Investigating obesity as a risk factor for influenza‐like illness during the 2009 H1N1 influenza pandemic using the Health Survey for England

Background Following the 2009 H1N1 influenza pandemic, obesity was shown to be associated with severe influenza outcomes. It remains unclear whether obesity was a risk factor for milder influenza‐like illness (ILI). Objectives To determine whether obesity was associated with an increased risk of self‐reported ILI during the 2009 H1N1 influenza pandemic using Health Survey for England (HSE) 2010 cross‐sectional data. Methods This study used HSE data collected from English households between January and December 2010. Weight and height measurements were taken by trained fieldworkers to determine obesity. ILI was defined as a positive response to the question “Have you had a flu‐like illness where you felt feverish and had a cough or sore throat?” with illness occurring between May and December 2009. Multivariable logistic regression was used to evaluate the association between obesity and ILI. Results The study comprised 8407 participants (6984 adults, 1436 children), among whom 24.7% (95% CI: 23.6–25.9) were classified as obese. Of obese participants, 12.8% (95% CI: 11.1–14.8) reported ILI compared to 11.8% (95% CI: 10.8–12.8) of non‐obese participants. The adjusted OR for ILI associated with obesity was 1.16 (95% CI: 0.98–1.38, P=.093). For adults and children, the adjusted ORs were 1.16 (95% CI: 0.97–1.38, P=.101) and 1.26 (95% CI: 0.72–2.21, P=.422), respectively. Conclusion Household survey data showed no evidence that obesity was associated with an increase in self‐reported ILI during the 2009 H1N1 influenza pandemic in England. Further studies using active prospective ILI surveillance combined with laboratory reporting would reduce bias and improve accuracy of outcome measurements.

factors. 2 Although obesity is associated with immune dysregulation at the cellular level, 3,4 it remains unclear whether it is associated with severe influenza outcomes through increasing the risk of acquiring influenza infection, promoting progression to severe disease after infection or both. 5,6 The vast majority of people with symptomatic influenza have mild illnesses and therefore do not seek medical attention 7 ; rates of laboratory testing increase with severity of illness. Studies based on medical records and/or laboratory testing thus fail to capture much of the burden of community influenza-like illness (ILI) 8 and may not yield insights transferable to the wider population. Studies using selfreported height and weight measurements to calculate body mass index (BMI) may be subject to recall bias and misclassification due to participants reporting a more socially desirable height and weight. 9 Other studies based on BMI measurements extracted from medical records may be limited by lack of timeliness and accuracy of BMI recording.
In this study, we use data from the Health Survey for England (HSE) 10 -a nationally representative population-based study in which obesity classification is based on weight and height measurements taken by trained fieldworkers-to determine whether obesity was associated with an increased risk of self-reported ILI during the 2009 H1N1 influenza pandemic.

| Data source and population
The 2010 HSE data set was used for this study. 11 It comprised data from a sample of adults aged 16+ years and children 0 to 15 years representative of private households in England. HSE methods are described elsewhere. 12 All data used in our study, including detailed information on social and demographic characteristics, lifestyle behaviours, health and physical measurements such as weight and height, were collected during household visits by trained interviewers, which took place throughout the 2010 calendar year. For inclusion in this study, HSE participants had to have both a valid measure of obesity and a valid response to the question used to identify influenza-like illness.

| Definition of outcome and exposure
The outcome of interest was ILI experienced during the first eight months of the H1N1 pandemic between May 2009 and December 2009. The question used to identify ILI was "Since May 2009, have you had a flu-like illness where you felt feverish and had a cough or sore throat?" Permitted responses were "yes" and "no" with the responses "don't know" and "refused" recoded to missing. If participants responded "yes," they were asked to report the month and year of illness. Illnesses reported to occur after December 2009 were not included in the outcome definition.
The primary exposure, obesity, was based on BMI measurements for adults aged 16 and over and on age-and sex-based population centiles for children aged 2-15 years. Those aged under two did not have height measurements so were excluded from analyses. Interviewers measured height on those aged 2+ and weight for all participants. For participants exceeding the weight limit of the scales (130 kg), self-reported weights were used to calculate BMI. BMI measurements considered unreliable (i.e. pregnant women, those who refused to be measured, measurements that were attempted but not obtained, measurements that were not attempted or measurements that were not useable) were excluded from analyses. Adult participants with a BMI 30 kg/m 2 or above were classified as obese; children were classified as obese if their weight exceeded the 95th centile. 13 A binary variable for obesity (obese vs non-obese) was generated that combined results for adults and children. We also conducted two sensitivity analyses: first we reclassified BMI for adults as underweight/ normal (BMI <25), overweight (BMI 25-29.999) and obese (BMI≥30), then removed the overweight category to compare obese with underweight/normal weight individuals; second, we used the waist hip ratio variable as an alternate measure of obesity in adults only, which was classified as either "not at increased risk" or "substantially increased risk" according to standard cut-offs. 14 Only participants with a valid waist hip ratio measurement were included in this analysis.

