Comparative community burden and severity of seasonal and pandemic influenza: results of the Flu Watch cohort study

Summary Background Assessment of the effect of influenza on populations, including risk of infection, illness if infected, illness severity, and consultation rates, is essential to inform future control and prevention. We aimed to compare the community burden and severity of seasonal and pandemic influenza across different age groups and study years and gain insight into the extent to which traditional surveillance underestimates this burden. Methods Using preseason and postseason serology, weekly illness reporting, and RT-PCR identification of influenza from nasal swabs, we tracked the course of seasonal and pandemic influenza over five successive cohorts (England 2006–11; 5448 person-seasons' follow-up). We compared burden and severity of seasonal and pandemic strains. We weighted analyses to the age and regional structure of England to give nationally representative estimates. We compared symptom profiles over the first week of illness for different strains of PCR-confirmed influenza and non-influenza viruses using ordinal logistic regression with symptom severity grade as the outcome variable. Findings Based on four-fold titre rises in strain-specific serology, on average influenza infected 18% (95% CI 16–22) of unvaccinated people each winter. Of those infected there were 69 respiratory illnesses per 100 person-influenza-seasons compared with 44 per 100 in those not infected with influenza. The age-adjusted attributable rate of illness if infected was 23 illnesses per 100 person-seasons (13–34), suggesting most influenza infections are asymptomatic. 25% (18–35) of all people with serologically confirmed infections had PCR-confirmed disease. 17% (10–26) of people with PCR-confirmed influenza had medically attended illness. These figures did not differ significantly when comparing pandemic with seasonal influenza. Of PCR-confirmed cases, people infected with the 2009 pandemic strain had markedly less severe symptoms than those infected with seasonal H3N2. Interpretation Seasonal influenza and the 2009 pandemic strain were characterised by similar high rates of mainly asymptomatic infection with most symptomatic cases self-managing without medical consultation. In the community the 2009 pandemic strain caused milder symptoms than seasonal H3N2. Funding Medical Research Council and the Wellcome Trust.


Introduction
Infl uenza causes roughly 250 000-500 000 deaths worldwide each year. 1 In the 20th century there were three infl uenza pandemics for which there are varying mortality estimates: 1918 A/H1N1 at least 20-40 million excess deaths, 1957 A/H2N2 about 4 million excess deaths, and 1968 A/H3N2 about 2 million excess deaths. [2][3][4] In 2009 a new pandemic virus, 5 infl uenza A(H1N1)pdm09, emerged in Mexico 6 and spread globally over 2009-10, causing an estimated 200 000 respiratory deaths and 83 000 cardiovascular deaths during the fi rst 12 months of circulation. 7 WHO declared an end to the pandemic on Aug 10, 2010. 8 However, a further pandemic wave occurred in some European and other countries outside North America 9 in 2010-11 with reports of excess deaths in, for example, England. 10 Internationally, infl uenza activity surveillance provides real-time information to inform prevention and control policy. 11 Surveillance focuses on cases seeking medical attention: the so-called tip of the iceberg of infection. Underestimation of the number of community cases leads to overestimates of severity. 12,13 Heightened concern during a pandemic can change patient consultation thresholds and clinician recording and investigation behaviour, thus distorting surveillance information. 14 Information on the community burden of infl uenza is key to informing control, 15 but is not routinely collected. For example, infl uenza transmission models, which are widely used to consider the effi cacy and cost-eff ectiveness of vaccines, antivirals, and non-pharmaceutical counter measures, depend on valid epidemiological estimates of the community occurrence of disease. The available data for periods of seasonal infl uenza are largely derived historically from household cohort studies of families with children in communities in the USA between 1948 and 1981, [16][17][18][19] and a more recent study from rural Vietnam. 20 There have also been some cohort studies reporting on the 2009 pandemic from Hong Kong, southeast Asia, and Mali 21-25 as well as several cross-sectional serosurveys from this period. 26 Case-ascertained household transmission studies can estimate the secondary attack proportion and eff ects of interventions within households; however, they are not designed to estimate community burden of infl uenza infection and disease. 27,28 The Flu Watch study is the fi rst national community cohort study of infl uenza occurrence enrolling households with and without children, with the additional benefi t of modern molecular diagnostic techniques.
