During influenza epidemics, school-aged children are amongst the first affected and then go on to spread the virus through the community [1-3]. Recognising influenza activity in schools could therefore be an important early indicator of seasonal activity in the wider community. During the 2005/06 season, influenza B was associated with high levels of morbidity in school children and over 600 outbreaks in schools were passively reported to the Health Protection Agency (HPA) by local Health Protection Units (HPUs) [4].

While there are considerable difficulties associated with directly monitoring influenza in schools, there is scope to investigate sentinel school absenteeism data as an indicator of influenza activity in the community. A similar study in New York [5] showed little benefit but analysis was based on overall absenteeism rates; absence data coded for illness, as in the United Kingdom (UK), could be a more sensitive tool for surveillance [6]. Following the large number of school outbreaks detected in the 2005/06 influenza season, the Department for Children, Schools and Families (DCSF) agreed to collaborate in a pilot study to test the feasibility of monitoring sickness absence in a sample of schools during the winter of 2006/07.

The main aims of this study were to test the feasibility of collecting school absenteeism data, to evaluate the usefulness of these data as an indicator of influenza activity and to improve the early detection of influenza outbreaks. The predominant circulating virus in the 2006/07 season (October 2006 to May 2007) was influenza A (H3) and indices of activity stayed well within normal limits. Compared to the 2005/06 season, in 2006/07 there were far fewer reported school outbreaks (n=20) to HPA Centre for Infections from HPUs [7], suggesting less influenza activity generally. If absenteeism due to illness could be shown to be an indicator of community activity in a low flu year, then not only would proof of concept of the pilot be achieved but the scheme would have applications beyond high activity, influenza B winters.

Methods

DCSF provided the HPA with a list of 90 schools that kept electronic registers and who could submit daily attendance data of pupils. The recruitment process was two-tiered: a direct approach to head teachers of the listed schools by the Centre for Infections and a request to local Health Protection Units to act as a first point of contact for recruited schools and to find other schools that kept electronic registers and were willing to participate in their area. The recruitment of schools based on DCSF listings was problematic, with the lines of communication from DCSF to HPA to school emerging as a key issue and, as a result, 17 were fully recruited but only 11 regularly provided data. Eight primary and three secondary schools from five of the nine HPA regions participated. Log-ins and passwords were provided to the schools in order for them to submit daily electronic register data for daily absenteeism due to illness, by age, via the Department of Health (DH) funded Health Protection Informatics (HPI) website. “Illness” is used by schools to account for absence when confirmed by parents and included respiratory and non-respiratory illness; absence for any other reason was not considered. Daily aggregated data were collected, episode incidence, i.e. the number of new absences and the duration of absence for an individual, were not collected. These data represent the total number of absentees for every school day in the study period: week ending 14/01/07 (week 02/07) to week ending 20/05/07 (week 20/07).

A number of schools were reluctant to join the surveillance scheme due to concerns of an extra workload, but with the development of clear user guides it was possible to show that this would involve little resource on the part of the school and could deliver high-impact results. The provision of support for online reporting was felt to be useful during the first few weeks. Although data collection should have started at the beginning of the scholastic year, i.e. the week ending 10/09/06 (week 36/06), due to logistical problems absenteeism data were only collected as of the week ending 14/01/07 (week 02/07).

The primary school population for this study comprised school year groups 1-6 (age group 4-11 years). Nursery and pre-school classes were excluded; they often comprised two groups (one group attending in the morning only and the other attending in the afternoon only). Similarly, school year groups 7-11 (age group 11-16 years) comprised the secondary school returns because the year groups 12 and 13 (age group 16-18 years) tend to be absent more than any other secondary school year groups, due to scheduled study leave. Electronic registering was split into two sessions per day and, accordingly, the HPI website was set up so that the number of sessions in a day would be twice the number of children that attend the school. However, this presumed that all children would be at school for five full days per week and would not accurately record the attendance of children who might only be present for five half days a week or those with scheduled study leave.

Weekly rates for overall absenteeism and illness-coded absenteeism were calculated and stratified in to primary and secondary schools. These data were then plotted against a number of indices of influenza activity; the Royal College of General Practitioners (RCGP) episode incidence rate (i.e. the weekly number of new consultations from sentinel general practitioners (GPs)) for influenza-like illness (ILI) per 100,000 population for all ages and the 5-14 years age group, comparable to the sample under consideration; the number of positive influenza samples taken by GPs involved in the RCGP sentinel scheme (the number of positive respiratory syncitial virus (RSV) samples were also considered to investigate the possible effect of confounding by this seasonal respiratory virus); the number of outbreaks in schools reported by HPUs; and the NHS Direct proportion of “fever” calls (5-14 years) and “cold/flu” calls (all ages). NHS Direct is a nurse led telephone helpline that can be used as a syndromic surveillance tool. Experience from several years of surveillance of NHS Direct calls has shown that rises in the proportion of “fever” calls in the 5-14 years age group and “cold/flu” calls in all ages may provide an early warning of a rise in influenza and influenza-like-illness in the community [8].