| Potential confounding and effect modifying variables
We considered variables as potential confounders or effect modifiers

| Statistical analysis
Data analysis was conducted using Stata version 13.0 (Stata Corporation, TX, USA). We described baseline characteristics of participants by calculating the frequency and survey weighted percentage for all categories of variables of interest. Weighting was used to correct the distribution of household members to match population estimates for sex/age groups and geographic region, as well as to correct for bias resulting from individual non-response within households. 15 Stata's "svy" suite of commands was used to account for the complex survey design. We used univariate logistic regression analysis to generate an odds ratio with 95% confidence interval for the association between obesity and influenza-like illness, with participants who were not obese (BMI <30 kg/m 2 ) forming the baseline reference group. Univariate associations between all potential confounders and obesity and separately ILI were investigated through chi-squared tests and logistic regression analysis. We also drew a causal diagram using the programme DAGitty v 2.3 (http:// www.dagitty.net/dags.html#) to inform choice of variables for inclusion in multivariable logistic regression models (see Supplementary material). Multivariable models were then generated: the main model included all theoretically relevant confounders associated with both outcome and exposure, for which adjustment was identified as necessary using our causal diagram. Each variable added was examined for multicollinearity. For asthma, COPD and smoking, we generated a composite binary variable which was positive if any of these variables was positive. For each model, we examined the change in effect size and the Wald test P-value compared to the crude model. Potential effect modifiers identified based on existing literature (influenza vaccination, type of diabetes and age) were also evaluated using interaction terms within the final logistic regression model. The Wald test P-value was used to assess the strength of the interaction. Analyses were repeated separately for adults and children. Other sensitivity analyses were conducted for adults only to investigate the effect of varying the definition of obesity. These were 1) removing the overweight category to compare obese individuals with normal/underweight individuals and 2) using waist-hip ratios in place of BMI as a measure of obesity.

| Participant characteristics
Between January and December 2010, the HSE selected 8736 households, of which 90.8% met the inclusion criteria and 66% participated.
From the core HSE sample of 10 494 participants interviewed, 2803 participants were excluded due to non-valid BMI measurements and a further four because of invalid influenza-like illness responses, leaving 8407 participants in our study (Fig. 1).
These were 6984 adults with a median age of 48 years (IQR 35-63 years), and 1436 children with a median age of 9 years (IQR

| Univariate analysis
Among obese participants, 12.8% (95% CI: 11.1-14.8) experienced ILI between May to December 2009 compared to 11.8% (95% CI: 10. 8-12.8) of participants who were not obese. The unadjusted OR was 1.11 (95% CI: 0.93-1.31, P-value: .241). The highest odds of reporting ILI were seen in the age group 25-34.9 years, and ILI was least reported among people aged 65 years and over. Asthma, COPD and current smoking were associated with an increase in ILI reporting on univariate analysis, while hypertension was associated with a small decrease. The frequency of ILI reporting also varied by interview season and was most common in spring. People with no qualifications or foreign/other qualifications reported less ILI than those with higher levels of education. Obesity was associated with age, ethnicity, influenza vaccination, high blood pressure, diabetes, asthma, smoking status in adults, alcohol consumption in adults, COPD, household size, IMD score and education on univariate analysis ( Table 2).

| Multivariable analysis
In multivariable analysis across the whole study population (adults and children), the adjusted OR for the effect of obesity on likelihood

| Sensitivity analyses
In sensitivity analyses of adults only, removing the overweight category (n=2627) to compare ILI reporting in obese adults (n=1908) with normal/underweight adults (n=2449) resulted in an adjusted OR of 1.14 (95% CI 0.

| DISCUSSION
We found no evidence that obesity was associated with an increase in self-reported ILI during the 2009 H1N1 pandemic in English households using representative population data from the HSE. In our data set, people aged 25-35 were most likely to experience ILI and the over 65 age group were least likely to report ILI, consistent with other data from the pandemic. 8 We have previously shown that ILIs reported in the HSE 2010 show a similar pattern and age distribution to infections identified in the Flu Watch cohort study, 16 although the overall magnitude of ILI was considerably less. The Flu Watch cohort was designed to estimate the community burden of ILI by collecting data using active weekly prospective follow-up but did not measure obesity.
Meta-analysis of hospitalization and death data from the 2009 pandemic suggests that obesity is an independent risk factor for severe outcomes of pandemic influenza. 1 Individual studies report similar findings for seasonal influenza. 2,17,18 One case cohort study found that obese adults aged 20-59 years had an increased risk of attending outpatient clinics with ILI symptoms than those of normal research will be invaluable for informing healthcare planning, guiding targeting of resources and informing governments to ensure a proportionate response to future influenza pandemics.