We aimed to compare the community burden and severity of seasonal and pandemic infl uenza across diff erent age groups and study years and gain insight into the extent to which traditional surveillance underestimates this burden. Our specifi c objectives were to measure the proportion of the population infected each season, the proportion of those infected who developed symptomatic disease attributable to infl uenza, the proportion who had detectable nasal shedding of infl uenza virus, the symptoms among those with confi rmed infl uenza, and the proportion who were medically attended. During the pandemic we also aimed to measure the development of type-specifi c immunity to the pandemic strain.

Participants
We did a household-level community cohort study of acute respiratory illness and infl uenza infection, recruiting households across England (appendix). We followed up successive cohorts over the 2006-07, 2007-08, and 2008-09 periods of seasonal infl uenza circulation, and the fi rst (spring and summer 2009), second (autumn and winter 2009), and third (winter 2010-11) waves of the pandemic. Households were recruited annually through written invitation sent to a random sample of people registered with 146 volunteer general practices as well as inviting previous participants (in . In England, most of the population is registered with a general practice. 29 At baseline (October-December) and follow-up visits (May-July) of each year we collected blood samples for serological testing (willingness to provide samples was a condition of participation in adults, voluntary in children aged 5-15 years, and not requested in children younger than 5 years). Follow-up samples from spring 2009 acted as baseline specimens for individuals who continued to participate through the 2009-10 pandemic.
Participants gave written informed consent (proxy consent for children). The protocol was approved by the Oxford MultiCentre Research Ethics Committee. (06/Q1604/103).

Procedures
We collected demographic and medical history data at baseline and self-reported vaccination status at baseline and end of follow-up. Admissions to hospital and deaths during follow-up were recorded with the end of season follow-up questionnaire completed by the lead householder, with deaths among participants also being reported to the study by participating practices and directly by families. We minimised recall bias of illness through weekly telephone or online surveys to record any "cough, cold, sore throat, or fl u-like illness" among household members. In addition to weekly surveys, participants were asked to complete detailed daily symptom diaries for the duration of any acute respiratory illness, including daily temperature measurement and reporting of several symptoms: feeling feverish, headache, having muscle aches, cough, sore throat, runny nose, blocked nose, and sneezing. Symptoms were allocated a numerical score on the basis of severity (0=absent, 1=mild, 2=moderate, 3=severe or for fever <37·8°C=0, 37·8-38·9°C=1, 39·0-39·9°C=2, ≥40°C=3). Review of participants' primary-care records was used to measure consultation behaviour in practices where a research nurse was available to extract the data. We asked participants to submit, by mail, nasal swabs on day 2 of any illness. These swabs were transported in viral transport medium and screened by RT-PCR for infl uenza A (subtypes H1, H3), infl uenza B, infl uenza A(H1N1)pdm09 (from 2009 onwards), and a panel of other respiratory viruses including respiratory syncytial virus, rhinovirus, coronavirus, adenovirus, human metapneumovirus, and parainfl uenza virus with methods described elsewhere. 30,31 We measured serum antibody titres against circulating infl uenza strains (appendix) in baseline and follow-up samples with haemagglutination inhibition assay using standard methods. 32,33 Outcomes Key outcomes of interest were infection with infl uenza, defi ned as a four-fold titre rise in serum samples of unvaccinated individuals (but not in vaccinated individuals since both vaccination and natural infection lead to titre rises); occurrence of any acute respiratory illness (selfreported "cough, cold, sore throat, or fl u-like illness"); occurrence of infl uenza-like illness, defi ned according to the US Centers for Disease Control and Prevention (CDC) defi nition of fever (temperature ≥37·8°C) and a cough or a sore throat in the absence of a known cause other than infl uenza; 34 occurrence of PCR-confi rmed infl uenza; symptom severity over the fi rst week of illness in PCRconfi rmed cases; and consultation with primary care. During the pandemic additional outcomes included monitoring the development of immunity (defi ned as antibody titre to infl uenza A H1N1 pdm2009 of ≥32). In this analysis key predictors of interest are age, study year, and circulating strain of infl uenza. The study size was chosen to give accurate annual estimates of infection and disease rates such that a sample size of 800 per year would allow a 25% risk of infection to be estimated within 95% CIs of 22-25 and a 10% risk of infl uenza-like illness within 95% CIs of 8-12. The study was expanded in the pandemic to provide accurate real-time measures of infl uenza-like illness rates recruiting as many participants as was practically possible.