Results

Of the 11 schools involved in the study, at least nine reported data in any given week between week 02/07 and 11/07. Between weeks 12/07 and 20/07 returns were provided by between seven and nine schools. These figures include nil returns made by schools for weeks of school holiday closure. Week 20/07 is regarded as the end of the influenza season; data after this week were not analysed. When illness-defined absenteeism data was stratified into primary and secondary schools, illness absence in primary schools peaked one week before that in secondary schools during weeks 05/07 and 06/07 respectively (Figure 1). The peak illness absence in both school types was of a similar magnitude at 9.8% in primary and 9.2% in secondary schools. The half term break for all recruited secondary schools was in week 11/07 and the Easter break for all recruited schools (both primary and secondary) was in week 14/07 and 15/07. As a consequence, no absenteeism data were collected for the specified school groups in these weeks. Combined illness-defined absenteeism data peaked in the same week as the secondary school data (week 06/07) at 8.3% absence in week ending 10/07.

Combined illness-defined absenteeism data for primary and secondary schools was assessed against the episode incidence rate for influenza-like illness (all ages) obtained from the Weekly Returns Service of the RCGP, the positive influenza and RSV from this scheme and the proportion of NHS Direct “cold/flu” calls (all ages). The series was extended retrospectively to consider data from these current surveillance schemes from week 48/06. The peak RCGP rate and number of positive influenza samples was one week after peak week for illness-defined absenteeism in week 07/07. According to the RCGP thresholds, influenza was circulating in the community at this time. NHS Direct “cold/flu” calls did not reach the threshold level which was established to give advanced warning of influenza circulating in the community (Figure 2).

Please Note: The threshold for NHS Direct “cold/flu” calls (all ages) is 1.2%

The threshold for RCGP ILI episode incidence rate (all ages) is 30 per 100,000 population.

School illness-defined absence data peaked during week ending 11/02/07, the same time as the RCGP rate for the 5-14 years age group and NHS Direct calls for “fever” in the 5-14 years age group,(Figure 3). This data and the other indices of influenza activity for school aged children peaked one week later than outbreak reports submitted to CfI.

Please Note: The threshold for NHS Direct “fever” calls (5-14 yrs) is 9%
No threshold for RCGP ILI episode incidence rate has been set for specific age groups.

School illness-defined absence data was broken down by HPA region (data not shown). In all regions, illness absence peaked in the same week or one week prior to the peak RCGP regional episode incidence in the 5-14 yrs age group. In two regions, illness absence peaked one week after that of the NHS Direct regional rate for “fever” calls in the 5-14 yrs age group while in the remaining three regions it peaked in the same week or one week prior.

Discussion

It is very encouraging that, in a year of low influenza activity [7], a small number of schools demonstrated that illness-defined absenteeism could be correlated with established indices of influenza activity, similar in age structure to the sample under consideration. As with other indices that look at school age children, school illness-defined absenteeism data peaks before that of the general community, i.e. indices of influenza activity based on all age groups. Virology data would suggest that early increases in school absenteeism, RCGP episode incidence rate and proportion of NHS Direct “fever” calls for 5-14 years age group data reflect influenza activity. The results suggest that expanding this scheme to collect more rigorous evidence of how illness related school absenteeism could be used as a proxy for influenza activity would be worthwhile. With a larger cohort, the data would better represent national illness-coded school absence, would allow more extensive analysis and, with a few seasons’ data, establish baseline activity from which control charts could be developed. Such control charts could be used to alert HPUs of any larger than normal increase in absenteeism for that time of year, enabling the early implementation of control measures to be applied.

A more extensive analysis on a larger prospective cohort would be useful for examining whether peak illness-defined absenteeism in primary schools was significantly earlier compared to secondary schools alone and primary and secondary schools combined. Although illness-defined absenteeism peaked a week earlier than established indices, because data was not collected during the period of initial increase in influenza activity, one cannot assume that the initial increase would also have been identified earlier, but it would seem to warrant investigation as to whether this may be the case. If proven, then surveillance of primary school illness absence data alone could give an earlier indication of influenza activity than current established surveillance schemes. The requirement for ever more timely data is in part driven by the possibility of an influenza pandemic situation, where early detection of influenza in schools could be crucial in informing policy such as school closures and the move by local resilience fora to start their emergency-only mode of operation.

