Effects of seasonal and pandemic influenza on health‐related quality of life, work and school absence in England: Results from the Flu Watch cohort study

Background Estimates of health‐related quality of life (HRQoL) and work/school absences for influenza are typically based on medically attended cases or those meeting influenza‐like‐illness (ILI) case definitions and thus biased towards severe disease. Although community influenza cases are more common, estimates of their effects on HRQoL and absences are limited. Objectives To measure quality‐adjusted life days and years (QALDs and QALYs) lost and work/school absences among community cases of acute respiratory infections (ARI), ILI and influenza A and B and to estimate community burden of QALY loss and absences from influenza. Patients/methods Flu Watch was a community cohort in England from 2006 to 2011. Participants were followed up weekly. During respiratory illness, they prospectively recorded daily symptoms, work/school absences and EQ‐5D‐3L data and submitted nasal swabs for RT‐PCR influenza testing. Results Average QALD lost was 0.26, 0.93, 1.61 and 1.84 for ARI, ILI, H1N1pdm09 and influenza B cases, respectively. 40% of influenza A cases and 24% of influenza B cases took time off work/school with an average duration of 3.6 and 2.4 days, respectively. In England, community influenza cases lost 24 300 QALYs in 2010/11 and had an estimated 2.9 million absences per season based on data from 2006/07 to 2009/10. Conclusions Our QALDs and QALYs lost and work and school absence estimates are lower than previous estimates because we focus on community cases, most of which are mild, may not meet ILI definitions and do not result in healthcare consultations. Nevertheless, they contribute a substantial loss of HRQoL on a population level.


| INTRODUCTION
Influenza epidemics have a major social and economic impact. As well as direct healthcare costs, influenza may lead to other indirect effects including school absenteeism, loss of workplace productivity and effects on health-related quality of life (HRQoL). 1 The quality of life of both patients and their families may be affected, especially when the patient is a child. 2 Quantifying indirect effects accurately is essential to inform cost utility analyses (CUA) of interventions to mitigate the population impact of influenza, including extension of seasonal vaccination policies.
In the United Kingdom, the National Institute for Health and Care Excellence (NICE) recommends that health effects of interventions are expressed in terms of quality-adjusted life years (QALYs) as this generic measure of health benefits reflect both mortality and HRQoL. 3 The standardised validated tool EQ-5D 4 is NICE's preferred measure of HRQoL. 3 NICE uses a cost utility threshold of £20 000-30 000 per QALY to judge whether or not interventions are deemed cost effective.
A systematic review of HRQoL in influenza showed a paucity of studies that used standardised well-validated methods to generate estimates of the quality-adjusted life days (QALDs) lost. 5 It identified 4 previous estimates of QALDs lost due to influenza, which varied from 1.57 to 10.69 depending on the population sampled and method of HRQoL measurement used. [6][7][8][9] Many of these studies did not measure HRQoL throughout the duration of illness. They tended to measure HRQoL once at baseline and once on the worst day of illness, which required assumptions to be made about the shape of the QALY loss over an illness. 5 Studies that measure HRQoL and work and school absence from influenza cases seeking medical attention may overestimate the indirect cost per case. A systematic review of studies of children's absences from school and day care due to influenza showed a gradient of days lost, with the longest absences reported by cases attending hospital emergency departments, then those in physician office-based studies followed by community cases. 10 Additionally, studies that estimate the population-level burden of HRQoL and absences from only severe cases miss the majority of influenza illnesses which, despite their mild nature, are likely to contribute substantially to the overall burden. 5,11 Although household studies may capture these milder illnesses that do not result in health-seeking behaviour, and therefore provide less biased estimates, their specificity is often limited by a lack of laboratory-confirmed diagnoses.
There is therefore a need for robust estimates of the indirect effects of influenza from community studies identifying illnesses through prospective active symptom and molecular surveillance.
We have previously described the community burden of influenza, ILI and acute respiratory infections not meeting the definition of ILI across multiple influenza seasons in a large household cohort in England. 12 Here, we present the effects of these illnesses on HRQoL and work/school absences using the same cohort. We also estimate the population-level burden of these outcomes among community influenza cases.

| Study design
Flu Watch is a previously described, household-based, community cohort study of acute respiratory disease and influenza infection in England. 12,13 In brief, the study followed up cohorts during 6 influenza seasons including 3 periods of seasonal influenza (winters 2006-2007, 2007-2008 and 2008-2009)  Baseline surveys collected demographic, socio-economic and occupation data. Participants were categorised into "working" (employed full-time, part-time or self-employed), "students" (self-classified, aged 5-15) and "not in work/education." Participants were contacted weekly and asked to record any "cough, cold, sore throat or flu-like illness", which we define as an acute respiratory illness. During these illnesses, participants reported daily symptoms and temperature measurements using prospective illness diaries. Parents/guardians completed surveys on behalf of their children as needed. Self-administered nasal swabs were requested on day 2 of any illness. Participants submitted the swabs by mail to be tested for circulating influenza A viruses (H1N1, H3N2 and from 2009 onwards H1N1pdm09) and influenza B viruses using RT-PCR. 14,15 and QALYs were measured using the EQ-5D-3L instrument, [16][17][18] which was completed at baseline and daily throughout illness.

