Despite this basic congruence, considerable differences in the output data remain (3,5,9). This can be explained by differences in the healthcare systems and other influences on consultation behaviour, for example, reimbursement issues for medication and consultations, and organisational needs (such as certificates needed for absenteeism).
Additionally, different emphasis of the main effort and presuppositions taken into account for the set up of each network led to the use of different case definitions. For the networks included in the study, the case definitions can be summarised in three groups:
- Influenza-like illnesses (ILI) with case definition*
- ILI without a case definition (illnesses that are considered to be influenza)
- acute respiratory tract infections (ARI) with case definition*
*The case definitions used are not uniform.

Different denominators are in use. A population based denominator is ideal, but cannot always be directly calculated, for instance, in healthcare systems where patients have a free choice of general practitioner (GP) to be consulted. Therefore, the number of reporting practices or the total number of consultations are used as denominators instead (10). The estimation of population based rates has been encouraged for networks where population based reporting is not possible. This was one harmonisation measure used by EISS after the index study had been started (3,5). Nevertheless, the numerical observations of the various national networks differ considerably, and the data and resulting graphs can be interpreted only with an in depth knowledge of each specific network (5,6).This study explores the usefulness of an index based on simple principles for harmonising the scaling of national data and assisting the interpretation and estimation of influenza activity.

Material and methods
For this pilot study data from eight countries which were able to provide the necessary historical data were chosen. The following morbidity indicators are recorded in these countries:
• ILI (without case definition) per population (England)
• ILI (with case definition) per population (Netherlands and Portugal)
• ILI (with case definition) per consultations (Belgium and Switzerland)
• ARI (with case definition) per population (Czech Republic)
• ARI (with case definition) per consultations (Germany and France).

The concept of the index was to assess the weekly influenza-attributable excess morbidity for each country, which is a reliable and commonly used indicator for influenza activity. This was put in relation to the excess morbidity expected for the peak weeks of 'usual' seasons in the respective networks.
Calculation of the background activity for each country (6,11,12).
As a simple and practical method, the mean value of recent years (minimum seven years) for each specific week of the period was calculated, excluding all weeks where influenza activity was considered to be more than sporadic (influenza-attributable excess morbidity is detectable) by those responsible for interpretation of the data (13).

In cases where data from under two weeks could be used for the calculation of a specific week - for example, due to frequent influenza activity during this week in recent years - adjacent (previous and following) week values were included for the estimation. A sliding mean value over three weeks was practical for smoothing periods where influenza circulated frequently. In periods with sufficient weekly values, smoothing was avoided because the typical weekly pattern (imprinted by public holidays, Christmas, etc.) might be softened.

Calculating the excess morbidity typical for the peak weeks of a 'usual season'.
This 'reference' value represents the excess during the peak period of a usual activity and was calculated from at least two epidemics considered as usual by the national scientists responsible. The mean of the influenza attributable excess (recorded value minus background) during the three peak weeks of all selected seasons represents the increase of the indicator that can be expected when the level of influenza activity is no higher than usual (13).
The seasons and weeks selected to assess this reference value for each country are listed in table 1.

Calculation of the index per week:
The index for each week is calculated as the excess morbidity (observed value for the week minus 'background' expected for the week) in relation to the reference value representing the increase indicating usual influenza activity. If the actual measured value is equal to, or lower than, the 'background', there is no more than background morbidity (no influenza attributable excess), and the index is set at zero. Consider, for example, the value of 50 ILI/100 000 persons in country X for week 45. Assuming the background for week 45 has been determined for country X to be 40 ILI/100 000 persons. Thus, the excess is 10 ILI/100 000 persons. If the reference, ie the average excess of the weeks with highest activity during usual influenza seasons, was 100 ILI/100 000 persons, then the index had a value of 10/100=10%.

Results
In Figure 1 (graphs 1 to 8) the morbidity indicators for the 2001/02 season are shown as recorded by the different systems. The number of positive influenza assays is also given, to indicate the period with increased circulation of the viruses. In most of the graphs, the beginning and end of the influenza epidemic can be recognised. In some network this is only possible with the additional consideration of the laboratory data. An assessment of the magnitude of influenza-attributable excess morbidity (as indicator for the influenza activity) and its development during the season is impossible without additional information.

In Figure 2 (graphs 9 to 16) the index for the season 2001/02 is shown for the respective countries along with the number of positive Influenza assays. The flow of the curves is generally unchanged, as only the estimated background morbidity is eliminated. This is relevant for ARI recording systems in particular, where the background is usually large (6). The uniform scaling leads to a stretching or compression of the curves forcing the indicator level typical for a usual influenza peak activity to an index value around 100. The beginning, peak, and end of the influenza epidemics can now be recognised more easily. The indicator aids a definite impression of the magnitude of the excess morbidity.