Given that fewer people with influenza-like symptoms now seek consultation with a GP than in the past, alternative systems for monitoring influenza, such as this, may become progressively more indicative of the disease burden experienced in these age groups. Different surveillance schemes for monitoring influenza activity will produce different estimates of influenza activity. A strength of school illness-defined absenteeism is that it can be used to differentiate between primary and secondary school populations unlike other surveillance schemes where there is stratification into age groups that transcend the schooling type. However, a weakness of this scheme compared to others is that it is unable to accurately gauge the true burden of disease, given the effect of other non–respiratory diseases also coded as “illness”. This could be addressed if respiratory disease was coded separately but there are currently no plans to do so. Considering data from several routine surveillance sources for a given period allows one to gain an estimation of influenza that is closer to true activity than any single system. This model may have applications in the surveillance of other seasonal diseases and could be used as a basis for similar surveillance schemes in European countries where timely illness-coded data was electronically available and was deemed to be culturally appropriate.

One of the key limitations to using school illness-defined absence data is that key periods of influenza activity are missed during holidays. Unlike respiratory virus infections with a more predictable season, such as RSV, influenza activity peaks at different times from one winter to the next. These missing weeks will therefore cause varying levels of difficulty when extrapolating influenza activity from illness-defined absence data. However, a larger cohort of schools would overcome this to some extent, since half term weeks vary across the country.

It was not possible to collect absenteeism data prior to week 02/06. Information on the background rate of illness-defined absenteeism throughout the scholastic year would have allowed inferences to be made on the significance (and specificity) of the peaks observed and on the sensitivity of this surveillance tool. The inclusion of pre-season months (i.e. October-December) would have allowed one to observe data for the start of influenza activity. This would have better allowed examination of any possible confounding effect that RSV circulation may have, although the declining number of positive RSV samples during the period of increasing and raised illness-defined absenteeism would suggest that this was negligible. In addition, RSV reports to the HPA mainly relate to illness in the <1 year age group and is therefore of questionable relevance to this study.

In order to improve detection of local outbreaks and possibly the start of influenza activity in the community, geographical representation is essential; currently lacking in the small number of schools recruited. While the pattern of absenteeism varies between regions, it is important to note that in two regions (Yorkshire and Humber and West Midlands) only one school participated in this pilot scheme. Therefore, little useful interpretation could be made from the comparison of school absenteeism data at HPA regional level against incidence data for the specific population of students under consideration (RCGP rate for the 5-14 years age group and NHS Direct “fever” calls for the 5-14 years age group) for a given region. With a larger cohort, breakdown of data to both region and HPU could be a useful tool in evaluating the effect of influenza on absenteeism at these levels.

With appropriate geographical representation, these data could be used to provide a real time public health response by informing Health Protection Units (HPUs) when and where to investigate. In the seasonal situation, early detection of influenza in a local community would trigger investigation and virological sampling by local HPUs which would allow us to analyse the evolution of the virus, to identify important drift variants, and to contribute information for vaccine recommendations and vaccine candidate viruses. It is likely to lead to prophylaxis or treatment being offered where appropriate, reducing morbidity and spread of infection. Ideally, virological investigation should be part of such a scheme, with samples taken from a proportion of participants when activity increases. Not only does it provide key virological information as previously described but it underpins the epidemiological information collected. However, due to finite resources such sampling was not undertaken and potential outbreaks in schools, as indicated by the absenteeism data, were not investigated.

School outbreaks reported to CfI from HPUs peaked before all other indicators of activity in the school age group. However, it is worth noting that due to the passive nature of the reporting of these outbreaks, they are unlikely to be representative and given that they are not consistently laboratory confirmed would be no replacement for routine sampling of a proportion of schools involved in this scheme in the future.

It is clear that good communication with the schools is essential if recruitment and compliance are to be maximised. Established relationships between the schools and local health protection units are also crucial; recruiting and supporting the schools centrally was more labour intensive than anticipated. The pilot also identified some required changes in relation to the dataset and recording and retrieving of information through the website, which are being addressed jointly by the HPA and the DH web team. With more obvious returns for the schools, in terms of local responses by HPUs, it is likely that more would sustain regular reporting even during weeks outside of the winter, vital in generating a dataset from which control charts could be developed.

During the 2007/08 season, the HPA will continue to work closely with both the DCSF and HPUs to identify schools who could participate in this surveillance scheme. Recruitment will be increased with a focus on primary schools and improving geographical representation throughout England. Securing access to retrospective illness absenteeism data from a large cohort would provide sufficient power to carry out a quantitative analysis of the relationship between illness absenteeism and influenza activity, as denoted by the indices described in this paper, and further investigation in to the usefulness of a scheme such as this.

Acknowledgements

We thank the schools who participated in this pilot, DCSF in full, and the regional epidemiologists and Health Protection Units from Local and Regional Services of the HPA who supported these schools. Data used in this paper from established influenza surveillance schemes were extracted from the agreed dataset shared with the HPA.