The rationale for use of clinically defined outcomes in assessing the impact of pneumococcal conjugate vaccines against pneumonia

ABSTRACT

Introduction: When evaluating the public health value of adult pneumococcal conjugate vaccine (PCV) for pneumonia, regulatory agencies and vaccine technical committees (VTCs) emphasize vaccine serotype (VT), radiologically confirmed community-acquired pneumonia (CAP) to the exclusion of clinically defined pneumonia and thus may underestimate PCV’s public health value.

Areas covered: We review the critiques that have been raised to using clinically defined pneumonia as a complement to VT-CAP in evaluating the public health value of adult PCVs.

Expert opinion: PCV13 efficacies for preventing hospitalized CAP ranged from 6% to 11% and for a combination of primary and secondary care from 4% to 12%, with relatively high associated rate reductions. These efficacy values are larger than estimated from multiplying PCV13 efficacy against vaccine-type CAP by the proportion of CAP identified as vaccine-type through tests, such as a serotype-specific urinary antigen detection assay. Current understanding of pneumococcal epidemiology and limitations of diagnostic tests suggest the efficacy values for clinically defined outcomes are plausible and potentially generalizable. Regulatory agencies and VTCs have accepted clinically defined outcomes for assessing pediatric vaccines and – while additional studies assessing adult clinical CAP VE are needed – they might consider existing data when evaluating adult PCV use.

KEYWORDS:

• Adult
• community-acquired pneumonia
• pneumococcus
• pneumococcal conjugate vaccine
• pneumonia
• vaccine efficacy

1. Introduction

Etiologically defined vaccine-serotype or radiologically confirmed pneumonia have been used to license pneumococcal conjugate vaccines (PCV) for the prevention of pneumonia in both adults and children. Etiologically confirmed outcomes are especially relevant and are the focus for regulatory trials and for determining if a PCV prevents disease against the serotypes contained in the formulation. However, these specific outcomes have limited sensitivity and thus, when they are the focus of assessments of the public health benefit of a vaccination program, may provide an overly conservative estimate. Clinically defined outcomes such as all-cause pneumonia, while less specific, can provide a broader view and estimate of a PCV’s public health benefit, including through measuring incident rate reductions, also known as vaccine-preventable disease incidence (VPDI; defined as control group incidence minus intervention group incidence) and its reciprocal, the number needed to vaccinate (NNV) (Table 1) [1,2].

Table 1. Definition of terms

Vaccine Preventable Disease Incidence (VPDI): The difference in outcome incidences between the unvaccinated (or control) and vaccinated populations, which is mathematically equivalent to the control or unvaccinated population incidence multiplied by Vaccine Effectiveness (VE). Also known as the attributable risk, absolute risk reduction, or risk difference. We report this as per 100,000 subjects per year or per 100,000 person-years of follow-up. When available, we report values directly from cited studies. Number Needed to Vaccinate (NNV): Calculated as 100,000 divided by VPDI divided by either average length of subject follow-up) or average duration of vaccine immunity (e.g. for adult PCV13 use in Table 4, NNV was estimated for 1, 5, and 10 years duration of immunity). Unlike VPDI, NNV is not a rate and thus should not be calculated for a single year unless that is the duration of the study in question or the documented duration of immunity.

Numerous recent studies – including randomized controlled trials (RCTs) and observational studies, from developed and developing countries – have reported vaccine efficacy/effectiveness and VPDIs for clinical outcomes. As we detail further down, regulatory agencies and VTCs have incorporated these data into vaccine indications and decision-making; while not justifying such an approach for a particular vaccine or outcome, the use of these data by VTCs and regulatory agencies illustrates that such groups already accept data based on clinical outcomes. Recent studies also confirm the often-substantial underestimation of the public health value afforded by vaccines that results from the analyses of only etiologically or radiologically confirmed outcomes. Studies on rotavirus, Haemophilus influenzae type b (Hib) conjugate, and PCVs have illustrated that clinical outcomes yield VDPIs 1.5-fold to 15-fold higher than those for etiologically or radiologically confirmed outcomes (Table 2). This principle has been demonstrated even for outcomes not generally considered difficult to confirm etiologically, such as invasive pneumococcal disease (IPD), Hib meningitis, and rotavirus gastroenteritis. The increase in VPDIs with clinically defined outcomes likely occurs due to the process of etiologic and radiologic confirmation inevitably excluding cases due to limited test sensitivity; lack of diagnostic specimen or chest x-ray collection including differential collection based on perceived severity or underlying health conditions; problems with test availability or implementation or specimen transport; variability in chest x-ray interpretation; and limited staff availability.

Table 2. Vaccine efficacy or effectiveness (VE), vaccine-preventable disease incidence (VPDI) (per 100,000 person-years of observation or 100,000 subjects per year), and numbers needed to vaccinate (NNV) based on duration of follow-up or age at which follow-up ended

