Objectives We examined whether the weekly number of newspaper articles reporting on influenza was related to the incidence of influenza in a large city.

Design Prospective, non-randomised, observational study.

Setting Registry data of influenza cases in Fukuoka City, Japan.

Participants A total of 83 613 cases of influenza cases that occurred between October 1999 and March 2007 in Fukuoka City, Japan.

Main outcome measure A linear model with autoregressive time series errors was fitted to time series data on the incidence of influenza and the accumulated number of influenza-related newspaper articles with different time lags in Fukuoka City, Japan. In order to obtain further evidence that the number of newspaper articles a week with specific time lags is related to the incidence of influenza, Granger causality was also tested.

Results Of the 16 models including ‘number of newspaper articles’ with different time lags between 2 and 17 weeks (x_{t-2 to t-17}), the β coefficients of ‘number of newspaper articles’ at time lags between t-5 and t-13 were significant. However, the β coefficients of ‘number of newspaper articles’ that are significant with respect to the Granger causality tests (p<0.05) were the weekly number of newspaper articles at time lags between t-6 and t-10 (time shift of 10 weeks, β=−0.301, p<0.01; time shift of 9 weeks, β=−0.200, p<0.01; time shift of 8 weeks, β=−0.156, p<0.01; time shift of 7 weeks, β=−0.122, p<0.05; time shift of 6 weeks, β=−0.113, p<0.05).

Conclusions We found that the number of newspaper articles reporting on influenza in a week was related to the incidence of influenza 6–10 weeks after media coverage in a large city in Japan.