In summary, 7vCRM efficacy against all-cause AOM episodes was an estimated 0%–9% in randomized trials and 17%–23% in nonrandomized clinical trials. Observational database studies showed that OM visit rates decreased 19% on average following 7vCRM introduction, with estimates ranging widely (+7% to −48%). Before 7vCRM introduction, OM visit rates were already declining in all but one study. These findings raise several issues to be considered when appraising policy options and designing studies, as discussed below.
Variability in Efficacy Trial Results
Efficacy against all-cause AOM assessed for 3 vaccine formulations (7vCRM, 7vOMPC, and 11Pn-PD) tested in 5 randomized trials yielded point estimates ranging from −1% to 34%. Whereas some of this variability is likely due to differing vaccine composition, it is difficult to distinguish this from confounding by local variability in viral and bacterial etiology, case ascertainment, diagnosis, and care-seeking behavior. Re-analyses of POET and FinOM trials, adjusting for severity of case definition and pathogen distributions, somewhat narrowed differences in PCV efficacy estimates [27
], but such reconciliation is not always feasible, and conclusions have to be based on central tendency across studies.
Observed Versus Theoretical Effectiveness
Theoretical maximum effectiveness in real-life settings can be calculated by assuming no replacement with nonvaccine types, a stationary commensal profile, and 100% vaccine uptake. On the assumption that 70% of AOM episodes are bacterial [2
], of which 50% are due to S. pneumoniae
], of which 7vCRM serotypes represent 75% [29
], and for which efficacy is 57% [3
], then 7vCRM should prevent approximately 15% of the episodes of all-cause AOM (70% × 50% × 75% × 57%).
Vaccination rates of >80% would be expected to induce herd protection via decrease in nasopharyngeal carriage of vaccination serotypes [30
]. Although there is strong evidence for herd protection with IPD, herd protection against vaccine-type AOM in nonvaccinated age groups has not yet been directly demonstrated because tympanocentesis is not routinely performed. Dilution of vaccine-type herd protection within all-cause OM makes it hard to show, and one study failed to detect it in overall AOM visits in older children 2 years after 7vCRM implementation [19
]. Near elimination of vaccine-type carriage some years after PCV use, on the assumption of maximum herd protection (with vaccine types eradicated), is a reasonable approximation. Effectiveness against vaccine types would then be 100% instead of 57%, yielding a theoretical effectiveness of approximately 26%.
The above calculations do not reflect any replacement with nonvaccine serotypes and bacteria, although some replacement is suggested in clinical studies and postintroduction surveillance [3
]. Indeed, a recent model that used actual nasopharyngeal carriage rates in US children for both vaccine and nonvaccine serotypes, taking into account their specific abilities to cause AOM, projected a maximum theoretical effectiveness of 7vCRM against overall AOM of only 12% [34
]. This suggests that estimates well beyond these theoretical limits may be substantially confounded and biased.
Variability in Baseline Incidence
Baseline AOM episode rates in clinical trials varied 10-fold. The high baseline rate in FinOM [3
] is similar to US rates (900–1500 AOM episodes per 1000 children) [20
], whereas the low rate in the POET trial is closer to those reported in other European studies (154–400 AOM episodes per 1000 PY) [37
]. In general, stronger vaccine effects would be expected on samples that use tighter diagnostic definitions and, hence, lower baseline case incidence, but they face sample size challenges. This, plus possible intrinsic differences in populations or differences in healthcare uptake beyond those of diagnostic definition, suggest that one should be cautious in considering between-study comparisons of vaccines. Among the database analyses, baseline rates also varied across studies, even after taking into account age differences [12
]. Strong evidence for demographic, immunological, or microbiological differences between such populations is lacking, so such baseline rate differences are more appropriately attributed to differences in case severity or diagnostic code for case definition.
Changes Before Versus After Vaccine Introduction
In observational database studies, OM visit rates decreased by 19% on average after 7vCRM introduction. However, among studies also presenting data before 2000, all [12
] but one [16
] observed OM visits declining by 15% on average before 7vCRM introduction. This suggests that long-term decreases in consultations before 7vCRM introduction, which are unlikely to have halted, have added to apparent postintroduction decreases. Poehling et al and Grijalva et al controlled for annual trend via differential effect by age, arriving at 4%–19% decreases due to 7vCRM [12
]. However, this minimizes any herd protection affecting the nonimmunized portion of the younger cohort. In addition, non–vaccine-related factors, such as age stratification of <2/≥2 years in antibiotic prescription guidelines, could affect OM visit rates over time differentially by age. De Wals et al moved in the appropriate direction by estimating a post-PCV rate with time-series regression to adjust for annual trend [19
]; the raw decrease in OM claims in 2000–2007 was 25%, but the adjusted decrease attributable to 7vCRM was only approximately 13%.
To determine whether the decrease in consultations is due to 7vCRM introduction, analysis must be made over a few years and according to when and to what extent the vaccine was introduced. For example, a recent study in an Athens hospital found that, beginning in 2005, emergency department visits by children aged <15 years decreased by 38% and 48% for all-cause and pneumococcal otorrhea, respectively [38
]. However, this drop occurred 1–2 years before mass pneumococcal vaccination in Greece, at the time of (presumably low) private market 7vCRM use, and, even after the decrease, vaccine serotypes still represented the majority of pneumococcal otorrhea. Upon implementation of mass vaccination in 2006, no further drop was seen, indicating that the reduction in 2005 was largely due to nonvaccine factors.