| HRQoL outcomes
Designed for self-completion, EQ-5D-3L has 2 components. The first describes health across 5 domains: mobility, self-care, usual activities, pain and anxiety. Participants rate each domain as "no problems," "some problems" or "extreme problems." Participants also record their overall health status on a visual analogue scale (EQ-VAS) from 0 (worst imaginable health state) to 100 (best imaginable health state). The online EQ-VAS question used in Flu Watch however asked participants to rate their health without the visual scale. The 3 possible responses for each of the 5 EQ-5D-3L domains results in 3 5 possible health states. These health states were mapped to an index value (representing a QALD weight) using a validated UK value set. 18 The QALD weights range between 1 (full health) and 0 (dead).

| Illness outcomes
All acute respiratory illnesses, regardless of swabbing or PCR result, were classified into 2 symptomatic outcomes. Those with confirmed fever (≥37.8°C) or symptoms of "feeling feverish" and either a cough or sore throat at any point were classified as influenza-like illnesses (ILI).
All other acute respiratory illnesses were classified as acute respiratory infections (ARI). Among the illnesses that had an accompanying swab, some were confirmed as PCR+ influenza cases and these were grouped into influenza A and influenza B viruses. In 2010/2011, when the EQ-5D-3L data were collected, all influenza A illnesses were H1N1pdm09, apart from 1 H3N2 case. The individual-level results report QALD loss for H1N1pdm09 cases only, but the population projections include H3N2.

| Time off work/education
The illness duration, percentages of illnesses with time off and mean number of days taken off were calculated for each illness outcome and stratified by age group and employment status. The latter 2 estimates were carried out separately for time off taken by the ill person, by someone caring for the ill person and a combination of both.

| HRQoL
Within each illness, the worst day of illness within each domain was identified. The percentage of respondents reporting no, some or extreme problems on their worst day in each domain was compared to the corresponding baseline responses, stratified by illness outcome.
T A B L E 1 Baseline characteristics of ill participants A sensitivity analysis was also conducted using the respondents' highest reported QALD weight as the comparison (baseline) group, regardless of when it was measured.

| Missing data
If a participant's baseline questionnaire was missing, then QALDs and QALYs could not be estimated for their subsequent illnesses. All illnesses with daily EQ-5D-3L measurements were included in the duration of illness, worst day EQ-VAS and QALD weight estimates. If a participant failed to complete illness diaries throughout their illness, then their illness duration would be truncated. We also investigated whether influenza cases actively reported no illness in the week following the last reported day of illness, or whether this weekly report was missing.

| Population impact
We estimated the total QALY loss experienced by community cases in the population and the number of days they took time off work/

| Time off work/education
Average illness duration, percentages of illnesses with time off and the symptom number of days per illness with time off were broadly comparable between influenza A and B cases although influenza A appeared slightly more severe ( T A B L E 2 (Continued)

| EQ-5D-3L
Those reporting problems and problem severity on the worst day of illness were broadly similar between H1N1pdm09, influenza B and ILI ( Figure 1 a-d). The most affected domains were "usual activities" and pain, followed by mobility, but all domains were affected.
The median and mean EQ-VAS background scores were between 84 and 90 for H1N1pdm09, influenza B and ILI, but dropped to between 40-50 on the worst day of illness (Figure 2, Table 3). Mean QALD weights were 0.93 and 0.92 at baseline for H1N1pdmo09 and influenza B, respectively, but dropped to 0.44 and 0.36 on the worst day of illness (Table 3). Median QALD weight for H1N1pdm09 (0.73) was much higher than the corresponding mean (0.44) suggesting that a few severe illnesses were greatly contributing to the mean ( Figure 2, Table 3).
For H1N1pdm09 and influenza B, daily EQ-VAS and QALD weights varied throughout illness, with a rapid decline in the first 2 days (Figure 3a-b). The lag time between symptom onset and the most severe day of illness appeared longer for H1N1pdm09 than for influenza B. Although the medians remain relatively low for the first week, over time these estimates reflected fewer illnesses, that is those with the longest duration (see bottom panels, Figure 3a-

| Missing data
One H1N1pdm09 and 2 influenza B illnesses were missing baseline EQ-5D-3L measurements. Among the 57 influenza cases with QALD data, all but 2 reported no illness in the week following their illness.

