Mortality attributable to seasonal influenza in Greece, 2013-2017: variation by type and age, and a possible harvesting effect

BACKGROUND Estimating the contribution of influenza to excess mortality in the population presents substantial methodological challenges. We combined environmental, epidemiological and laboratory surveillance data to estimate influenza-attributable mortality in Greece, over four seasons (2013-2014 to 2017-2018), specifically addressing the lag dimension and the confounding effect of temperature. METHODS Associations of influenza type-specific incidence proxies and of daily mean temperature with mortality were estimated with a distributed-lag non-linear model with 30 days of maximum lag, separately for each age group. Total and weekly deaths attributable to influenza and cold temperatures were calculated. RESULTS Overall influenza-attributable mortality was 23.6 deaths per 100,000 population (95%CI: 17.8, 29.2), and varied greatly between seasons, by influenza type and by age group, with the vast majority occurring in persons 65 years or older. Most deaths were attributable to A/H3N2, followed by type B influenza. During periods of A/H1N1 circulation, weekly attributable mortality to this subtype among older people increased rapidly in the first half, but then fell to zero and even negative, suggesting a mortality displacement (harvesting) effect. Mortality attributable to non-optimum temperatures was much higher than that attributable to influenza. CONCLUSIONS Studies of influenza-attributable mortality need to take distributed-lag effects into account, stratify by age group and adjust for circulating influenza types and daily mean temperatures, in order to produce reliable estimates. Our approach is useful and readily applicable in the context of influenza surveillance.