Estimation of seasonal influenza vaccine effectiveness using data collected in primary care in France: comparison of the test-negative design and the screening method

Abstract

ObjectivesWe discussed which method between the test-negative design (TND) and the screening method (SM) could provide more robust real-time and end-of-season vaccine effectiveness (VE) estimates using data collected from routine influenza surveillance in primary care. MethodsWe used data collected during two influenza seasons, 2014–15 and 2015–16. Using the SM, we estimated end-of-season VE in preventing medically attended influenza-like illness and laboratory-confirmed influenza among the population at risk. Using the TND, we estimated end-of-season VE in preventing influenza among both the general and the at-risk population. We estimated real-time VE using both methods. ResultsFor the SM, the overall adjusted end-of-season VE was 24% (95% confidence interval (CI), 16 to 32) and 12% (95% CI, −16 to 33) during season 2014–15, and 53% (95% CI, 44 to 60) and 47% (95% CI, 23 to 64) during season 2015–16, in preventing influenza-like illness and laboratory-confirmed influenza, respectively. For the TND, the overall adjusted end-of-season VE was −17% (95% CI, −79 to 24) and −38% (95% CI, −199 to 13) in 2014–15, and 10% (95% CI, −31 to 39) and 18% (95% CI, −33 to 50) in 2015–16, among the general and at-risk population, respectively. Real-time VE estimates obtained through the TND showed more variability across each season and lower precision than those estimated with the SM. ConclusionsAlthough the worldwide use of the TND allows for comparison of overall VE estimates among countries, the SM performs better in providing robust real-time VE estimates among the population at risk.