Background. We estimated the effectiveness of inactivated influenza vaccines for the prevention of laboratory-confirmed, medically attended influenza during 3 seasons with variable antigenic match between vaccine and patient strains.

Methods. Patients were enrolled during or after a clinical encounter for acute respiratory illness. Influenza infection was confirmed by culture or reverse-transcriptase polymerase chain reaction. Case-control analyses were performed that used data from patients who were ill without influenza (hereafter, “test-negative control subjects”) and data from asymptomatic control subjects from the population (hereafter, “traditional control subjects”). Vaccine effectiveness (VE) was estimated as [100 × (1 − adjusted odds ratio)]. Influenza isolates were antigenically characterized.

Results. Influenza was detected in 167 (20%) of 818 patients in 2004–2005, in 51 (14%) of 356 in 2005–2006, and in 102 (11%) of 932 in 2006–2007. Analyses that used data from test-negative control subjects showed that VE was 10% (95% confidence interval [CI], −36% to 40%) in 2004–2005, 21% (95% CI,−52% to 59%) in 2005–2006, and 52% (95% CI, 22% to 70%) in 2006–2007. Using data from traditional control subjects, VE for those seasons was estimated to be 5% (95% CI, −52% to 40%), 11% (95% CI, −96% to 59%), and 37% (95% CI, −10% to 64%), respectively; confidence intervals included 0. The percentage of viruses that were antigenically matched to vaccine strains was 5% (3 of 62) in 2004–2005, 5% (2 of 42) in 2005–2006, and 91% (85 of 93) in 2006–2007.

Conclusions. Influenza VE varied substantially across 3 seasons and was highest when antigenic match was optimal. VE estimates that used data from test-negative control subjects were consistently higher than those that used data from traditional control subjects.