Statistical analysis
Analyses were done in STATA version 12. Analyses of serological data were restricted to those with serological samples available (no children younger than 5 years had serological specimens). We weighted analyses to the age and regional structure of England to give nationally representative estimates. In this weighting, children younger than 15 years were considered as a single group, so measures of age-adjusted population rates of infection (but not of PCR-confi rmed disease or illness) apply the rates in the 5-15 year age group to the 0-15 population. We did not weight on ethnic origin or social deprivation because there was no evidence of a strong association with infection or disease rates (data not shown) and small or zero numbers in some groups would have led to instability of weighted measures.
Participants were assumed not to have a respiratory illness in weeks with missing illness status reports. After excluding illnesses where PCR identifi ed a non-infl uenza virus we plotted rates of respiratory illness, infl uenza-like illness, and PCR-confi rmed infl uenza (per 100 000 personweeks). We estimated the percentage of the population infected each season by calculating age and season-specifi c rates of serological infection and PCR-confi rmed disease per 100 person-seasons. A person-season was defi ned as the time from the fi rst PCR isolation of infl uenza in the cohort to the last isolation in any given season, rates therefore accounted for diff erential follow-up time during periods of infl uenza circulation. We did not undertake follow-up blood samples for serological testing for participants recruited before the fi rst (spring/summer) pandemic wave until after the second (winter) wave had fi nished because we had not predicted separate summer and winter pandemic waves. Methods to derive infection rates in the fi rst wave are described in the appendix.
We estimated the percentage of serological infections leading to illness by two independent methods. First, we calculated age-adjusted attributable rates of illness due to infection (subtracting rates of respiratory illness in nonseroconverters from those in seroconverters). 35   analyses infl ated these adjusted attributable rates to account for the recorded level of under-reporting (based on the proportion of expected weekly illness status reports received during periods of infl uenza circulation). Second, we measured the proportion of unvaccinated seroconverters with PCR-confi rmed infl uenza. We compared symptom profi les (an ordered categorical variable) over the fi rst week of illness for diff erent strains of PCR-confi rmed infl uenza and non-infl uenza viruses using ordinal logistic regression with symptom severity grade as the outcome variable, adjusting for age group and strain type and accounting for repeated measures in individuals using robust standard errors (Stata ologit commands with cluster option).

Role of the funding source
The sponsors had no role in study design, collection analysis, interpretation of data, or writing of the report. ACH, EBF, and ERCM had access to the raw data. The corresponding author had full access to all data and fi nal responsibility to submit for publication.

Results
Roughly 10% of invited households agreed to participate (appendix). Table 1 presents the comparison of unweighted cohort demographics with those of the England population showing good geographical spread but underrepresentation of young adults; people living in socially deprived areas, north England, west Midlands, and London; and people of non-white ethnic origin.
The highest rates of infl uenza-like illness and PCRconfi rmed infl uenza were during the epidemic of H3N2 in 2008-09 before the pandemic and in the 2010-11 third pandemic wave. Compared with other seasons, illness rates in the fi rst pandemic wave were low (fi gure 1).