Study	Reference	Study design	Age of follow-up	Outcome	VE	Incidence unvaccinated	VPDI	NNV *	Fold increase in VPDI (and decrease in NNV) compared to etiologically confirmed disease
Indonesia Hib	[3]	Cluster randomized	<2 years	Hib meningitis	86%	19	16	3125	Referent
				Hospitalization for meningitis or clinic assessment for seizures	22%	701	158	316	9.9
Gambia 9-valent PCV	[4]	Individually randomized	<2 years	Vaccine-serotype bacteremic pneumonia	70%	200	140	357	Referent
				X-ray confirmed lobar consolidation or pleural effusion	37%	3510	1300	38	9.3
Finland 10-valent PCV	[5]	Cluster randomized	<2 years	All serotype invasive pneumococcal disease	94%	80	75	671	Referent
				Clinically suspected invasive pneumococcal disease	50%	426	210	238	2.8
				Hospitalized pneumonia	23%	1234	271	185	3.6
The Netherlands PCV13	[6]	Individually randomized	65+ years, 5 years assumed immunity duration	PCV13 serotype community acquired pneumonia with radiologic confirmation	37.5%	67	25.1	798	Referent
				Hospitalized clinically defined community acquired pneumonia	8.1%	891	72.2	277	2.9
Finland, rotavirus	[7]	Pre/post	<1 year	Hospitalized rotavirus acute gastroenteritis	80%	490	390	256	Referent
				Hospitalized all-cause acute gastroenteritis	54%	1980	1070	93	2.7
US rotavirus (2008–9)	[8]	Cohort	8 to 20 months	Hospitalized rotavirus acute gastroenteritis	87–92%	100 - 121	87 - 111	901 - 1149	Referent
				Hospitalized all-cause acute gastroenteritis	40–56%	349 - 491	139 - 275	364 - 719	1.6 to 2.5
South Africa inactivated influenza virus in pregnant women	[9]	Individually randomized	≤175 days	Influenza confirmed acute lower respiratory infection	100%	218	218	458	Referent
				All acute lower respiratory infection	43%	7200	3100	32	14
Africa RTS,S malaria vaccine	[10]	Individually randomized (comparing R3R vs. C3C)	Enrolled at age 5–17 months, followed to 48 months	Malaria hospitalization			1000	25	Referent
				All-cause hospitalization			1475	17	1.5
			Enrolled at age 6–12 weeks, followed to 39 months	Malaria hospitalization			554	56	Referent
				All-cause hospitalization			1108	28	2.0

With regards to pneumonia, clinically defined outcomes are used when assessing the benefit of PCVs in pediatric populations, largely because no sensitive and specific serotype-specific diagnostic test exists for non-bacteremic pneumococcal pneumonia, which constitutes most pneumonia caused by the pneumococcus. A Cochrane Review of five pediatric PCV RCTs reported vaccine efficacies of −11% to 35% against radiologically confirmed pneumonia with a meta-estimate of 19% (95% CI, 0 to 34%); efficacy increased to 27% (95% CI, 15% to 36%) when excluding an outlier study among American Indians [11]. Four of these studies also reported efficacies for clinically defined pneumonia ranging from 0% to 7% (only one of which was statistically significant at the 95% confidence level) with a meta-estimate of 6% (95% CI, 2% to 9%). Nevertheless, for the three studies with a clinically defined pneumonia efficacy >0%, the VPDI was 1.5- to twofold greater for this outcome than the radiologically defined outcome [4,12,13]. As we review in more detail below, PCV efficacy data for clinical pneumonia have informed pediatric decision-making such as the 2010 US recommendations for PCV13 use, which cited efficacy against ‘IPD, x-ray-confirmed pneumonia, and clinically diagnosed pneumonia’ [14].

While clinically defined outcomes are used to assess PCVs in pediatric populations, this is not generally true for PCV use in adult populations. Consequently, here we review the rationale for use of clinically defined pneumonia as a useful public health outcome for assessing the value of PCVs in adults. We provide calculations of VPDI and NNV for clinically and etiologically/radiologically defined pneumonia and lower respiratory tract infection (LRTI) outcomes to illustrate the large multiple in values for these estimates that results from use of the former and thus how clinically defined endpoints can inform vaccine program evaluations. We then address concerns with plausibility and propose a framework for using clinically defined pneumonia in the overall evaluation of PCVs as a complement to the more conservative estimates derived from etiologically and radiologically confirmed pneumonia endpoints.

2. Current evidence for pcv13 efficacy/effectiveness against clinically defined cap among adults

For adults, many VTC evaluations and economic and impact models rely on vaccine efficacy or effectiveness against vaccine serotype CAP, and incidence data for this same outcome. Adult vaccine serotype CAP in turn has been based largely on Pfizer’s serotype-specific urinary antigen detection (UAD) assays, which we discuss in more detail below.

PCV13 efficacy against vaccine serotype CAP was determined from a large double-blind RCT conducted in The Netherlands [15,16]. This study demonstrated direct PCV13 efficacy of 46% (95.2% CI, 22% to 63%) against vaccine-type hospitalized CAP (VT-CAP) – including efficacy for non-bacteremic CAP of 45% (95.2% CI, 14% 5o 65%) – among persons age ≥65 years, leading to licensure of PCV13 for non-bacteremic pneumonia by the US Food and Drug Agency (FDA) and the European Medicines Agency (EMA). A subsequent observational study from Louisville, Kentucky of adults hospitalized with radiologically confirmed CAP found a vaccine effectiveness of 71.2% (95% CI, 6.1% to 91.2%) [17] against VT-CAP.

More recent assessments have shown adult PCV13 efficacy or effectiveness against clinically defined pneumonia including: a) an adult RCT from the Netherlands with two study reports, one on hospitalized CAP [6] and another on CAP and LRTI identified in primary and secondary care [18]; b) an observational study among US Medicare recipients assessing PCV13 effectiveness against hospitalized CAP [19,20]; and c) an observational study in Saxony, Germany among persons included in an insurance claims database assessing PCV13 effectiveness against all-cause hospitalized and outpatient pneumonia [21]. These studies have largely reported results in line with those observed for pediatrics, with vaccine efficacy/effectiveness estimates ranging from 6% to 11% for hospitalized CAP and 4% to 12% for outpatient or combined outpatient and hospitalized CAP (Tables 3, 4). These estimates were associated with large VPDIs, and in some cases (e.g. for combined inpatient and outpatient LRTI in The Netherlands RCT) are among the largest reported PCV-associated VPDIs for CAP whether in pediatric or adult populations. We will now address the critiques of these data as delineated previously.