| Population impact
The estimated number of QALYs lost due to influenza A and B in England was 24 300 (95%CI: 16 600-34 700), of which two-thirds occurred in the 16-64 years age group (

| Comparison with other studies
Previous studies show substantial variation in the HRQoL asso- In general, our estimates for individual-level QALDs lost due to influenza were lower than earlier findings. This is unsurprising, as our study captured mild illnesses including cases of confirmed influenza that neither consulted for care nor met the symptom definition of ILI.
F I G U R E 2 EQ-VAS and EQ-5D QALD weights comparing background and worst day of illness by illness outcome  Additionally, our study included children who typically have less severe disease as well as a large number H1N1pdm09 cases which in our cohort were less severe than H3N2 cases. 12 This work and previous studies have shown that more QALDs are lost when estimates are derived from medically attended case, and in particular hospitalised cases. Our findings for work and school absences were also generally lower than previous estimates; for most illnesses, people did not take time off, although there were differences by age and illness definition. We showed however, that illness in a household member caused a substantial proportion of people take time off work to care for unwell household members. A study in the USA on school and parental absenteeism showed that for every 3 days a child took off school a parent missed on average 1 day of work. 22 The aforementioned British and Spanish studies are not directly comparable as they estimated the population-level burden of QALY loss due to influenza for more severe cases in a different season (2009/10). 5,11 They do however contextualise our findings as they report burden of QALY loss due to hospitalisations and deaths, which when combined with our results for community cases provides an indication of the scale of QALYs lost in a given season and the proportion attributable for different levels of disease severity. For example, the British study estimated that 40% (approximately 11 000 QALYs) of their total QALYs lost came from 337 reported influenza deaths. 5 Similarly, the Spanish study estimated their 318 deaths lost 12 000 QALYs. 11 It also estimated burden of QALY loss for influenza in-patients and primary care patients, demonstrating that less severe yet more numerous primary care patients lost far more QALYs (6778) than the more severe but less common in-patients (94 QALYs). Given these findings it seems that at least for these 2 seasons, the biggest contributors of population-level QALY loss are community cases (medically and nonmedically attended) and deaths. The true burden and contribution by level of severity are likely to vary substantially between seasons and populations as it dependent on population size and age-specific rates of illness and death. The estimated burden is also highly dependent on severity of cases included in the model.

| Strengths and weaknesses
Our estimates of HRQoL and work and school absence were derived from a large community cohort study using active molecular and symptom surveillance to identify episodes of influenza, ILI and ARI.
We captured a broad spectrum of illnesses including mild cases of laboratory-confirmed influenza that did not meet the syndromic definition of ILI and/or did not consult a healthcare professional, which gave less biased estimates of the overall HRQoL and absences associated with influenza. A key strength was that participants completed the EQ-5D-3L daily over the course of an illness. This directly meas- as H3N2 was associated with more severe symptoms than H1N1, its effects on HRQoL might have been greater. 12 Despite the large cohort size, the numbers with confirmed influenza and EQ-5D were relatively low (N = 58) and not sufficient to draw conclusions on differences in HRQoL by strain. The uncertainty in our QALD and QALY estimates is reflected in the 95% confidence intervals of our population projections. We previously showed that most influenza infections are either asymptomatic or produce only mild illness. 12 It is possible that we failed to capture very mild cases that did not shed enough virus to be PCR detectable. This would lead to a slight  1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25  T A B L E 4 Population-level burden of HRQoL lost and work/education absences due to community cases of influenza overestimation of individual-level QALD loss associated with influenza illness. Conversely, our population-level estimates should be considered minimum estimates because if we missed cases (eg from low viral shedding), this would reduce our estimated disease rates and thus overall burden estimates. We found some people reported worse HRQoL at baseline than during illness and our sensitivity analysis showed that when we took the participants' best reported measure of HRQoL as the comparison group, regardless of its timing, the oldest age group had much higher estimates of QALY loss. A further limitation is that children's HRQoL was reported by their parents. Previous studies show significant differences when both parents and adolescent measure children's quality of life. 23 Instruments such as EQ-5D-3L have not been validated for use in infants and very young children, which is a challenge of assessing HRQoL in this age group. 24

| Implications
Estimates of QALDs lost and work and school absences associated with influenza differ depending on the setting in which cases are identified; community illnesses result in smaller effects but contribute substantially to the population-level burden. Accurate assessment of both the number of expected cases and their QALDs/QALYs is essential to inform CUAs for decision-making bodies such as NICE. While for some interventions, such as antiviral treatments of severe influenza cases, it is appropriate to use utility estimates derived from medically attended cases, we believe that our estimates are more appropriate for assessing cost utility of community preventive interventions such as vaccines.

| CONCLUSIONS
We present new estimates of individual-and population-level QALDs and QALYs lost and work and school absences due to community cases of influenza to inform CUAs of community interventions to prevent influenza.

COMPETING INTERESTS
EBF, CWG, MZ and PJW report no conflict of interests. ACH serves on