This study confirms that estimates of absenteeism due to seasonal influenza typically ranged from 5% to 20%; higher absenteeism rates were associated with mixed seasons. These results are in reasonable agreement with general assumptions on the clinical attack rate for influenza, though it is noted that not everyone with symptoms consistent with an influenza like illness (ILI) [15, 24] (sore throat, fever and cough are the most frequent symptoms) would necessarily take time off work. Symptoms can be mild for some; not everyone experiences a fever and many infections are believed to be asymptomatic [25, 26]. Absenteeism rates for the 2009 pandemic were similar to those occurring for seasonal influenza. Employees, however, took more time off due to the pandemic strain than was typical for seasonal influenza. Employment characteristics had less impact on hours lost due to influenza than on the total hours lost for all illness and disability (Figure 2) .
Four special questions were added to the labour force survey in December 2009 through to February 2010 in order to estimate the impact of influenza on hours worked. Labour force survey participants were asked how many hours they took off work as a result of the 'flu' in the previous month due to their own illness as well as for the care of others. An estimated 9.0%, 4.4% and 3.5% of employed people were absent from work as a result of the 'flu' for November 2009, December 2009, and January 2010 respectively [27, 28]. Workers also reported working additional hours due to the flu. In comparison, in this study we estimated an absenteeism rate due to influenza of 6.7% and 0.3% for the months of November and December 2009 respectively, and a negligible amount for January 2010 (Table 1). The two estimates, though both based on LFS participants, are different. Because of sample rotation, only about 5 out of 6 households surveyed in the November panel participated in the December survey. The November panel was asked about absences during the November reference week, while the December panel was asked about absences during the December reference week and about flu related absences during the whole month of November. Estimates of influenza-attributed absenteeism for the non-reference weeks were calculated based on the weekly level of influenza activity. The survey estimate from the four special questions included hours lost due to the respondent's own flu-related illness, care for others, and any flu-related medical appointments. However, the assessment that influenza-like symptoms were due to flu was at the respondents' discretion. There are many viruses that cause influenza like symptoms, and most 'flu' symptoms in December and January were most likely due to other viruses. After accounting for the differences in definitions, the two estimates of absenteeism for the month of November appear to be consistent. Our estimate of the average number of hours lost per absence was slightly higher than the estimate of hours lost due to 'flu' from the special survey (25 hours compared to 20 hours for 'flu'), and again this difference was possibly due to the potential inclusion of other ILI by the respondents in the special survey.
Economic studies of the benefits of influenza vaccination programs in the workplace avoid the costly process of directly measuring absenteeism due to influenza by comparing the number of days lost due to ILI in vaccinated and unvaccinated workers . Confirmation of an influenza infection is possible through laboratory testing; however, this limits potential studies to a small population. The advantage of our approach is that the absenteeism estimates are specific to influenza, can be generalized to the Canadian labour force and include many seasons; however, the indirect estimation of hours lost due to influenza has other limitations, including the relatively large confidence intervals for sub-populations. This statistical approach has been used to estimate other characteristics of the disease burden attributable to influenza on the Canadian population, such as hospitalization [1, 17] and mortality rates [6, 8, 16]. In comparison, the LFS is a relatively small sample of employed persons in Canada, with a data point available for only one week per month. As a result, this study did not have sufficient statistical power to assess the relative effects of the various employment characteristics on absenteeism rates, though the studies mentioned above were able to provide estimates for sub-populations with more precision. Proxy variables for the level of activity of other respiratory viruses such as parainfluenza and respiratory syncytial virus (RSV) were initially included in the model, but were dropped due to lack of statistical significance and because, based on these previous modelling experiences, it is reasonable to assume that hours lost due to other, non-influenza ILI would be captured in the seasonal baseline.
The inclusion of a scale parameter in the model inflated the confidence intervals of the estimated parameters, so it is unlikely that the level of statistical significance is overstated, however, the less than ideal model fit still suggests caution in the interpretation of model results. As this is a population-level study design, other explanations than those included in the model may be possible. The effect of public health messaging advising the public to stay home if sick is uncertain, as the proportion of hours lost was actually lower in the summer months of 2009 than for previous years. Would employees have taken less time off for other reasons in anticipation of possibly needing additional sick days due to a future infection with the pandemic strain? Absenteeism rates were not statistically significant for all seasons; it is not clear whether the lower peak absenteeism rates for the 2005/06 season were due to limited illnesses related to influenza that season as the model suggests, or due to other causes not included in the model. The robustness of annual estimates of disease burden is known to be less than ideal.
Despite higher vaccination coverage in recent years (increasing from approximately 10 to 25% of the working age Canadian population ), a slight upward trend in absenteeism rates due to seasonal influenza was noted for recent years (not shown). As the four seasons where a single antigenic strain dominated occurred early in the study period (1997/98, 1999/00, 2002/03 and 2003/04) and these seasons were associated with relatively low absenteeism (Figure 1), the apparent lack of association between vaccination coverage and absenteeism could to be explained by higher overall attack rates associated with the co-circulation of multiple influenza strains in recent years or, perhaps, increasing social pressures for self-isolation at home when sick.