Table 3. Characteristics of three studies evaluating the efficacy or effectiveness of 13-valent pneumococcal conjugate vaccine (PCV13) against pneumonia in older adults

Study characteristic	United States observational study	Netherlands randomized controlled trial (RCT)	Germany observational study
Reference	19, 20	6, 18	21
Design/data source	Time survival model of Medicare data	RCT with 1:1 PCV13:placebo	Population based cohort study of health insurance data with propensity score-based 3:1 matching of controls to cases for vaccine effectiveness calculation
Population size	81,009,094 person-years of observation; PCV13 coverage at study start (9/14) was 0.8% and at study end (12/17) 41.5%	42,240 received PCV13 42,256 received placebo	11,395 persons received PCV13 34,185 controls did not receive PCV13
Population Characteristics	Age 65+ years No exclusion of immunocompromised	Age 65+ years Non-immunocompromised	Age 60+ years No exclusion of immunocompromised
Duration	3.25 years	4 years	3 years
Outcome ascertainment	Hospital	Almost entirely hospital (ref. 3) and hospital + outpatient (ref. 89)	Hospital and outpatient
Setting	United States	Netherlands	Saxony
Outcome	Community acquired pneumonia based on a primary diagnosis of pneumonia or a secondary diagnosis of pneumonia with a primary diagnosis of septicemia, meningitis, empyema, or acute respiratory failure	Community acquired pneumonia and lower respiratory tract infection based on based on at least 2 pre-specified clinical criteria independent of radiographic findings	Pneumonia based on ICD-10-GM coding*

*ICD-10-GM coding: International Classification of Diseases 10^th Revision, German Modification

Table 4. Incidences and vaccine-preventable disease incidence (VPDI) (per 100,000 person-years of observation [Netherlands] or 100,000 subjects per year [Germany and United States]) as well as numbers needed to vaccinate (NNV) based on either 5 years or 10 years of 13-valent pneumococcal conjugate vaccine (PCV13) duration of protection against pneumonia. NA = not available

Study	Reference	VE (95% confidence interval or p-value)	Incidence vaccinated	Incidence unvaccinated	Incidence overall	VPDI	NNV (5 yrs PCV13 immune duration)	NNV (10 yrs PCV13 immune duration)
United States PCV13-type etiologically confirmed hospitalized CAP	19,20	43% (12% to 63%)	NA	NA	17 to 76	7 to 33	600 to 2800*	300 to 1400*
United States hospitalized clinical CAP	19,20
Flu season/flu vaccine+		6.0% (4.3% to 7.8%)	NA	NA	1776**	107	188	94
Flu season/flu vaccine-		11.4% (9.4% to 13.4%)	NA	NA	1776**	202	99	49
Non-flu season/flu vaccine+		9.3% (7.5% to 11.5%)	NA	NA	1776**	165	121	61
Non-flu season/flu vaccine+		10.2% (6.8% to 13.5%)	NA	NA	1776**	181	110	55
Netherlands hospitalized CAP	6	8.1% (1.0% to 14.6%)^†	819	891	--	72	277	138
Netherlands 1^o and 2^o care	18
Clinical CAP 1^o care		5.5% (-2.6% to 13.0%)	1,910	2,020	--	110^‡	182	91
Clinical CAP 1^o and 2^o care		7.4% (0% to 14.4%)	3,120	3,370	--	250^‡	80	40
LRTI 1^o care		3.8% (-1.1% to 8.4%)	11,120	11,550	--	430^‡	47	24
LRTI 1^o and 2^o care		4.4% (-0.3% to 9.0%)	12,320	12,890	--	570^‡	39	20
Germany all pneumonia 1^o and 2^o care	21	11.9% (p=0.028)	1556	1767	NA	211	95^Ω	47

* The original presentation calculated NNV over just a single year; however, because this is not the recommended method for NNV calculation as it does not take into account duration of protection (Refs. 22,23), we do not present these results

** VE was presented only by the four strata of influenza season and influenza vaccine, while incidence was presented only for the population overall (calculated as 148 per 100,000 beneficiary-months for CAP multiplied by 12 months [Figure 4, ref. 20])

† Based on exact methods (ref. 6, Table 2a)

‡ The original publication did not present VPDI or NNV. These were calculated as described in the text from original data

Ω The original publication reported an NNV of 158, calculated over the 3-year study duration

3. Framework

While vaccine technical committees (VTCs), such as the US Advisory Committee on Immunization Practices (ACIP), and other expert groups, have reviewed data on PCV efficacy or effectiveness against adult clinically defined pneumonia, they generally do not base policy decisions on these data. This may occur for several potential reasons: 1) vaccine efficacy or effectiveness estimates for clinically defined CAP are inconsistent with the proportion of CAP documented to be due to vaccine serotypes; 2) vaccine efficacy or effectiveness for etiologically confirmed pneumonia is a characteristic of the vaccine while these measures for all-cause pneumonia reflect the epidemiological characteristics in which a study is performed; 3) pediatric PCV programs have eliminated circulation of vaccine serotypes; 4) to the extent vaccine serotypes remain, only serotype 3 remains important and PCV13 may have more limited efficacy against serotype 3; and 5) all-cause clinical outcomes are not generally favored versus etiologically derived outcomes for use in product labels or when evaluating vaccination programs in general. In the next sections, we will review each of these concerns.