Risk of PCR-confi rmed disease tended to decrease with increasing age (fi gure 2, appendix). For infl uenza A this age dependence was most apparent during the H3N2 epidemic of 2008-09 and the 2009-10 second wave of the H1N1 pandemic when children had signifi cantly higher rates of PCR-confi rmed disease and serological infection with infl uenza A than older adults (appendix). For infl uenza B, children had signifi cantly higher rates of PCR-confi rmed disease than adults in 2008-09 and 2010-11 seasons (appendix). The 2010-11 third wave of the H1N1 pandemic was unusual in having markedly higher rates of infl uenza A in young adults than any other season. PCRconfi rmed infl uenza was very rare in people older than 65 years in all seasons.

Figure 1: Rates of illness or PCR-confi rmed infl uenza standardised by age and region
Rates of acute respiratory illness, infl uenza-like illness, and PCR-confi rmed infl uenza per 100 000 person-weeks. Excludes illnesses known to be due to non-infl uenza viruses.  Infection rates were typically highest in children aged 5-15 years (not measured in younger children) and decreased with age (fi gure 2, appendix). This agedependence was most apparent during the 2009 fi rst pandemic wave when children aged 5-15 years was the only age group with measurable risk of infection: 26% (0-58). Age dependence of infl uenza A was also strong during the 2009-10 second wave of the pandemic when children had signifi cantly higher rates of serological infection with infl uenza A than older adults (appendix). The 2010-11 third pandemic wave was the only season when young adults aged 16-44 years had the highest risk of infection: 34% . During periods of seasonal infl uenza A age dependence was strongest during the H3N2 epidemic of 2008-09 (appendix). Children also had signifi cantly higher rates of infl uenza B infection than older adults in 2008-09 (appendix).
Most infections were asymptomatic. 192 respiratory illnesses including 70 infl uenza-like illnesses were reported from 327 participants with serological evidence of infection over 280 person-seasons of follow-up (69 respiratory illnesses, 25 infl uenza-like illnesses per 100 personseasons). There were 623 respiratory illnesses including 95 infl uenza-like illnesses reported from 1742 participants with no serological evidence of infection over 1423 personseasons of follow-up (44 respiratory illnesses, seven infl uenza-like illnesses per 100 person-seasons). The rate of respiratory illness attributable to infl uenza (age-adjusted incidence rate diff erence) 35   (95% CI 13-34) including 18 infl uenza-like illnesses per 100 person-seasons (95% CI 12-24). There was insuffi cient power to test the hypothesis that the asymptomatic proportion varied by age or strain type. Sensitivity analyses adjusting for the fact that 85% of illness status reports were returned during periods of infl uenza circulation gave estimates of the rate of respiratory illness or infl uenza-like illness attributable to infection of 27 respiratory illnesses and 21 infl uenza-like illnesses per 100 person-seasons, respectively. These estimates of infections leading to disease are similar to the 25% (18-35) of people with serological infections who had PCR-confi rmed infl uenza from nasal swabs. PCR-confi rmation levels seemed to be lower in adults aged 65 years or older (9%, 95% CI 1-60) and for infl uenza B infections (5%, 1-28), although this was not statistically signifi cant.
Most people with PCR-confi rmed infl uenza did not consult and among those who did, infl uenza or infl uenza-    Of 133 PCR-confi rmed cases of infl uenza, with data available from end of season surveys, there was one admission to hospital potentially attributable to infl uenza (febrile convulsions in a child younger than 5 years within 2 weeks of a positive swab for infl uenza A H1N1 pdm2009). There were no deaths among these 133 PCR-confi rmed cases. This single admission gives a maximum estimated hospitalisation rate of 0·75% (95% CI 0·02-4·19). Of 226 seroconverters to infl uenza, with data available from end of season surveys, there were two admissions to hospital that were potentially attributable to infl uenza: one in a young adult with a four-fold titre rise to infl uenza A H1N1 pdm2009 admitted with a chest infection in the winter of 2010-11 and one in an individual aged 45-64 years with a four-fold rise in titre to infl uenza B admitted with pneumonia during the winter of 2010-11. These two admissions give a maximum estimated hospitalisation rate for serological infection of 0·88% (95% CI 0·11-3·19). This compares with three respiratory hospitalisations in 1730 participants who did not have a four-fold titre rise (0·17%, 95% CI 0·04-0·51). There were two respiratory deaths in the cohort, both of which occurred in vaccinated participants older than 65 years during the 2008-09 winter season; one was partly attributable to chest infection and the other was attributable to pneumonia. It is not possible to infer whether or not infl uenza contributed because there were no nasal swab samples and post mortem serum samples were not sought.
