ABSTRACT

Influenza is a major cause of mortality due to cardio-respiratory diseases in temperate regions, as well as in tropical and subtropical regions where seasonal patterns of influenza epidemics are less identifiable. There is a need to adjust for influenza epidemics in air pollution daily time-series studies, but because of lack of virology data they are defined according to occurrence of overall respiratory mortality without validation. In this study, we assess the validity of overall respiratory mortality as a measure for influenza epidemics using virology data as gold standard. Weekly numbers of deaths were derived from Hong Kong Census and Statistics Department and weekly numbers of positive isolations of influenza A and B were obtained from the Queen Mary Hospital Microbiology Laboratory for the years 1996-1999. Influenza epidemic periods were defined when weekly numbers of positive isolations of influenza were greater than 4% of the annual total for two or more consecutive weeks in each year. Various cut offs in terms of percentile in the upper tail of weekly respiratory mortality counts were defined as a positive test for influenza epidemics. Sensitivity, specificity and areas under ROC curves were used in the evaluation. The area under the ROC curve is 90%. Influenza defined according to the cut off at >70^th, 80^th and 90^th percentiles are close to the left hand top corner of the ROC curve with sensitivity and specificity of 90% and 78%, 74% and 86%, and 55% and 96% respectively. Respiratory mortality data can be used for identification of influenza epidemics, and the method using percentiles of weekly counts is valid and could be used for adjustment of influenza epidemics in air pollution time series studies.