4. Epidemiological arguments

4.1. Clinical CAP PCV13 efficacy/effectiveness can be reconciled with data on the confirmed proportion of CAP due to vaccine serotypes

In the US, a study among persons age 65 years and older hospitalized with radiologically confirmed CAP reported that 4.2% were due to a vaccine serotype based mainly on Pfizer’s serotype-specific UAD assay [22]. Assuming either a 45% efficacy against this outcome as reported from The Netherlands RCT [15] or a 71% adjusted effectiveness as reported from an observational study in the US [17] implies that PCV13 should prevent 1.9% to 3.0% of all CAP. However, this value is inconsistent with the reported 6% to 11% reduction of all-cause CAP among the US Medicare population [19, 20]. Similarly, a study in Germany found that 7.4% of adults age 65 years and older hospitalized with radiologically confirmed CAP had a PCV13 serotype confirmed [23], primarily by UAD; applying the same 45% efficacy or 71% effectiveness to this value implies PCV13 should prevent 3.3% to 5.2% of all CAP, not the 11.9% observed [21]. Data internal to The Netherlands RCT were more consistent [15]: for the modified intention to treat analysis population, of 787 placebo recipients with radiologically confirmed CAP, 106 (13%) had a vaccine type (again, based primarily on the UAD), giving an expected reduction in clinical CAP of 6% compared to the 8% found. These data suggest that more vaccine-preventable disease exists than can be detected solely by evaluation of more specific outcomes, specifically CAP based on a consistent radiologic picture and a vaccine serotype identified by UAD. Several lines of evidence allow reconciliation of these data.

Regulatory authorities have validated Pfizer’s UAD assay for diagnosing VT non-bacteremic pneumonia in adults [24] (Public Health England has also developed a UAD test [25]). While Pfizer’s UAD has a reported sensitivity of 97% and specificity of 100%, these test characteristics were derived using a gold standard of bacteremic, chest x-ray confirmed CAP. UAD sensitivity and specificity for VT non-bacteremic CAP are unknown because no gold standard for non-bacteremic pneumococcal CAP exists. The UAD is a limit assay and as such cut-points for defining a normal value for each serotype are based on a statistically valid sample of 400 healthy controls for each study [24,26]. Consequently, it is possible that cut-points adequate to separate antigen levels in healthy controls from patients with bacteremic pneumonia may work less well for non-bacteremic pneumonia (e.g. if non-bacteremic pneumonia leads to lower urinary antigen levels). Moreover, the UAD was developed to measure PCV13 VE against VT-CAP for licensure purposes and thus the methodology for setting cut-points was designed to maximize specificity rather than sensitivity [24,26]. Similar issues have been documented for other urinary antigen detection tests. For example, the BinaxNOW pneumococcal assay has a reported sensitivity of 77–92% for bacteremic and 52%-78% for non-bacteremic pneumonia [27]. Consequently, while in principle BinaxNOW positive pneumonia could define a category of all pneumococcal pneumonia that might be an intermediate step between VT-CAP and clinical CAP, in practice it does not avoid the primary issue of reduced sensitivity that occurs when a diagnostic test or radiological examination is part of an outcome definition. Additionally, BinaxNOW is not completely specific because the C-wall polysaccharide may be expressed by non-pneumococcal Streptococci sp. Nevertheless, if a pan-pneumococcal test were developed with high sensitivity and specificity for both bacteremic and non-bacteremic pneumonia, and if it were widely employed in countries using PCVs in adult populations, it could provide a useful tool for better assessing a PCV's public health value.

Sensitivity may also be reduced by inclusion of a positive chest x-ray based on specific radiologic criteria as part of the case definition for adult CAP. Some pneumonia patients may not have a chest x-ray obtained for a variety of reasons and even among those who have an x-ray, its interpretation for CAP, especially in elderly adults, can be difficult [28], as demonstrated by relatively low interobserver reliability for interpretation of chest x-rays for adult CAP [29]. The requirement for a specific set of radiologic findings assumes that pneumococcal CAP cannot present with other findings; this assumption, however, has been challenged by pediatric data from Israel where PCV7 and PCV13 prevented a substantial burden of lower respiratory infection (LRI) not associated with alveolar infiltrate [30]. Whether a patient with a pneumococcal LRI presents with an alveolar infiltrate – and thus meets a CAP definition – may depend on a variety of factors, such as time from symptom onset to presentation, antibiotic use before a CXR is obtained, underlying disease, immune status, and other factors. As a result of these issues, requiring a positive CXR likely misses some vaccine-preventable CAP or LRI episodes among persons that either had a negative or equivocal CXR or did not have a CXR at all.

As noted above, antibiotic use before a CXR is obtained may alter CXR findings. Similarly, antibiotic use may affect sensitivity of other common pneumococcal diagnostic assays such as blood culture, sputum, and UAD. This issue is of importance since up to 40% of adult patients with pneumonia receive antibiotics before collection of diagnostic assays [31].

Besides imperfect UAD and chest x-ray sensitivity, other theoretical explanations exist for the inconsistency between the calculated fraction of CAP preventable by PCV13 based on etiologically confirmed disease and the measured fraction of PCV13-preventable clinically defined CAP. Pneumococcal infection may occur earlier in the causal chain, and thus not be identifiable by UAD or blood culture. For example, some of the reported adult CAP may represent chronic lung or heart disease exacerbations resulting from an earlier and undocumented pneumococcal infection [32]. A more intriguing hypothesis is that pneumococci and viruses interact in the upper airway not just to increase the risk of secondary pneumococcal pneumonia, but also to increase the risk of viral pneumonia or LRI. Previous studies in children and adults have found that PCV may prevent viral acute respiratory infection [33–35], including most recently endemic human coronaviruses [36]. Other studies have found that upper airway carriage of bacteria – including pneumococci – can increase the risk of viral infection and symptomatic disease [37]. A mechanism has been suggested by a study documenting reduced mucosal IgG and IgA to live attenuated influenza vaccine in the presence of pneumococcal carriage [38].

