Estimates of influenza vaccine effectiveness in primary care in Scotland vary with clinical or laboratory endpoint and method--experience across the 2010/11 season

Aim This study examines estimation of seasonal influenza vaccine effectiveness (VE) for a cohort of patients attending general practice in Scotland in 2010/11. The study focuses on the variation in estimation of VE for both virological and clinical consultation outcomes and understanding the dependency on date of analysis during the season, methodological approach and the effect of use of a propensity score model. Methods For the clinical outcomes, three methodological approaches were considered; adjusted Poisson multi-level modelling splitting consultations in vaccinated individuals into those before and after vaccination, adjusted Cox proportional hazards modelling and finally the screening method. For the virological outcome, the test-negative case–control study design was employed. Results VE was highest for the most specific outcomes of ILI (Poisson end-of-season VE = 47% (95% CI: −69%, 83%); Cox VE = 34% (95% CI: −64%, 73.2%); Screening VE = 52.8% (95% CI: 3.8%, 76.8%)) and a virological diagnosis (VE = 54% (95% CI: −37%, 85%)). Using the Cox approach, adjusted for propensity score only gave VE = 46.5% (95% CI: −30.4%, 78.0%). Conclusion Our approach illustrated the ability to achieve relatively consistent estimates of seasonal influenza VE using both specific and less specific outcomes. Construction of a propensity score and use for bias adjustment increased the estimate of ILI VE estimated from the Cox model and made estimates more similar to the Poisson approach, which models differences in consultation behaviour of vaccinated individuals more inherently in its structure. VE estimation for the same data was found to vary by methodology which should be noted when comparing results from different studies and countries.