According to estimates made in 2003 by Statistics Netherlands, 6 out of 10 inhabitants of the Netherlands older than 12 years were regular internet users. Internet usage was high (75%–85%) between the age of 12–45 years, steadily decreasing thereafter by 10–15% per decade. About 5–8% of elderly persons over 75 years still used internet regularly. Internet usage by men was slightly higher than by women in 2003 and the higher the level of education, the higher the use [10]. Where the accrual of persons willing to participate in GIS was concerned, the enterprise was definitely a success. Following substantial recruitment efforts in the media, almost 26,000 persons replied to the requests for participation by answering the intake questions at least, and 53% of these became serious participants. This percentage compares favorably with the 51.2% of serious participants who finished a Swedish internet survey on sexual behavior following an open invitation via a website, and the 50%-plus response rate obtained in a population web survey on lifestyle issues reported by Augustsson et al. [16, 17]. Much lower rates have been reported for online health surveys, however, possibly because they involved less appealing subjects (than sex or influenza), had a poorer design, or were less creative in their motivation and incentive strategy [18, 19].
The serious participants in our study were not equally distributed with regard to gender and age. The Dutch population has an almost equal male-female distribution up to the 71–80 years age group. In the GIS population, however, women between 11 and 50 years were clearly overrepresented, whereas the reverse was observed after the age of 50 and men became overrepresented. Apart from gender, there was, by and large, a clear overrepresentation of GIS participants between 21 and 70 years and a marked under-representation before and after these age limits. The Swedish internet survey mentioned above found an almost identical overrepresentation of participants between 25 and 65 years and under-representation beyond these age limits [17]. A disproportionate distribution of this kind is not a great problem for many health-related issues, however, because there is a good chance that most relevant age groups are included. On the other hand, this imbalance may be a problem for the surveillance of topics such as ILI, because the very young, and the very old to a lesser extent, are largely underrepresented, while these are precisely the age groups that suffer most from ILI. We found, nevertheless, that this concern did not translate into a difference between the course of the ILI incidence curves from the Dutch Sentinel Practice Network and GIS. The incidence of ILI in the very young and old appears to have been less than had been expected, or was outnumbered by the magnitude of the ILI incidence in the rest of the GIS population. This was supported by the fact that the similarities between the 2 curves remained the same when very young and very old patients were excluded from the data provided by the GPs (manuscript in preparation). The major difference between the two curves is that the GIS curve started from a much higher background level and remained at a higher level throughout the season.
The most likely reasons for this difference are that only some patients with ILI complaints will go to their GPs, and that GPs are more critical in making the diagnosis. Part of the difference in amplitude between the GIS curve and the GP curve may also have been due to overrepresentation of GIS participants between 25 and 65 years of age. They represent the majority of the working population, and may be less likely to consult their GP for ILI complaints than the non-working population. On the other hand, they also represent the more healthy proportion of the population. It has been estimated earlier that in the Netherlands the incidence of ILI in the community at least is 6 times higher than presented to the GP [20], which corresponds with the present observation that the overall difference between GP consultation rate and GIS incidence is 10-fold.
It was a remarkable and encouraging observation that the course of ILI monitored by GIS so closely resembled the official survey of ILI by GPs participating in EISS. The implications of this finding for the surveillance of ILI itself are limited, however. The monitoring of ILI and influenza by EISS is well established and has not only proved to be reliable over the years, but also to function as an important source of scientific information with regard to virology and vaccination [15]. The key role of GPs in the EISS surveillance system ensures a high level of continuity and scrutiny, characteristics that are questionable in GIS. As far as continuity is concerned, GIS is being carried out for the third time in the Netherlands and Flanders in the 2005–2006 season and was started in Portugal for the first time [21]. It remains to be seen how successful these surveys will continue to be in the long run.
A comparable population-based approach for ILI surveillance, in which the GP is bypassed and information is provided by callers to a helpline, (NHS Direct), has been used in the UK. NHS Direct is a nurse-led telephone helpline providing health advice 24 hours a day. On receipt of a call and description of the caller's symptoms, nurses use their judgment to select the appropriate diagnostic algorithm in their computer, leading to self-care advice or a referral to another part of the NHS [22]. In principle, the call data, which can be analyzed daily or even by the hour, can also be used for community surveillance, including ILI. The usefulness of NHS Direct for ILI surveillance has been evaluated for two influenza seasons thus far and, although the results have been reassuring, the system was hindered by insufficient population coverage, lack of uniformity and poor definition of ILI [23, 24].
There was some indication that the incidence of ILI started to increase 1–2 weeks earlier than reported by EISS, but this could not really be substantiated, due to the small number of early events and the high background level of ILI in GIS. For a better understanding of the natural course of ILI it might have been better if GIS had started much earlier than the start of the influenza season. It might have provided us with a badly needed threshold value.
It should be noted that in GIS it took about 4 days before reliable incidence rates could be deduced from real-time data, giving the internet-based approach a head start of about 5 days on the information provided by GPs. This difference between mail and web is likely to disappear as soon as Dutch GPs start reporting their data on a real-time basis. The success of an approach of this kind has been reported for GPs participating in a real-time influenza surveillance projects in other countries [25, 26].
The major finding of the current study is that it is possible to recruit of a high number of persons, willing to participate in on-line health surveillance. This has been previously demonstrated for well-defined groups such as students [27], but only occasionally for the general population [16]. Our study demonstrates that not only was recruitment successful, but also that the information provided was reliable, because it paralleled the information that was gathered via a different, acknowledged route. In addition, the demographic and health characteristics of the participants were remarkably similar to the general Dutch population, apart from some typical deviations related to internet use. It can therefore be envisioned that this type of interactive on-line surveillance is extended to other health-related issues with, preferentially, a high public appeal such as obesity, life style and stress. Actually, questions about stress and stressful behavior have been asked to participants of the third GIS that took place during the 2005–2006 influenza season [8].
Bälter et al. recently made some wise remarks on the appropriate use of internet as a tool for epidemiological research [28], but their remarks were mainly devoted to technical subjects such as length and layout of the questionnaires. The GIS study allows us to tentatively point to another cornerstone that may be important for success: feedback of information to keep the participants involved and motivated.