The contribution of any of the above to the measured efficacy or effectiveness against clinically defined pneumonia remains unknown. However, these explanations offer biologically plausible mechanisms to explain the epidemiologic inconsistences that critics of these data point out and why clinically defined pneumonia could provide additional information to the highly specific outcome of vaccine serotype, radiologically confirmed pneumonia. As an example, a recent public health analysis of The Netherlands PCV13 RCT measured the degree to which the latter outcomes underestimate vaccine-preventable disease incidence, with rate reductions 2.9-fold lower for serotype and radiologically confirmed CAP versus clinically defined CAP (Table 2) [6].

4.2. While influenced by epidemiological context and case definitions, vaccine efficacy/effectiveness for clinical CAP may be generalizable

It has been argued that vaccine efficacy/effectiveness for etiologically confirmed outcomes is a characteristic of the vaccine while for clinically defined outcomes this is dependent on local epidemiology, applicable only to the time and place of the study. This, however, is a relative rather than an absolute difference. PCV13 efficacy against VT-CAP in The Netherlands RCT was calculated based on the aggregate of all 13 vaccine serotypes, with the relative ratio of serotypes dependent on the epidemiological circumstances where the trial was conducted. In this study, among the five serotypes for which each had at least 10 confirmed cases in the placebo arm, efficacy for individual serotypes varied from 20% (serotype 1) to 62% (serotype 3) to 73% (serotype 7 F) [39]. Since efficacy differs by serotype, the aggregate VE measurement will change if serotype distribution changes. Also, it is possible that PCV13 has a different efficacy/effectiveness for more severe pneumonia (as seen in the South African pediatric trial [12] but not The Gambian pediatric trial [4]), so that measured efficacy/effectiveness against etiologically confirmed disease would vary based on hospital access and study inclusion criteria. Consistent with these two hypotheses, the two efficacy or effectiveness measures for PCV13 against etiologically confirmed pneumonia among adults age ≥65 years show substantially different values: The Netherlands RCT reported adult PCV13 efficacy of 37.5% (95.2% CI, 14.3 to 54.5) [6] to 45.6% (95.2% CI, 14.2 to 65.3) [15] (depending on analytical approach) while an observational study in Louisville, Kentucky reported an adjusted effectiveness of 71.2% (95% CI, 6.1 to 91.2) [17]. The use of an RCT vs. an observational design likely explains some of the difference, however, local epidemiology may also contribute. Moreover, VPDI - and NNV as its reciprocal - for etiologically confirmed pneumonia often incorporate background pneumonia incidence, which may vary based on pediatric PCV use and implementation, strength of the influenza and RSV season, degree of co-morbidities and average age among elderly persons, residence among elderly persons at home versus long-term care facilities, and other factors. In sum, measured efficacy or effectiveness estimates are likely to vary for both etiologically confirmed and clinically defined outcomes, and the measured variation of 37% to 71% against hospitalized CAP for the former is not substantially different from the variation of 6% to 12% for the latter.

We refer throughout this document to clinical CAP. However, as noted (Tables 3 and 4), clinical CAP case definitions may vary, which likely will influence endpoints such as vaccine efficacy or VPDI because different case definitions can have varying specificity or sensitivity for vaccine preventable outcomes. Despite these differences, and as discussed above, vaccine efficacy/effectiveness results for all three studies were consistent over a relatively narrow range, and yielded associated rate reductions that were substantially higher than for etiologically confirmed CAP, suggesting there is justification for use of clinical CAP as a general category. Moreover, a similar issue exists for etiologically or radiologically confirmed CAP, since these outcomes include clinical CAP as part of their definitions. Etiologically or radiologically confirmed CAP actually may multiply issues with reproducibility due to additional variation in specificity and sensitivity of chest x-rays and laboratory tests. Finally, use of standard of care clinical outcomes may have more relevance for clinicians and payors than more specific study outcomes. For example, it may be more useful for decision-making purposes to report a 6% to 10% reduction in clinical pneumonia episodes or hospitalizations than a 45% reduction in pneumonia defined as vaccine serotype based on a Pfizer UAD combined with a positive CXR as determined by an adjudication committee (as was done with the Dutch RCT).

4.3. Despite herd protection from pediatric PCV programs, circulating vaccine serotypes still cause substantial disease in adults

When PCV13 was introduced, hope existed that infant and early childhood immunization would nearly eliminate vaccine serotypes from circulation, thus providing indirect protection to unvaccinated persons. If this had occurred, reductions in clinically defined pneumonia following direct PCV13 vaccination of adults would not be biologically plausible. Recent evidence, however, indicates that for some vaccine serotypes, substantial disease remains among unvaccinated persons. In the US, during 2015–2017, approximately seven and 17 years after PCV13 and PCV7 introduction in children, respectively, 20% of all IPD among persons age ≥65 years was due to VTs, most commonly serotypes 3, 7F, 19A, and 19F [40]. The UK reported that 22% of IPD during 2016/2017 was due to PCV13 serotypes [41]. A study of 13 sites in 10 European countries (none of which have age-based PCV13 immunization programs for adults) reported that after 5 to 6 years of pediatric PCV13 use, initial VT IPD declines in older adults were followed by a plateau and a subsequent increase [42]; among sites using PCV13 and PCV10 in their pediatric program, PCV13 serotypes represented 20–29% of older adult IPD in the former and 32–53% in the latter.

Similar data exist for VT-CAP. A US study of 23 acute care hospitals in 10 cities reported that 4 and 14 years after infant PCV13 and PCV7 implementation in infants, 4.2% of all-cause CAP among persons age ≥65 years was due to a documented PCV13 serotype [29]. This same study found that over the 3-year study period the proportion of hospitalized VT-CAP in adults age ≥65 years and immunocompromised adults age 18–64 years, both of whom were recommended to receive PCV13 immunization, declined by 30% to 40%, respectively; by contrast, no decline was seen among healthy persons and persons age 18–64 years with chronic co-morbidities, neither of whom were recommended to receive PCV13 immunization [22].

