Measuring the impact of influenza on mortality

Abstract

Statistical methods for the analysis of time series data are commonly applied to quantify the population mortality burden of influenza. Here, I examined the methodology used to make these estimates, and applied modified methods to data from China including mainland and Hong Kong. I first reviewed studies of influenza-associated excess mortality, identifying 103 published papers, and classified the methods used into time series models with an influenza virus activity proxy (52.4%), Serfling-type models (27.2%), other time series models (27.2%), and multiplier methods (3.9%). The impacts of methodological choices on the estimates of influenza-associated excess mortality were evaluated using a random-effect meta-regression model. The influenza-associated excess mortality rates estimated from Serfling-type models were found to be higher than the estimates based on multiplier methods. The proportions of pneumonia and influenza (P&I), respiratory, and respiratory and cardiovascular (R&C) excess mortality in all influenza-associated excess mortality were estimated to be 16.5% (95% confidence interval (CI): 13.1%, 20.7%), 32.1% (95% CI: 26.9%, 38.2%), and 71.3% (95% CI: 64.1%, 79.3%), respectively for older adults.

I analyzed data on 244,800 respiratory deaths in mainland China between 2010/11 and 2014/15, from a mortality reporting system covering approximately 4.2% of the whole population. Based on the linear regression models fitted for the 22 provinces and the meta-regression models constructed to extrapolate the influenza-associated excess respiratory mortality rates in the remaining nine provinces, it was estimated that an average of 88,100 (95% CI: 84,200, 92,000) influenza-associated respiratory excess deaths occurred in mainland China annually, corresponding to 8.2% (95% CI: 7.9%, 8.6%) of all annual respiratory deaths.

In Hong Kong, it was estimated that influenza virus was on average associated with 1,053 (95% CI: 982, 1,122) all-cause excess deaths annually. The excess mortality in people aged ≥85 years and ≥65 years accounted for approximately 50% and 95% of influenza-associated excess deaths, respectively. The all-age mortality per unit influenza virus activity proxy for A(H3N2) and B increased with the proportion of people aged ≥65 years. The impact of influenza A(H3N2) virus infections on population mortality increased steadily under ‘extreme’ high temperatures.

In conclusion, influenza virus results in substantial mortality. Standardization of the methodology used for estimating influenza-associated excess mortality could improve comparability of estimates among different study locations. Estimates of influenza-associated all-cause excess mortality tended to be approximately three times the estimates of influenza-associated respiratory excess mortality. Substantial influenza-associated respiratory excess mortality burden was observed in mainland China and the burden varied by age group and across provinces. Ambient temperature could be an effect modifier of the associations between influenza A(H3N2) virus activity and population mortality in tropical or subtropical regions.