Case-control Study Shows Flaw in Flu Shot Efficacy Assumption

A University of Alberta case-control study published Sept 2008 in the American Journal of Respiratory and Critical Care Medicine examined the efficacy of influenza vaccine against deaths that occur outside the flu season. In analyses which didn’t adjust for confounders, vaccinated pneumonia patients were found to be significantly less likely to die than those unvaccinated. However, when adjustments were made using all available data including that for socioeconomic status and frailty, the difference was insignificant.
Since, in a non-flu-season, the pneumonia cases examined couldn’t have been preceded by influenza, the researchers explained that the difference may have been due to a “healthy-user effect”, meaning that those who accept vaccinations are likely to be more health conscious than those who don’t. To equate acceptance of a flu shot with health consciousness is questionable; a flu shot may be seen by some as an easy way to avoid illness while continuing to live unhealthily. A more likely reason for the loss of significance when socioeconomic status and frailty were accounted for is that it is mainly these factors, not lack of vaccination, that lead to death. To his credit, co-investigator, Dean Eurich, PhD, admitted, “only about 10 percent of winter-time deaths in the United States are attributable to influenza, thus to suggest the vaccine can reduce 50% of deaths from all causes is implausible in our opinion.”
In an analysis and comment in the Oct 28, 2006 British Medical Journal, Dr Tom Jefferson stated: “A metaanalysis of inactivated vaccines in elderly people showed a gradient from no effect against influenza or influenza-like illness to a large effect (up to 60%) in preventing all-cause mortality. These findings are both counterintuitive and implausible, as other causes of death are far more prevalent in elderly people even in the winter months.15 16 It is impossible for a vaccine that does not prevent influenza to prevent its complications, including admission to hospital.
A more likely explanation for such a finding is selection bias, where one half of the study population (hemi-cohort) systematically differs from the other in one or more key characteristics.14-16 In this case, the vaccinated hemi-cohort may have been more mobile, healthy, and wealthy than the control hemi-cohort, thus explaining the differences in all-cause mortality.11 14 The same effect is seen in stronger study designs (such as cluster randomised trials) that are badly executed, which introduces bias.10 Its presence seems to be a marker of confounders that persist even after adjusting for known ones, and it makes accurate interpretation of the data difficult. Caution in interpretation should thus be the rule, not the exception. This problem (in the opposite direction—with frailer people more likely to be vaccinated) has been identified before but not heeded.17 The only way that all known and unknown confounders can be adequately controlled for is by randomisation.”
http://www.medpagetoday.com/InfectiousDisease/URItheFlu/tb/10731
http://www.bmj.com/cgi/content/full/333/7574/912