Equivalent data exist for countries without a publicly funded age-based PCV13 recommendation for older adults. An Israeli study reported that following four to five years of pediatric PCV13 use, 8% of CAP among adults age 50 years and older was confirmed as PCV13 serotypes [43]. Canada reported a decline in PCV13 serotypes among adults hospitalized with CAP from 8% in 2010 to 5% in 2014 but increasing to 6% in 2015 [44]. In Sweden, 11% of CAP cases among adults age 18+ years were identified as caused by six different PCV13 serotypes [45] similar to the 12% seen in the UK [46] and the 11% seen in Germany [23].

In summary, while pediatric pneumococcal vaccination has induced indirect protection among unvaccinated age cohorts, particularly for some serotypes, this has not been enough to prevent a substantial amount of vaccine-type IPD and CAP among older adults. This likely has occurred because for some vaccine serotypes transmission continues to occur from children to adults or among adults. Supporting data are available. For example, PCV13 has led to dramatic reductions in pediatric vaccine serotype carriage prevalence in some settings [47,48] but not others [49]. Moreover, a recent modeling study reported that transmission within the same age strata was estimated to explain one-third of IPD among elderly UK adults [50]. Finally, recent studies have found that adult pneumococcal carriage point prevalences can be as high as 20% to 40%, albeit at a substantially lower density than children [51,52].

4.4. Vaccine serotype CAP is often due to serotype 3, against which PCV13 provides direct protection, helping to explain efficacy/effectiveness against clinical CAP

Even if vaccine serotype disease persists, if a substantial proportion is caused by serotype 3, efficacy or effectiveness would be limited based on a view of limited protection conferred by PCV13 against this serotype. At the hypothetical extreme, if all remaining vaccine type CAP was due to serotype 3, and PCV13 efficacy against serotype 3 were zero, PCV13 should have a measured efficacy or effectiveness against clinical CAP equal to zero. Published evidence, however, contradicts both assumptions underlying this criticism.

First, vaccine serotypes other than serotype 3 continue to circulate. Studies conducted in US, UK, Israeli, and Swedish adults showed that at least 50% of VT-CAP is not due to serotype 3 [21,43,45,46].

Second – while the most recently published cost-effectiveness models from the US ACIP [53,54] assumed as a base case no PCV13 efficacy against serotype 3 non-bacteremic pneumonia and in Germany the Robert Koch Institute assumed half the efficacy against serotype 3 as for other serotypes [55] – recent data are consistent that PCV13 provides direct protection against serotype 3 CAP in adults. The Netherlands RCT and a pooled analysis, the latter including the RCT and two other observational studies in the US and Argentina, found VEs against serotype 3 CAP hospitalization to be 62% (95% CI, 18% to 83%) [39] and 53% (95% CI, 6.2% to 76%) [56], respectively. These data, although limited, are supported by additional data documenting PCV13 effectiveness against pediatric serotype 3 IPD [57,58].

5. Regulatory issues

5.1. Regulatory agencies incorporate clinical outcomes into product labels and thus VTCs may also consider using these outcomes

Regulatory agencies indeed have included clinical outcomes for pediatric populations in PCV labels. The PCV13 EMA Summary of Product Characteristics (SMPC) includes a 16% decline in all CAP following a switch from PCV7 to PCV13 reported from France as well as a 68% reduction in outpatient visits and a 32% reduction in hospitalizations for alveolar CAP in Israel during the PCV13 period compared to the pre-PCV7 period [59]. Likewise, the SMPC for PCV10 (Synflorix®) includes VE against clinically suspected CAP of 6.7% (95% CI, 0.7 to 12.3%) [60,61]. The US FDA’s package insert cites a US pediatric PCV7 RCT reporting a 7% reduction in all-cause otitis media [62]. The New Zealand [63] and Australian [64] package inserts refer to reductions in all-cause pneumonia from multiple studies and for multiple age groups.

It is therefore not surprising that VTCs have included clinical outcome data in their decision-making processes for PCV use in pediatric populations. The UK JCVI cost-effectiveness analysis for pediatric PCV7 included a VE against all-cause radiologically confirmed pneumonia of 18% and against all-cause otitis media of 7% [65]; the subsequent model for PCV13 fitted changes in IPD to changes in outpatient visits and hospitalizations for clinical pneumonia and otitis media [66]. The US ACIP has cited a pooled PCV13 VE from clinical trials against pediatric radiologically confirmed pneumonia of 27% and against clinical pneumonia of 6% [14]. The World Health Organization’s 2019 update on recommendations for PCVs cites the same pooled 27% VE, as well as a single study reporting a 34% reduction in pediatric radiologically confirmed pneumonia, and states: ‘such proportions are considered to be more precise than estimates based on data from studies of laboratory-confirmed cases’ [67].

5.2. Regulatory agencies incorporate clinically defined pneumonia among adults into product labels and thus VTCs may also consider using this outcome

Regulatory agencies also have accepted PCV vaccine efficacy or effectiveness data against clinical outcomes for adults. The product labels from EMA [59], New Zealand [63], and Australia [64] have recently included PCV13’s 8% VE against clinical CAP from The Netherlands RCT.

6. Conclusion

An increasing body of data indicates consistent efficacy or effectiveness of PCV13 against clinically defined hospitalized or outpatient CAP and LRTI among adults, with values of 4–12% in the context of pediatric PCV and population-based influenza immunization programs. While these efficacy/effectiveness estimates may seem low, they are for a common condition and thus associated VPDIs are large. We have reviewed biologic, epidemiologic, and regulatory critiques surrounding use of these data for decision-making and propose a rationale for why use of these data is epidemiologically and methodologically sound.

