Sensitivity Analysis for an Unmeasured Confounder

In this week's issue of NEJM, a group of investigators examine the association between influenza vaccination and outcomes in multiple HMO cohorts. However, a major limitation of the HMO cohorts was the lack of covariate data on socioeconomic status and other sociological factors that may potentially confound ...

Thus, the analysis conducted special sensitivity analsysis assuming the existence of an unknown confounder. Using the confounding triangle paradigm, they 1) assumed that a confounder is associated with the exposure with a RR=2 (i.e. low SES individuals are twice as likely to not receive vaccination), and 2) simultaneously assumed that the confounding factor of low SES associated with outcome under varying prevalence scenarios of the confounder, ranging from RR=2 to RR=3, and finally 3) varied the prevalence of the confounder from 0% (no confounding) to 60% under different scenarios. (The sensitivity analysis method applied utilized the method of Lin et al. Biometrics 1998)

All in all, the protective association between influenza vaccination and outcomes (hospitalization for influenza or pneumonia, or death) remained significant across all scenarios, suggesting that unmeasured confounding by a single powerful confounder, with associations of RR=2-3 strength in the confounding triangle, could not have explained the association.

In future epidemiologic settings, perhaps we should all consider performing such detailed sensitivity analyses to assess the robustness of our own findings...

References:

Nichol KL, Nordin JD, Nelson DB, Mullooly JP, Hak E. Effectiveness of influenza vaccine in the community-dwelling elderly. N Engl J Med 2007;357:1373-1381
http://content.nejm.org/cgi/content/abstract/357/14/1373

Lin DY, Psaty BM, Kronmal RA. Assessing the sensitivity of regression results to unmeasured confounders in observational studies. Biometrics 1998;54:948-963