Receipt of the 2010–2011 seasonal influenza vaccine was associated with a 42% reduced risk of laboratory-confirmed influenza hospitalization among community-dwelling elderly adults. Although significant VE against influenza A subtypes (A/H3N2 and A/H1N1) was demonstrated, we detected no significant VE against influenza B. With only 33 individuals who tested positive for influenza B, our study may not have been sufficiently powered to detect VE against influenza B. Our estimates of VE remained robust to numerous subgroup and sensitivity analyses, and the absence of an association between influenza hospitalization and our tracer exposure suggests that these results are not due to underlying differences in healthcare-seeking behavior between test-positive and test-negative individuals.
An estimated VE of 42% is consistent with the only RCT evaluating the efficacy of TIV against serologically confirmed influenza infection among elderly patients (VE = 58%; 95% CI, 26%–77%) [6
]. Our results are also consistent with 2 studies using the test-negative design to evaluate the 2010–2011 seasonal influenza vaccine, where a VE of 38% (95% CI, −16% to 67%) was estimated for adults ≥65 years in the United States [28
], and a VE of 60% (95% CI, 17%–81%) was estimated for adults ≥60 years in a multicenter European study; both used laboratory-confirmed influenza infection as the outcome [29
]. Although our VE estimate is similar to previous evaluations in elderly adults, those studies did not examine VE against influenza hospitalizations.
Only 2 prior studies have examined VE against laboratory-confirmed influenza hospitalizations in older adults. Talbot et al reported a propensity score-adjusted VE of 61% (95% CI, 18%–82%) among community-dwelling adults aged ≥50 years [30
]. However, in that study only 28% of test-positive influenza hospitalizations occurred among those aged ≥65 years (most were in those aged 50–64 years), and VE was imprecisely estimated for the ≥65 years age group (VE = 61%; 95% CI, −48% to 90%). Another study estimated a VE of 59% (95% CI, 4%–83%) against laboratory-confirmed influenza hospitalizations for 2010–2011 [31
], but only 45% of the test-positive patients were ≥65 years and VE estimates were not stratified by age group.
A recent study of the 2011–2012 seasonal influenza vaccine suggests that vaccine-induced protection may wane over time within an influenza season [32
]. In this study, the VE point estimates declined both by month (51% in December 2010 vs 47% in January 2011 vs 34% in February/March/April 2011) and when dividing the study period into periods before and after peak influenza activity, but the confidence intervals overlapped considerably and interaction tests were not statistically significant. Future studies with larger samples sizes will be required to investigate this possibility.
In the most recent Cochrane Collaboration review of influenza vaccines for elderly adults, vaccinated community-dwelling elderly were found to be at reduced risk of pneumonia and influenza (P&I) hospitalizations during defined influenza seasons compared with unvaccinated individuals (VE = 27%; 95% CI, 21%–33%) [9
]. However, only 30% of all P&I hospitalizations during discrete influenza seasons have been statistically estimated to be attributable to influenza viruses [33
]. Our study estimated VE against laboratory-confirmed influenza hospitalizations, which is a much more specific outcome for serious influenza infection than P&I-coded hospitalizations in discharge abstracts. Estimating VE against nonspecific outcomes can magnify the bias inherent in observational studies of VE in the elderly, as measured and unmeasured characteristics may differ between those receiving the vaccine and those not receiving the vaccine, leading to exaggerated vaccine benefits [14
Several limitations of this study merit emphasis. First, as symptom onset date was available for only 16% of specimens, we could not restrict our study sample to individuals for whom symptom onset was within 7 days of specimen collection; this is a common inclusion criteria for most test-negative studies to ensure influenza can still be detected [17
]. However, among participants with a recorded symptom onset date, 96% of test-positive individuals and 89% of test-negative individuals were tested within 7 days of symptom onset, suggesting the absence of a large difference between the 2 groups. Second, the specimens were collected as part of routine clinical care rather than through systematic screening and enrollment. With no standard case definition for testing hospitalized patients for influenza, it is possible that a physician's decision to order a clinical test may have been influenced by the patient's prior vaccination status. However, in a post hoc analysis, we explored differences in influenza PCR testing rates among vaccinated and unvaccinated hospitalized patients in Ontario. Among 81 398 hospital admissions of vaccinated individuals occurring between 1 December 2010 and 30 April 2011, 1194 were tested for influenza (1.47% of admissions), compared to 1045 individuals tested for influenza during 62 975 admissions of unvaccinated individuals (1.66% of admissions). Therefore, unvaccinated individuals are more likely to be tested for influenza than vaccinated individuals (crude odds ratio = 1.13; 95% CI, 1.04–1.23). In a multivariable model incorporating the variables described in the “Covariates” section of the Methods, the adjusted odds ratio was 1.21 (95% CI, 1.11–1.32). The 21% relative increase in testing for unvaccinated individuals compared to vaccinated individuals suggests that PCR testing for influenza in inpatient settings is not dramatically affected by a patient's vaccination status. Such a difference in testing patterns would likely have only a small impact on VE estimates; using a simulation model, Ferdinands et al demonstrated that if unvaccinated patients are 10% more likely to be tested than vaccinated patients, the estimated VE would be 73% instead of a true VE of 70% [35
]. Third, vaccination status could have been misclassified due to receipt of influenza vaccines outside of physician offices (estimated to be about 25% of those vaccinated among elderly adults). Test-positive individuals have been found to be similar to test-negative individuals in terms of healthcare-seeking behavior [17
], so any misclassification would likely be nondifferential and cause underestimation of VE, although the possibility of differential misclassification exists. Fourth, we were unable to control for potential bias that might result from viral interference, as infection with a noninfluenza virus may offer protection against influenza infection [36
]. Finally, as with all observational study designs, the possibility of residual confounding remains, although no studies have demonstrated substantial confounding bias when using the test-negative design.
Despite the limitations, the test-negative design allowed for the assessment of influenza vaccine benefit against a highly specific and serious outcome of influenza infection. Linking routinely collected laboratory and health administrative data enabled us to conduct, to our knowledge, the largest evaluation of VE against laboratory-confirmed influenza hospitalizations among elderly adults in an inexpensive and efficient manner. Such studies complement prospective test-negative studies that involve primary data collection.
Although the benefits of influenza vaccines for preventing serious influenza outcomes in older adults have been uncertain given the scarcity of RCT evidence and the selection bias and outcome specificity issues noted in prior observational studies [9
], the results of this study suggest that the 2010–2011 influenza vaccine was 42% effective in reducing laboratory-confirmed influenza hospitalizations among elderly adults; this estimate may better inform decision making and vaccination policy in this current area of controversy. Future studies that link routinely collected laboratory data to assess influenza VE should incorporate additional clinical information to ensure greater homogeneity of the study population, if it is not possible to institute systematic selection of hospitalized patients for influenza testing. Similar analyses are needed annually due to antigenic drift and frequent changes in influenza vaccine composition. These findings support the current recommendations of vaccinating adults aged ≥65 years to prevent serious outcomes of influenza infection.