Primary-care-based surveillance greatly under estimated the extent of infection and illness in the community (fi gure 3). Under ascertainment was lower during the summer wave of the pandemic. The rate of PCR-confi rmed infl uenza across all winter seasons was on average 22-times higher (95% CI 17-28) than rates of PCRconfi rmed disease from the Royal College of General Practitioners Sentinel Infl uenza-Like Illness/Virological Surveillance Scheme. 36 During the pandemic summer wave the rate was only three-times higher.
In children aged 5-15 years, protective antibodies were mainly acquired as a result of natural infection over the fi rst and second pandemic wave, in young adults (16-44 years) protective antibody levels increased mainly as a result of natural infection in the second and third wave, in people older than 45 years protective antibodies were mainly acquired as a result of vaccination during the second and third waves (fi gure 4).

Discussion
Flu Watch is one of the largest and most comprehensive studies of its type since the 1980s (panel). 19 Through providing reliable information on the epidemiology and burden of seasonal and pandemic infl uenza we hope to inform future decisions on seasonal disease control and pandemic planning. For example, better measures of community disease burden will increase the validity of population models that are widely used to infer the effi cacy and cost-eff ectiveness of countermeasures such as use of antivirals, vaccines, and behavioural  interventions. On average infl uenza infected 18% of unvaccinated people each winter. Up to three-quarters of infections were asymptomatic and about a quarter of infections had PCR-confi rmed disease. 17% of people with PCR-confi rmed disease had medically attended illness. These data did not vary signifi cantly when comparing pandemic with seasonal infl uenza. People infected with the 2009 pandemic strain had markedly less severe symptoms than those infected with seasonal H3N2.
The study covers six infl uenza seasons, including seasonal and pandemic periods, but the variable nature of infl uenza means that we cannot exclude substantially more severe periods of both seasonal and pandemic infl uenza in the future. The study was limited by the diffi culty in obtaining a fully representative sample because, although selection was random, acceptance rates were low. Weighted analyses ensured results represented the age and regional structure of the country. Weighted analyses of overall population rates of infection (but not illness or disease) assume the level of infection recorded in children younger than fi ve years were similar to those in those aged 5-16 years. However, because children younger than 5 years represent only about 6% of the UK population the lack of serological analysis in this group will not have made a major diff erence to overall reported rates. Our estimates of infection in people older than 65 years are limited to unvaccinated individuals because we cannot reliably infer infection from titre rises in vaccinated people.
Diff erences in volunteers such as propensity to consult if ill and uptake of interventions might aff ect fi ndings. Participants were generally highly diligent at completing weekly illness reports and submitting nasal swabs during illness (>85% completion). That the proportion of infections judged symptomatic was very similar using two independent methods (proportion of seroconverters PCRconfi rmed and attributable-rate of respiratory illness in seroconverters compared with non-seroconverters) 35 suggests that the low proportion with PCR-confi rmed disease is not simply a matter of low test sensitivity. Nasal swabs have similar sensitivity to the gold-standard nasopharyngeal aspirates. 37,38 Self-taken swabs have similar sensitivity to those taken by health-care workers. 39 The extent of postal delay is not associated with likelihood of PCR positivity for infl uenza. 40 After adjusting for time from symptom onset to swab date, a recent comparison showed similar levels of viral detection and viral load in primary-care samples from patients with infl uenza-like illness to those in postally submitted samples from patients consulting the National Pandemic Flu Service. 41 Because of the high level of completion of weekly illness status reports, adjusting our attributable rates for the recorded level of under-reporting made minimal diff erence to our conclusions on the proportion of infections leading to illness. The attributable-rate method might underestimate the proportion of infections leading to illness if having infl uenza reduces the risk of other respiratory infections through viral interference, 42 but we did not identify any evidence of this in our data.
