General
Data from a median of 17 (14–28) countries from the EISS database were analysed for eight influenza seasons: 1999–2000 (14 countries); 2000–2001 (15 countries); 2001–2002 (16 countries); 2002–2003 (17 countries); 2003–2004 (22 countries); 2004–2005 (22 countries); 2005–2006 (22 countries) and 2006–2007 (28 countries). Countries were included in the analysis if they were at least 5 years active member of EISS and if weekly data were available for the full season. The assessment of influenza activity presented in this paper is largely based on data reported by sentinel GPs. The GPs report clinical cases of influenza-like illness (ILI) and/or acute respiratory infection (ARI) to a central registry and take respiratory specimens that are sent to a national reference laboratory for testing. This ensures that the clinical data reported by the sentinel physicians are validated by virological data on influenza. The national reference laboratories also report laboratory test results on non-sentinel respiratory specimens e.g. specimens from hospitals or non-sentinel physicians. These data were collected to have an additional indicator of influenza activity and to validate the sentinel virological data. The national reference laboratories participate in the 'Community Network of Reference Laboratories for Human Influenza in Europe' (CNRL), which is coordinated by EISS [9]. CNRL works closely with the WHO through its network of National Influenza Centres and collaborates with the Centre for Reference and Research on Influenza at Mill Hill, London, UK.
In the current study only the time points of highest clinical and virological activity were used, because only peak levels represent undisputable markers of activity in a given country. The reason not to take whole incidence curves into account is that incidence rates in Europe vary considerably, because: 1) case definitions are not yet harmonised across Europe; most countries report cases of ILI, but some report cases of ARI, 2) the denominator calculations vary by country and, 3) consultation rates for ILI and ARI vary among countries. They not only depend on cultural factors, but also on the delivery of health care. For example, in some European countries a doctor's certificate is required for a single day of absence from work (leading to a higher consultation rate), whilst in others a certificate is only required after absence of 5 days or more, leading to a lower consultation rate.
The weeks of peak activity were selected by plotting the clinical and virological data available for each country. If the clinical and virological activity was very low during a season (e.g. below or around the baseline level; defined as level of influenza activity in the period when no influenza virus was detected), it was difficult to identify the peak week and no peak was selected. Most countries reported cases of ILI to EISS (13 out of the 17 countries); four countries used the less restrictive case definition of ARI (Czech Republic, France, Germany and Romania). Since 2004 the Czech Republic and Romania also report cases of ILI [10]. The case definitions of ILI and ARI have been described by EISS and discussed by Aguilera et al. [11, 12]. Briefly; the general criteria for ILI are: sudden onset of fever > 38°C, with respiratory (i.e. cough, sore throat) and systemic symptoms (headache, muscular pain); the criteria for ARI are: sudden onset of respiratory symptoms, accompanied by fever and headache in the absence of other diagnosis.
Validity analysis
For the validity analysis of sentinel reports, we defined a good match as a situation where the sentinel and virological peaks occurred in the same week, or when there was a difference of only one week. For example the peak of the incidence of ILI consultations in the Netherlands during the 2002–2003 season was week 10 and the peak of positive laboratory reports of the dominant influenza virus was week 9. This difference of 1 week was considered to be a good match. Therefore, a time difference between peaks of 1 week or less was taken as a measure for a good match (irrespective of which peak presented first); a difference of 2 weeks was taken as a reasonable match and a longer period as a poor match. The analysis was based on the percentage of countries fulfilling the criteria of a good or moderate match during 8 influenza seasons. In a second validity analysis, using similar criteria, the sentinel clinical incidences were compared with non-sentinel laboratory reports. Because the case definitions of ILI and ARI differ considerably, the validity analyses were performed separately for countries using ILI (n = 13) and countries using ARI criteria (n = 4). Differences in matching percentages between ILI and ARI were statistically evaluated using the Chi-square test or Fisher exact test if the expected value in one of the cells was less than five.
Length of influenza season
The length of an influenza season was roughly calculated by subtracting the earliest and latest week of peak clinical activity across Europe for each season. Per season the aggregated data of participating countries were used. Knowing that using peak weeks as indicator for activity would lead to underestimation of the length of the epidemic, because periods of high activity at the beginning and the end were not taken into account, 4 weeks were added: 2 weeks before the earliest peak and 2 weeks after the last peak. These periods still represent a rather conservative estimate of the slopes of increased activity around the incidence peaks. Eight countries participated throughout the 8 seasons (their longitudes ranging from -4 to 15.3); 6 countries during 7 seasons (longitudes -8 to 19.3) and 14 countries were included during less than 7 seasons (longitudes -3 to 25).
Spread of influenza (course of peak activity across Europe)
The sequence of peak activity of influenza in the various European countries was taken as a measure for the spread of influenza across Europe. We are well aware that the use of the word 'spread' is based on the liberal assumption that the sequence of peak activity across Europe parallels the actual spread of influenza. Therefore, as we have no clear insight into the dynamics of influenza between countries, in the present study 'spread' should be appreciated with some caution and in a very general context. In order to assess a possible west-east spread or a south-north spread of influenza activity in Europe, the peak week data of influenza activity in EISS countries were plotted against the longitude and latitude of the central point in each country. For the purpose of finding the appropriate geographic centre of a country, rounded longitude and latitude figures were used, based on The Gazetteer of Conventional Names, third Edition, August 1988 [13]. For Northern Ireland, Scotland, Wales and England such central points were not available. Therefore we took the capital cities of these regions: Belfast, Edinburgh, Cardiff and London as best substitutes. Considering it was difficult to identify a peak during seasons of low influenza activity, the sentinel virological data, if available, were used to select the peak. If no sentinel virological data were available, no peak was selected.
Regression analysis and analysis of significance was performed using SPSS 11.5. The variance was expressed as squared correlation coefficients (R2), interpreted as follows: < 0.1 very weak correlation; 0.1–0.25 weak; 0.25–0.50 moderate; 0.5–0.75 strong; 0.75–0.9 very strong; > 0.9 exceptionally strong correlation.