Potential Nonvaccine Factors
Several other factors might explain why OM rates decreased before PCV introduction and continued decreasing after. First, changes in AOM perception, consultation rates, and frequency and type of antibiotic use date from the early 1990s. The increasing acceptance by parents and physicians of observation without antibiotic use (“watchful waiting”), which is officially recommended for some AOM patients [39
], could reduce the apparent AOM incidence if parents do not consult physicians for mild AOM if they expect little benefit for their child. Stricter diagnostic criteria [39
] may have reduced not only inappropriate antibiotic use [13
] but also apparent AOM consultation rates. Second, a shift to higher antibiotic dosage or the doubling of long-acting macrolide use in US children around the same time as 7vCRM introduction [40
] could have reduced relapses and, therefore, reduced the total number of AOM visits per episode, thus reducing the healthcare burden [17
Third, awareness of vaccination status could affect care-seeking behavior. In a recent observer-blinded randomized trial in Sweden of children at risk for recurrent AOM conducted before universal PCV, receipt of 7vCRM reduced overall reported AOM episodes by 26% and AOM hospital visits by 36% [41
]. Because these apparent effects are larger than the above theoretical effectiveness estimate, there may have been some differential contribution from parents seeking medical assistance depending on vaccination status, with less care-seeking for vaccinated children because of the belief that vaccine would probably prevent the more serious forms or complications of disease.
Fourth, the decline in OM rates has paralleled the decreasing exposure of children to secondhand tobacco smoke, a strong AOM risk factor [42
]. Fifth, influenza vaccination can reduce AOM incidence during the influenza season by reducing viral coinfection [43
]. However, influenza routine vaccination in the US began in 2004, with the sharpest increase around 2007–2008 [44
], after the attributable post-7vCRM decrease in OM.
Possible differences among populations, chiefly their relative risks, cannot be overlooked in explaining the heterogeneity of results. However, convincing demonstrations are lacking. The failure of O'Brien et al to detect a statistically significant 7vCRM impact on AOM in high-risk American Indians may be due to the lack of statistical power [20
]. Likewise, a favorable, although nonsignificant, vaccine effect (adjusted relative risk, 0.88 [95% CI, .69–1.13]) was found in successive cohorts of 51 nonvaccinated and 97 vaccinated (7vCRM plus a 23-valent polysaccharide booster dose) high-risk Australian aboriginal children [45
]. Finally, the authors of the nonrandomized, nonblinded 7vCRM German trial [23
] suggested that the achievable efficacy was possibly biased against the vaccine because more children in the 7vCRM group than the control group had a medical risk factor (66% vs 18%) or were born prematurely (40% vs 6%). The assumption behind all these studies is that high-risk, otitis-prone children generate a weaker immune response, for which there is some evidence [46
]. Limited statistical power currently prevents clear conclusions, but possible differences in vaccine effectiveness between populations deserve consideration.
Diagnostic Codes Included as OM
Observational database studies identify OM cases according to broad diagnostic codes that are based often on a single clinician's judgment rather than on precise protocols and measurements. In the International Classification of Diseases, Ninth Revision
coding system, codes 381.x refer mainly to nonsuppurative AOM, codes 382.x to suppurative AOM, and codes 383.x to mastoiditis. Code choice could greatly affect absolute OM visit count, and study-specific differences in case definition or even OM type distribution could influence 7vCRM effectiveness estimates [28
]. Unfortunately, no studies reported the proportions of the different codes used. Grijalva et al defined OM diagnosis as 381.x–382.x in one study [12
] and as 381.x–383.x in the other [13
], whereas Poehling et al used 381.0–381.4 and 382.x [15
]. Zhou et al used 381.00–381.6, 382.00–382.02, 382.3, and 382.9 [16
] but, unlike the other studies, only considered first-listed codes, possibly explaining why they reported the largest decrease (43%) [16
]. Indeed, where AOM antibiotic use is strictly controlled, some physicians may use AOM less as a primary code, preferring a symptom-based equivalent code.
Design Considerations for Future Studies
Vaccine impact will always be assessed by large observational studies. However, one key requirement is adjustment for non–vaccine-related confounders. Adjustment for secular trends [48
], preferably via time-series modeling [19
], should always be performed. Modeling would also allow distinction between year-to-year variation (random and viral) and longer-term trends. At a minimum, projections from prevaccine trends should provide the expected null value from which an observed deviation may be taken as evidence for vaccine effect [48
]. In addition, measurement of time trends of other diseases could provide additional control, with the caveat that some nonvaccine trends could affect unrelated diseases differently.
The central public health questions are whether vaccination causes an overall decrease in AOM and associated healthcare burden. Tympanocentesis-based efficacy studies, even at the population level, would at least help specify how much of an overall decrease is limited to target pathogens/serotypes, but it remains unusual and ethically problematic to perform routinely, and determining vaccine effectiveness against individual serotypes necessitates large sample sizes.
PCV effects on AOM can be measured economically and with good control in case-control studies, as for IPD [51
]. However, finding appropriate controls in a well-immunized population is difficult. The presumably present herd protection is seen as a depressed incidence in controls and is not directly measurable with this design, meaning the effect is nearer to an efficacy than to an effectiveness estimate.
Finally, the problem of quality of case definition has long been remarked in AOM studies. Some hope of reducing variability from this source is given by 2 recent high-quality randomized studies on AOM treatment that used stringent and reproducible criteria applicable to all designs except routine practice databases [54
In conclusion, observed OM visit rates have decreased by approximately 19% following 7vCRM introduction, but long-term reductions in OM visits preceding 7vCRM introduction of approximately 15% suggest that continuing influences other than PCV vaccination have caused some of the subsequent reduction. Caution is therefore needed in the report and interpretation of these data, and no single study should be quoted as representing the “true” effect of 7vCRM on AOM. Study methods need to be improved to more accurately estimate true PCV effectiveness.