Like the pediatric indication, use of clinically defined pneumonia for decision-making related to direct PCV vaccination of adults has large public health implications. For example, the Netherlands RCT has reported a VPDI 2.9-fold higher and NNVs 2.9-fold lower for clinically defined versus etiologically and radiologically confirmed CAP [6] (Table 4). In the US, when combining hospitalized CAP data across several studies, VPDI was 6.1-fold to 15-fold higher and NNV 74-fold to 158-fold lower (when calculated for a 5-year duration of immunity [68] rather than for a 1-year duration of immunity) for clinically versus etiologically and radiologically defined CAP. VPDI and NNV values from Germany were consistent with data from The Netherlands and the US [69].

Furthermore, VPDI and NNV values for hospitalized clinically defined pneumonia outcomes in Table 4 may still underestimate the direct medical benefit of an adult PCV13 immunization program. VPDI and NNV incorporate background disease incidence and this is often underestimated due to issues such as patients being held in the emergency room or hospitalized and discharged (including due to death) before study enrollment can occur; lack of a complete workup; refusal of consent; or presentation to a non-study hospital. For example, the latter issue was identified as a cause of losing 37% of CAP episodes in The Netherlands RCT [70]. The data presented in Table 4 represent only VPDI and NNV associated with acute CAP events, and thus exclude prevention of other outcomes such as hospital-acquired and ventilator-associated pneumonia; chronic disease complications following pneumonia such as chronic lung disease exacerbations or myocardial infarction as has been demonstrated for several respiratory viruses [71,72]; the contribution of antibiotic treatment for otherwise preventable pneumonias on antibiotic resistance; and long-term reductions in quality of life following pneumonia. Lastly, as seen in data from Germany and The Netherlands that included outpatient pneumonia episodes, inclusion of these data substantially increases calculated PCV13-associated VPDIs and thus public health value.

We recognize that limitations and concerns exist. For example, as noted above, generalizing data from one setting to another may be problematic when a new PCV is introduced and population-based serotype distribution has not stabilized. Distribution and stabilization of circulating vaccine serotypes in turn may depend on serotype distribution at the time of vaccine introduction, vaccine uptake, and the specific age groups targeted for vaccination (including the use of mass catch-up vaccination). An additional concern is that trends in respiratory viruses causing CAP may influence measured PCV VE against clinical CAP, with higher values during mild respiratory virus seasons and lower values during more severe seasons. Finally, VE estimates reported for clinical outcomes (Tables 3 and 4) have relatively wide confidence intervals, reflecting their relatively nonspecific nature. It is somewhat reassuring that point estimates across studies for hospitalized clinical CAP are relatively similar. However, additional studies in other settings are needed to better define the true rate reductions in inpatient and outpatient pneumonias achievable with adult PCV use, and the extent to which these vary by time and place.

Nevertheless, as data on adult PCV efficacy, effectiveness, and rate reductions for clinically defined pneumonia continue to accumulate, it will become increasingly untenable to underestimate the value of PCVs in economic and impact models by limiting these models to inputs based on diagnostic assays. Consequently, methodologies should be developed to allow models to include metrics based on clinical outcomes, and two examples come to mind. Models could be built using serotype-specific data (as is done currently) and then adding in a multiplier, for example the ratio of VPDIs for clinical versus etiologically/radiologically confirmed disease. E.g. in The Netherlands RCT, the VPDIs for clinical CAP and VT-CAP were 72 and 25 per 100,000 person-years of follow-up, respectively, yielding a ratio of 2.9 [6]. This method has the substantial advantage that most extraneous factors cancel in the equation, with differences across settings reflecting primarily differences in the use of UAD and CXR interpretation (as UAD and CXR represent the two differences between clinical and VT-CAP). This method has the substantial limitation that to date only one study reports efficacy/effectiveness or rate reduction values for both clinical CAP and VT-CAP. Alternatively, models could be built by directly imputing clinically defined CAP VE or VPDI. This method, however, has some of the limitations noted in the preceding paragraph. If higher valency PCVs are used in adults before pediatric introduction, the first option may be preferred as it allows accounting for geographic variations in serotype distribution. Regardless, though, the discrepancy in model outputs when using VT-CAP versus clinical CAP is likely to be so large that it demands some methodology be developed and agreed upon to account for the latter, and this is an area for further research. To do otherwise is likely to result in inaccurately conservative models that distort the decision-making process.

In sum, current data support that exclusive reliance on etiologically and radiologically confirmed CAP will substantially underestimate the benefit of adult PCV use. By including nonspecific outcomes like clinical CAP as a complement to the highly specific outcomes for assessing benefits of an immunization program, policymakers and VTCs can arrive at a more complete view of a vaccination program’s public health value. In the future, as additional data are gathered and depending on results, we propose that assessment of the public health value of adult PCVs also include prevention – if any – of pneumonia-associated declines in quality of life, long-term mortality, chronic disease exacerbations, family disruption, and long-term use of health care and long-term residential care services. As has been increasingly recognized [73], it is only through employing the most complete evidence base that the full potential value of a vaccination program can be assessed and thus better inform public health decisions.

7. Expert opinion

If regulatory agencies, and more so VTCs, focus exclusively on etiologically/radiologically confirmed pneumonia when assessing adult PCVs, they will underestimate vaccine benefits, and perhaps severely. For example, a recent manuscript from The Netherlands RCT added primary care outcomes [18] to previously published data on secondary care outcomes [6]. Vaccine efficacies were generally low, and all were <10% given the nonspecific outcomes evaluated such as all LRTI. Consequently, few outcomes achieved statistical significance, because this study did not have power to assess efficacies in this range. However, taken at face value the VE multiplied by the placebo group incidence indicated large VPDIs because the placebo group incidence was so high. For example, the VPDI associated with combined primary and secondary care LRTI – calculated by multiplying a VE of 4.4% (95% CI, −0.3% to 9.0%) times a placebo group incidence of 12,890 per 100,000 person-years of observation (PYOs) – was 570 per 100,000 PYOs. This value is almost eightfold higher than the VPDI value of 72 per 100,000 PYOs for hospitalized clinical CAP previously published [6], and higher than VPDIs for PCV-associated clinical pneumonia published for young children from South Africa [12] and Finland [5].

