1 Entropy Research Institute, Upper Saddle River, NJ.
2 National Institute of Allergy and Infectious Diseases, Bethesda, MD.
3 Vyteris, Inc., Fairlawn, NJ.
4 Becton Dickinson and Company, Inc., Franklin Lakes, NJ.
5 Sergy Haut, France.
6 Fogarty International Center, National Institutes of Health, Bethesda, MD.
In economically developed countries, mortality increases distinctly during winter. Many causes have been suggested, including light-dark cycles, temperature/weather, and infectious agents. The authors analyzed monthly mortality in the United States during the period 1959–1999 for four major disease classes. The authors isolated the seasonal component of mortality by removing trends and standardizing the time series. They evaluated four properties: coincidence in mortality peaks, autocorrelation structure and autoregressive integrated moving average (ARIMA) models, magnitude, and age distribution. Peak months of mortality for ischemic heart disease, cerebrovascular disease, and diabetes mellitus coincided appropriately with peaks in pneumonia and influenza, and coefficients of autocorrelation and ARIMA models were essentially indistinguishable. The magnitude of the seasonal component was highly correlated with traditional measures of excess mortality and was significantly larger in seasons dominated by influenza A(H2N2) and A(H3N2) viruses than in seasons dominated by A(H1N1) or B viruses. There was an age shift in mortality during and after the 1968/69 pandemic in each disease class, with features specific to influenza A(H3N2). These findings suggest that the cause of the winter increase in US mortality is singular and probably influenza. Weather and other factors may determine the timing and modulate the magnitude of the winter-season increase in mortality, but the primary determinant appears to be the influenza virus.
Key Words: age distribution; cause of death; influenza; models, statistical; mortality; respiratory tract infections; seasons
Abbreviations: ARIMA, autoregressive integrated moving average; ZLS, z-like score.