Visual inspection suggests that worker and school absenteeism data might be used as an early indicator of influenza epidemics in Belgium. For the two worker absenteeism data sources investigated, a rise in absenteeism rates preceded the onset of the influenza epidemic by 2 to 3 weeks. The usefulness of the school absenteeism data in predicting the onset of an epidemic with the available data is harder to explore, because the recording started only a few weeks before or even after the onset of the influenza epidemic, which was partially due to the summer holidays. Nonetheless, it is promising to see that the peaks of the primary and secondary school absenteeism rates preceded the ILI peak, suggesting that school absenteeism data can also be used to predict the onset of an influenza epidemic.
Since the preliminary results are promising, it seems worthwhile to collect absenteeism data on longer time periods covering several influenza epidemics. Once such data are available, early warning algorithms can be developed taking into account e.g. seasonality, day-of-week effect, holiday effect and spatial variability. Evidently, the accuracy (i.e. the false positive and false negative rate) of such an early warning system needs to be scrutinised before implementation. In particular, high false positive rates are to be expected because several causes of increased absenteeism rates exist, especially for the younger age groups. However, by developing an early warning system that uses different data sources simultaneously, the accuracy is expected to improve sub-stantially. Nevertheless, developing early warning systems based on absenteeism data has several advantages as well. First, because absenteeism data (like other 'syndromic' data sources) are not disease-specific, they can be used to detect a variety of circulating diseases other than influenza. Second, absenteeism data (like other 'syndromic' data sources) are prediagnostic, implying a gain in time compared to diagnostic data sources (e.g. laboratory-confirmed influenza cases). Third, and in contrast to e.g. 'syndromic' surveillance using hospitalisa-tion or mortality data, absenteeism surveillance can potentially detect outbreaks of diseases causing minor illnesses. A fourth advantage of (school) absenteeism surveillance is that it allows monitoring age-specific sub-populations. Indeed, by monitoring nursery school absenteeism rates, a strong increase in RSV-circulation was detected, which would likely have remained unnoticed when using non-age-specific data sources. Finally, be-cause absenteeism surveillance (like other 'syndromic' data sources) makes use of existing data, it is relatively economically beneficial and easy to collect the data. In particular, for the worker absenteeism surveillance, existing data files are sent to and analysed by the WIV-ISP. For the school absenteeism surveillance, schools summarise the collected data, report the summary data electronically in a database hosted by the corresponding authorities. Thereafter the data, aggregated over schools, are sent for analysis and reporting to the WIV-ISP.
To conclude, we would like to mention the usefulness of absenteeism data for purposes other than early warning of new epidemics. Indeed, the absenteeism surveillance system is of particular importance in providing essential information to policy makers during crisis management. For instance, absenteeism data can serve as a basis to decide whether to close down schools (locally or nationwide) as an epidemic control measure. In addition, absenteeism data are important sources to adequately assess the socio-economic burden of influenza epidemics or other disease outbreaks. However, to this end, methodological improvements (e.g. standardisation of the school/company selection criteria, standardisation of the criteria used to register workers/students as present/absent) are needed.