Background — Estimates of influenza vaccine effectiveness in elderly individuals are largely from observational studies, which are susceptible to bias. Instrumental variable (IV) methods control for overt and hidden biases in observational studies.
Methods — The researchers used linked health administrative databases in Ontario to examine the association between influenza vaccination and all-cause mortality among community-dwelling individuals older than 65 years for 9 influenza seasons (2000–2001 to 2008–2009). They examined the composite of hospitalization for pneumonia and influenza and all-cause mortality as a secondary outcome, and used logistic regression modeling and IV analysis to remove the effect of selection bias.
Results — The researchers included 12 621 806 person–influenza seasons of observation. Logistic regression produced adjusted odds ratios of 0.67 (95% CI, 0.62–0.72) for all-cause mortality during influenza seasons and 0.85 (0.83–0.86) during post–influenza seasons when influenza is not circulating, suggesting the presence of bias. In contrast, IV analysis yielded adjusted odds ratios of 0.94 (95% CI, 0.84–1.03) during influenza seasons and 1.13 (1.07–1.19) during post–influenza seasons. For the composite of hospitalization for pneumonia and influenza and death, logistic regression produced adjusted odds ratios of 0.74 (95% CI, 0.70–0.78) during influenza seasons and 0.88 (0.87–0.90) during post–influenza seasons, whereas IV analysis produced adjusted odds ratios of 0.86 (95% CI, 0.79–0.92) and 1.02 (0.97–1.06), respectively.
Conclusions — Influenza vaccination is associated with reductions in the composite of hospitalizations for pneumonia and influenza and all-cause mortality during the influenza season but not mortality alone. Compared with standard modeling, IV analysis appears to produce less-biased estimates of vaccine effectiveness.
View full text