/PCV13 programme. Possible reasons for this are differences in coding practices between the two countries, in serotype distribution of pneumococcal-attributable sepsis, or in the contribution of non-pneumococcal pathogens to admissions with a non-specific sepsis code. The analyses from Finland excluded laboratory-confirmed IPD cases, but we were unable to do this due to a lack of common patient identifiers in our HES and laboratory-confirmed IPD surveillance datasets. However, removal of laboratory-confirmed IPD cases from admissions with a non-specific sepsis diagnosis should reduce the proportion of such cases attributable to the pneumococcus and thus the potential impact of PCV on this disease endpoint. For lung abscess with pneumonia, the rIRRs against the composite control and against each control condition individually were less than one in age groups under 15 years and similar to that for empyema.
We were unable to detect a large reduction in admissions for all otitis media diagnoses in infants, echoing the efficacy trials of PCV7 on all-cause otitis media in Finland [
27] and in the USA [
28]. For placement of tympanostomy tubes, which reflects recurrent or more serious otitis media, the US trial showed a higher efficacy of 20.1% (95% CI 1.5–35.2%), and late follow-up of the Finnish trial cohort to 5 years of age showed an efficacy of 39% (95% CI 4–61%) against tube placements [
29]. However, further follow-up of the Finnish cohort showed no sustained efficacy between 6 and 13 years of age [
30]. Although we observed reductions in the incidence of otitis media with tympanostomies in vaccine-targeted cohorts, reductions were also observed in the older age groups which are unlikely to be due to the PCV programme. Ear tube placements for otitis media are uncommon in adults (Additional file
1: Table S4), and our findings are therefore difficult to interpret.
Within the risk group analysis the composite control was essential to use as a benchmark, as shown by the increase in admissions for the unrelated control conditions in those with risk factors. This may be an artefact due to better recording of comorbid conditions in the available HES ICD codes, or a real increase due to changes in admission practice, but the comparison with the control conditions allowed adjustment for such secular trends. The significant reductions in pneumococcal pneumonia in high-risk individuals of all ages confirms the herd immunity benefit of the PCV programme in these vulnerable groups and that the additional benefit of direct vaccination is limited [
31]. The increase in older individuals in pneumonia of unspecified organism shown in Table
2 was largely among those in a risk group (Table
3). This may reflect their greater susceptibility to develop pneumonia from less pathogenic serotypes that have replaced the PCV13 vaccine types in the nasopharynx as has been shown for IPD in high-risk children [
32].
The use of additional ICD codes to identify the outcomes of interest should increase sensitivity, but it would likely be at the cost of reduced specificity. The contribution of additional true positive or false positive cases when the ICD-10 codes are expanded will likely vary with age group, disease outcome and time period, so the effect is unpredictable, as shown by our analyses. On balance we preferred to restrict our main analyses to outcomes of interest in the first diagnosis field, as this is intended to capture the primary reason for admission and so should maximise specificity.
Our method of assessing the impact of the PCV programme in England by comparing the IRR for the outcome of interest with that in control conditions has some advantages over other methods such as time-series analyses that have been used to assess the effect of PCV on pneumonia. The latter method does not take account of factors other than vaccination that may have resulted in secular changes in admission, and the results are sensitive to the model used to fit the pre-and post-intervention trend lines [
33]. Another variation in the use of control conditions against which to assess the impact of PCV on pneumonia has recently been evaluated in a methodological study using hospitalisation data from five countries in the Americas [
34]. In that study the authors derived a composite control by aggregating pre-vaccination data across 17 disparate ICD-10 chapters plus 20 additional conditions and then assigned weights to each to generate a "synthetic" composite control whose pre-vaccination trend best matched pre-vaccination pneumonia trends. The post-vaccination data from the weighted synthetic controls were then used as a counterfactual against which to assess the impact of PCV. The weights applied to each ICD chapter/condition varied substantially between the five countries and within each country by age. In some age groups, the results were particularly sensitive to the inclusion of specific codes in the synthetic control. Overall the results when using a country-specific synthetic control were somewhat heterogenous between countries, though, in common with our findings, there was no evidence of a decline in all-cause pneumonia admissions in older adults in any country. The synthetic control method is computationally demanding, especially when dealing with very large datasets such as the national HES dataset and does not take account of interventions apart from vaccination that may have affected the incidence of the control conditions. Furthermore, a split into subgroups such as our analysis of risk- and non-risk groups is harder to achieve. Our results demonstrate that individual comparisons between the studied diseases and the constituent conditions of the composite control are informative. For example, when using urinary tract infections as the control, the increase in admissions for pneumonia of unspecified organism in adults aged 65+ years was minimised (21% relative increase); both are common infections in this age group and are likely affected by the same underlying changes in admission practices. However, when compared to trends for fractures which showed little change in 65+ year olds over the same period (Additional file
1: Table S7), the relative increase was maximal (88% increase). Based on the knowledge of how admission or management practices have changed in an individual country, it may be more efficient to select control conditions that will be unaffected by the PCV programme but similarly affected by known changes in health service utilisation.