There are few technical barriers to considering existing data on efficacy, effectiveness, or rate reductions against clinically defined pneumonia or other respiratory disease outcomes. Most of the barriers rather are conceptual, as detailed in the current manuscript. However, none of these barriers are insurmountable, either scientifically or otherwise. If VTCs were to use existing data on clinically defined pneumonia, in line with current practice for pediatric vaccines, it would provide a more complete measure of the utility of PCV use in older individuals and allow a more accurate prioritization of health care and immunization program resources. This also would align with a current global emphasis on use of vaccines across the lifespan and not just during early childhood [74].

Additional evaluations would be useful. While current data for clinically defined hospitalized CAP VE are consistent within the range of 6% to 12%, more studies could help inform variation by time and population. Observational studies, however, are limited by the small number of countries that have implemented age-based PCV recommendations for older adults; also, residual confounding may be substantial depending on issues such as health care access or vaccination uptake among population strata with different pneumococcal risk. Additionally, in some countries with publicly funded programs, coverage is extremely low making these sites inappropriate for effectiveness evaluations. Another RCT would be preferable but impractical as these are expensive and time-consuming. Nevertheless, an opportunity for such a study may exist given that two new extended valency PCVs may be licensed in adults (a PCV15 and a PCV20).

A second avenue for study is PCV13 efficacy, effectiveness, or rate reductions on non-hospital outcomes. The German observational study of PCV13 effectiveness against clinical CAP [21] and the Dutch RCT [18] both showed efficacy or effectiveness against outpatient disease but with relatively wide confidence intervals. Moreover, few studies have evaluated efficacy or effectiveness against the entire range of potentially important outcomes such as chronic disease exacerbations, quality of life, custodial and in-home care services, long-term mortality, and family economics. Such evaluations also could help disentangle whether the occurrence of downstream effects after a pneumonia episode are caused by the pneumonia episode itself – and thus potentially preventable with vaccines – or due to the underlying conditions that resulted in the pneumonia.

As noted above, one implication of efficacy/effectiveness estimates for adult clinically defined pneumonia is that a larger proportion of this outcome than previously realized may involve the pneumococcus as part of the causal chain. New testing methods have found point prevalences of pneumococcal carriage over 20% in older adults [51,52], and these results make it at least plausible that pneumococcus could be a preventable cause of pneumonia other than through the direct prevention of pneumococcal pneumonia at the time of presentation. This could occur, for example, if pneumococcus precipitates secondary viral pneumonias or leads to exacerbations of chronic lung disease that are mistaken for pneumonia. More mundanely, current serotype-specific UAD technology, and chest -x-ray interpretation, may be insensitive for identifying the majority of non-bacteremic pneumococcal pneumonias. Future research could address all these issues to assist in providing a biological basis for the epidemiologic findings.

Over the next 5 to 10 years, it is likely that a wealth of new epidemiological and biological data will be available on the role of pneumococcus in adult disease. This will be motivated in part by the current COVID-19 pandemic, and by the introduction of new PCVs from additional manufacturers targeting adults. Such data will inform vaccine use by VTCs in adult populations. In the meantime, however, enough data exist for consideration of using VEs and VPDIs for clinically defined pneumonia as a complement to etiologically confirmed CAP when developing models and generating recommendations for PCV use in older adults. While this approach has limitations, these likely are relatively small compared to the underestimate of adult PCVs’ public health value when clinically defined outcomes are not considered.

Article highlights

• Regulatory agencies and vaccine technical committees (VTCs) routinely use clinically defined outcomes as a complement to etiologically defined outcomes in their assessments of vaccines used for children; however, this has not occurred for adult vaccines and particularly for pneumococcal conjugate vaccine (PCV) use among older adults.

• One randomized controlled study and two large observational studies of PCVs have shown vaccine efficacy/effectiveness (VE) of 6-12% against hospitalized or hospitalized plus outpatient clinically defined pneumonia, yet VTCs often do not use these data to assess PCV cost-effectiveness of adult PCV recommendations.

• We review potential epidemiologic and regulatory critiques to the use of PCV VEs for clinical CAP among older adults and provide a rationale for use of this increasingly large body of data in adult PCV decision-making.

• We describe the recently documented plateau in reductions of vaccine-type pneumococcal disease among adults following PCV use in children; PCV VE against serotype 3 non-bacteremic CAP; the relationship between PCV VEs for clinically defined CAP and the fraction of CAP with a confirmed vaccine-type pneumococcal etiology; and the lack of sensitivity of currently available diagnostic tests (such as urinary antigen detection) for serotype-specific pneumococcal identification.

• Because incorporation of clinical CAP outcomes into VTC decision-making and economic models provides a more comprehensive and accurate assessment of the expected public health value for PCV use in adults, regulatory agencies and VTCs might consider this step as a complement to the use of etiologically and radiologically defined CAP.

• Additional research should confirm efficacy/effectiveness estimates from these initial studies; expand outcomes to broader benefits such as outpatient disease, chronic disease exacerbations, and quality of life; and further assess the biologic basis for the observed data.

Declaration of interest

All authors are employees of Pfizer, which sponsored the manuscript. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

Peer reviewers on this manuscript have received an honorarium for their review work. Peer reviewers have no other relevant financial or other relationships to disclose.

Additional information

Funding

This paper was funded by Pfizer.