Reported cases of infl uenza represent the tip of a large clinical and subclinical iceberg that is mainly invisible to routine surveillance systems (fi gure 3). Surveillance does not aim to capture the totality of community cases, but low ascertainment means that changes in consultation and reporting behaviour during periods of increased concern can make interpretation of trends highly problematic. It is also likely that propensity to consult if ill might vary by country, making international comparisons diffi cult. Our fi ndings show this during the summer wave of the pandemic, where comparison of disease rates with national surveillance suggest a much higher proportion of cases were identifi ed with surveillance than at other times. This greater identifi cation probably represents an increase in propensity for people to consult and be reported during periods of increased national concern; however, because only two confi rmed cases were identifi ed in the study  18 the methods of which were adapted for the Flu Watch study. These studies are summarised elsewhere. 19 We also reviewed fi ndings from more recent cohort studies using similar designs. [20][21][22][23] We then did three separate systematic literature searches in Embase, Global Health, and Medline databases (up to Aug 13, 2013) to identify community-level cohort studies of infl uenza with PCR or serological confi rmation (search terms in appendix).
To identify work relevant to establishing population-level infl uenza infection rates, we reviewed a recent review and meta-analysis of age-specifi c cumulative incidence for the 2009 infl uenza pandemic. 26 We also searched published work for community-level cohort studies of infl uenza that collected paired sera (search terms in appendix). Studies were included in this review if they were community-level cohorts but excluded if they only focused on subsections of the population.
To add to the work identifi ed above, we did a further two systematic literature searches in the same databases to identify relevant community-level prospective cohort studies that had PCR confi rmation of infl uenza illnesses (search terms in appendix). Studies were included if they were community-level cohorts with prospective follow-up of respiratory illnesses with swabbing for PCR confi rmation of infl uenza. Studies were excluded if they only focused on subsections of the population or if they only did PCRs on a subset of illnesses (ie, only consulting illnesses or only those meeting a case defi nition which would exclude less severe infl uenza).

Interpretation
Our rates of infection and disease were similar to those found in historical cohort studies and in more recent, mostly pandemic, studies. [18][19][20] Our fi nding that a high proportion of infections were asymptomatic was consistent with other research. 35 Infl uenza A H3N2 has previously been reported as causing more severe symptoms than Infl uenza A H1N1. 35 Others have also reported that a high proportion of people with infl uenza-like illness do not seek medical attention. 14 during this summer period we cannot test this directly. From a population perspective there was little diff erence in burden of disease in the 2009 pandemic compared with seasonal infl uenza periods. Both seasonal and pandemic infections were common, especially in children, but most infection was asymptomatic and most symptomatic cases did not consult. The symptoms associated with 2009 pandemic infl uenza were substantially milder than those of seasonal H3N2. The completeness of return of weekly illness status reports did not alter during the pandemic, suggesting that change in reporting to the study was not a source of bias (appendix). Hospitalisation and death rates were very low with insuffi cient events to establish if this varied by strain type.
After the high attack rates during the 2009-10 pandemic, WHO advised that the virus had largely run its course. 8 Despite this, a further wave of A(H1N1)pdm09 infection was noted across Europe in 2010-11. 9 In England this wave was associated with an upwards age shift in infected people admitted to hospital (median age 35 years during the 2010-11 winter compared with 20 years over the 2009-10 pandemic period) and pressure on intensive care. 43 We did not identify any evidence of increased symptom severity in the third wave, but there was a marked increase in the number of cases in young adults. This upwards age shift is probably explained by the fact that before this third wave, protective antibody titres were present in about threequarters of children, but only about a third of working-age adults. However, it remains unclear why young adults were not more aff ected during earlier waves. The higher numbers of deaths and admissions noted nationally during the pandemic third wave were therefore probably due to increased numbers of cases in young adults rather than, as reported by others, increased severity. 44 The occurrence of a third wave in the pandemic reinforces the value of investing in vaccine development even if vaccines are unlikely to be available suffi ciently early to alter the course of early waves. The proportion of serologically confi rmed infections that are asymptomatic is an often neglected variable, which is an important component of severity. Our fi nding that only 23% (95% CI 13-34) of infections are symptomatic is lower than is sometimes assumed, but is consistent with fi ndings from other studies of seasonal infl uenza 35 and human challenge studies. 45 Measurement of the proportion of serologically confi rmed infections that are asymptomatic should be an early priority for any emerging infection of pandemic potential. This provides an additional index of severity complementing population level data on admission to hospital and deaths. Our study was limited by our inability to identify virus shedding during asymptomatic infection because this would have needed frequent regular nasal swabbing throughout follow-up. Comparison of symptom profi les in community cases of an emerging infection with those noted with other viruses can also provide important information on severity, complementing population level data on numbers of admissions to hospital and deaths.