Limitations
Although the HES database has been running prior to the start of the study period, we were unable to extract data for episodes prior to April 2004 because patient identification fields were not available in the data, precluding identification of repeat admissions in the same individual. We were therefore only able to assess the impact of PCV7 using 2 years' worth of data prior to its introduction mid-2006. Our results would be more robust if additional data prior to April 2004 were able to be linked to our existing dataset. The HES database does not include information on any laboratory testing conducted for each individual, so we were unable to confirm the aetiology of each case.
We identified risk groups using only the information available within the selected admissions, rather than examining prior admissions for individuals in the years before the outcome diagnoses. This may have resulted in a lower sensitivity for risk group identification. However, we performed a subanalysis (unpublished) where more information from the HES dataset was extracted from previous years on a patient-by-patient basis for a sample of 1000 patients. Despite being much more work, adding information from previous years increased the risk group identification only from just below 80% to just over 90%, which we deemed acceptable given the purpose of this analysis.
Another potential limitation is that we did not try to estimate the impact of PCV7 and PCV13 separately. We did this because our laboratory-confirmed IPD analyses showed that the impact of serotype replacement post-PCV7, particularly in relation to 19A and 7F, which offset some of the benefit of PCV7 was mitigated by the subsequent introduction of PCV13 [
8], and thus it was more informative to focus on the impact of the sequential PCV7 and PCV13 programme as a whole than each separately.
Our analysis of the potential change in the case burden for each of the diagnoses of interest may have underestimated the reported changes in Table
4, as the initial period from September 2006 through April 2007 was missed due to the use of HES data years.
Our use of five control conditions is a novel approach in assessing the impact of a pneumococcal vaccination programme on hospitalised disease outcomes. Our analysis could be improved if we had used more than five control conditions and conducted a more comprehensive analysis of the effect of using as controls conditions that were likely or unlikely to be affected by similar changes in admission practices as the pneumococcal-attributable outcomes. However, we were unable to extract data for more than five control conditions due to resource limitations, and with data from only 2 pre-PCV years available to us in HES, we were unable to conduct further investigations into the similarities between the control conditions and the disease outcomes of interest beyond the age distributions of cases and not being subject to other public health interventions (see Additional file
1: Figures S1 and S2). Furthermore, we did not assess the seasonality of the control conditions. However, we minimised the impact of seasonality by averaging over a 24-month period and including the same months pre- and post-PCV (April to March) for both outcomes of interest and control conditions. Despite being restricted to five controls, our analysis is an improvement on impact analyses that use no controls or the use of a single control condition which can be sensitive to changes in secular trends unrelated to the introduction of the PCV programme and therefore a source of bias in programme impact estimates [
34].
We have reported disease trends for many age groups and many diagnoses of interest, so it is possible that our results may include some IRR estimates for which the upper 95% confidence is less than one by chance. However, a formal correction for multiple comparisons is not straightforward in this instance given the method that we have used. Furthermore use of only five controls in deriving rIRRs precluded any formal statistical comparison such as computation of confidence intervals, so uncertainty in this measure could only be depicted by showing maximum and minimum rIRR values across the control conditions.
It is also important to note that the overall burden of non-invasive pneumococcal disease will be determined not only by the impact of the PCV programme on disease caused by the pneumococcus but also by the epidemiology of other causative pathogens for pneumonia. Changes in the latter may mask an overall reduction in pneumococcal-attributable non-invasive disease, but this cannot be assessed by the type of impact assessment conducted here.