[Estimates of influenza-associated excess mortality by three regression models in Shanxi Province during 2013-2017]

Objective: Using three models too estimate excess mortality associated with influenza of Shanxi Province during 2013-2017. Methods: Mortality data and influenza surveillance data of 11 cities of Shanxi Province from the 2013-2014 through 2016-2017 were used to estimate influenza-associated all cause deaths, circulatory and respiratory deaths and respiratory deaths. Three models were used: (i) Serfling regression, (ii)Poisson regression, (iii)General line model. Results: The total reported death cases of all cause were 157 733, annual death cases of all cause were 39 433, among these cases, male cases 93 831 (59.50%), cases above 65 years old 123 931 (78.57%). Annual influenza-associated excess mortality, for all causes, circulatory and respiratory deaths, respiratory deaths were 8.62 deaths per 100 000, 6.33 deaths per 100 000 and 0.68 deaths per 100 000 estimated by Serfling model, respectively; and 21.30 deaths per 100 000, 16.89 deaths per 100 000 and 2.14 deaths per 100 000 estimated by General line model, respectively; and 21.76 deaths per 100 000, 17.03 deaths per 100 000 and 2.05 deaths per 100 000, estimated by Poisson model, respectively. Influenza-related excess mortality was higher in people over 75 years old; influenza-associated excess mortalityfor all causes, circulatory and respiratory deaths, respiratory deaths were 259.67 deaths per 100 000, 229.90 deaths per 100 000 and 32.63 deaths per 100 000, estimated by GLM model, respectively; and 269.49 deaths per 100 000, 233.69 deaths per 100 000 and 31.27 deaths per 100 000, estimated by Poisson model,respectively. Conclusion: Excess mortality associated with influenza mainly caused by A (H3N2), Influenza caused the most associated death amongold people.

Estimating the Baseline and Threshold for the Incidence of Diseases with Seasonal and Long-Term Trends

In epidemiology, it is very important to estimate the baseline incidence of infectious diseases. From this baseline, the epidemic threshold can be derived as a clue to recognize an excess incidence, i.e. to detect an epidemic by mathematical methods. Nevertheless, a problem is posed by the fact that the incidence may vary during the year, as a rule, in a season dependent manner. To model the incidence of a disease, some authors use seasonal trend models. For instance, Serfling applies the sine function with a phase shift and amplitude. A similar model based on the analysis of variance with kernel smoothing and Serfling's higher order models, i.e. models composed of multiple sine-cosine function pairs with a variably long period, will be presented below. Serfling's model uses a long-term linear trend, but the linearity may not be always acceptable. Therefore, a more complex, long-term trend estimation will also be addressed, using different smoothing methods. In addition, the issue of the time unit (mostly a week) used in describing the incidence is discussed.