Application of consistent methods across periods of seasonal and pandemic infl uenza has allowed assessment of the 2009 pandemic in context. Although A(H1N1) pdm09 infected most children and a high proportion of unvaccinated adults of all ages over three pandemic waves, infection and disease rates were similar to or lower than prepandemic periods. Symptoms due to infection with the 2009 pandemic strain were milder than those with seasonal infl uenza strains. Our fi ndings are consistent with the very low case-fatality rates recorded during the pandemic. Whereas overall, pandemic illnesses were mild there was a shift in the age distribution of deaths leading to an increase in years of life lost during the 2009 pandemic compared with some seasonal periods. 7 Despite its mild nature, the 2009 pandemic caused enormous international concern, expense, and disruption. We need to prepare for how to respond to both mild and severe pandemics. To do this we need more refi ned assessments of severity, including community studies to guide control measures early in the course of a pandemic and inform a proportionate response.

Contributors
ACH is the principal investigator of the Flu Watch study. He conceived the idea for and designed the original seasonal infl uenza study and the pandemic extension in discussion with JMW, AMJ, and MZ. AB, WJE, AMcM, JSN-V-T, IN, JMW, AMJ, and MZ were coapplicants on the seasonal and pandemic grants, members of the steering group, and contributors to study design. AC, NF, and RP were coapplicants for the pandemic extension and joined the steering group in 2009, contributing to the design of the pandemic phase. EBF, JK, and ERCM also contributed to the study design. Data collection was led by Flu Watch project managers JK (2006-08) and GH . EBF, AB, LW, ERCM, and FBW also contributed to data collection. EBF, Flu Watch statistician 2008-11, led the data management with contributions from JK, MSCL, ERCM, and FBW. ACH and EBF led the development of the overall analytical strategy with contributions from all steering group members as well as NG, ERCM, and FT. AB and MZ led the laboratory analysis with contributions from LW. AC was the statistical adviser for the study. FT also contributed statistical advice. ACH and EBF analysed the data. All steering group members and EBF, LW, NG, JK, MSCL, and ERCM contributed to interpretation of the fi ndings. ACH wrote the report with contributions from EBF and AB and input on drafts from all authors. Literature search was done by ACH and EBF. Tables and fi gures were created by ACH, EBF, and ERCM.

Declaration of interests
JSN-V-T has served on speaker bureaus for, served as a consultant to, and received grants and support for travel from Roche and GSK, but all personal remuneration stopped in Sept, 2010. He also received support for travel from Roche. IN was funded by GSK as chief investigator on a prospective, observational, multicentre, cohort, post-authorisation safety study of GlaxoSmithKline Biologicals A/California/7/2009 (H1N1) v-like pandemic vaccine adjuvanted with AS03. This study was part of the Medicines and Healthcare products Regulatory Agency's procedures for approval of the vaccination from 2009 to 2011. MZ has received funding from vaccine companies (Sanofi , Novartis, CSL, Baxter, GSK) and Roche for antiviral work. AMJçs been a Governor of the Wellcome Trust since 2011. ACH, WJE, NF, JSN-V-T, RP, JMW, and MZ have served on UK national advisory committees relevant to planning and response for seasonal and pandemic infl uenza. Authors not specifi cally mentioned declare that they have no competing interests.