The categories currently used by EISS to describe the level of influenza activity derive from assessments for the entire season and discriminate (5):
• Low: no influenza activity or influenza activity is at baseline level
• Medium: level of influenza activity usually seen when influenza virus is circulating in the country, based on historical data
• High: higher than usual influenza activity compared to historical data
Very high: influenza activity is particularly severe compared to historical data
This does not provide categories for the range 'no activity' to 'medium', which is the range for most of the season and the phase of increasing and decreasing activity. The index shown allows further discriminations particularly in this range and additional categories are suggested:
• Low activity: the excess morbidity is insignificant or low
(0-40%).
• Moderate activity: the excess morbidity is significant but still clearly below a usual peak level (41-80%).
• Usual (or medium) activity: level of influenza activity usually seen in the peak period of epidemics considered as usual based on historical data (81-120%).

Discussion
Harmonisation is a key issue in international collaboration with regard to surveillance systems. Minimum requirements regarding representativity and common surveillance principles are important aspects of harmonisation. Morbidity data based on at least 0.5 to 1% of the population under surveillance are generally considered to be sufficient for influenza that causes symptomatic illness in approximately 5% to 10% of the population during average seasons (15-18). The networks collaborating in EISS fulfil these requirements.
The study investigates whether an index projecting the indicators used in the different networks to a uniform and relative scaling is in principal useful for harmonisation. For this harmonisation measure the individual turn of the scale for each system that indicates 'usual' peak activity has been estimated for each system. Usual activity can be understood as representative for influenza seasons disregarding the 'non influenza seasons' and seasons with unusually high activity (13).

The methods used are simple, practical and easy to apply. The use of these seemingly crude methods, together with the arbitrary momentum from the assessment of national experts, appears tolerable considering the reliability of the recorded data which are affected by unclear selection steps during registration due to consultation behaviour, interpretation of the criteria given by individual physicians, etc. A fine tuning of the method has not yet been performed, and the differences, and advantages and disadvantages of alternatives to estimate the background, reference value, etc. have not been assessed. (for example, for the estimation of the reference the average of a number of peak weeks from all seasons would be a practical alternative if an appropriate number of seasons is available.)
The index shown assists the interpretation of the national data without the need to present historical data or detailed knowledge of each national network and healthcare system. The assessment of the influenza activity is supported and the better harmonisation may allow a finer discrimination of the intensity levels for each week. Additional categories for influenza activity to allow for discrimination are suggested, particularly in the range when the usual activity has not been exceeded. This is the case in many seasons and during the phase of increasing and decreasing activity. This could enable establishment of a more widely accepted terminology.

Compared to the activity levels currently used by EISS the "usual activity" indicated by the index is somewhat lower than the 'medium' activity due to the exclusion of seasons with high activity for the calculation of the reference value. For the consideration of the seasonal peak phase it should be considered that the reference value of the index takes three peak values into account, and hence the mean of the three peak values of the current season is decisive for influenza activity.
Despite the uniform scaling of the different national data, direct comparability of the index is limited. Basic differences between networks are not compensated or affected by the indexing eg the different sensitivity to certain syndromes, due to the consultation behaviour in the country and the case definition used. If, for instance, a wider definition such as acute respiratory infection is used (and people frequently consult with mild symptoms) this system is probably much more sensitive to influenza A/H1N1. This subtype is currently mainly infecting the younger age groups and causes mild symptoms in the majority of the cases. For a system using a strict definition with high fever, cough, joint pain etc. in a healthcare system where people consult only when they are severely ill, a relatively lower sensitivity against Influenza A/H1N1 illness can be expected (9). Disadvantages of the indicators used at present due for example to aggregation levels - with regards to regions and their geographic dimension or age groups - are not compensated by the index. For instance interferences of other co-circulating agents (eg respiratory syncytial virus (RSV)) with the morbidity indicator persist, because the index only projects the indicator to a uniform and relative scaling. The key problem is the unknown link - or function - between the true incidence in the population and the indicators registered for surveillance, which probably differs for the various networks. The index gives weight to the impression based on the indicators currently used without compensating differences in that function. This leads to a declining reliability of the index particularly when the influenza activity is unusually high. This may be further explored taking into account historical experiences of severe epidemics. Thus the index may be helpful in making those basic differences more apparent and so encourage their further investigation.

The link between the true incidence of the disease in the population and the registered indicator is mainly dependent on the selection steps for the registration of the cases. Differences in case definitions and denominators used are perhaps the most obvious factors, and their harmonisation is an important option. It should, however, be considered that a shift of these criteria might interrupt the continuity of the national observation and hence comparability to historical data. An adjustment phase of unpredictable duration might follow requiring additionally costly quality control measures. On the other hand many differences are a consequence of the healthcare systems, consultation behaviour, cultural beliefs, etc. Those influences will remain, and a satisfactory comparability may not be achieved. This is intimated by countries using very similar indicators, such as England and the Netherlands, or Portugal with a still very limited comparability of the recorded values. For example, the average of the three peak weeks of all epidemics from 1992/93 to 1996/97 in England was 163 ILI per 100 000 while in the Netherlands it is 291 ILI per 100 000 (10). A generally higher influenza activity in the Netherlands is not plausible. Applying the index calculation to these rates, the result for England would be 137% of the reference value and 138% for the Netherlands. Considerable regional differences in some countries exist despite the use of uniform numerators and denominators and a uniform healthcare system (8,14). Such measures should be carefully explored in pilot studies and are not considered to be short term options. Indexing the data may be a useful short term alternative and may allow a gradual harmonisation.