We used three datasets: patients hospitalized with 2009 pandemic influenza A(H1N1), deaths from 2009 pandemic influenza A(H1N1), and the National Health and Nutrition Examination Survey (NHANES), a representative survey of the general U.S. population. For each dataset we defined individuals according to BMI category (see below), presence of ACIP-recognized chronic medical conditions (defined for this analysis as cardiovascular disease, pulmonary disease, liver disease, cancer, and diabetes) and age group. We did not include neurologic, neuromuscular, hematological, non-diabetes metabolic disorders or immunosuppression as ACIP-recognized chronic medical conditions in this analysis due to lack of this information from NHANES. We used two age groups, 2–19 years and ≥20 years: dividing the ≥20 year old age group into additional age categories resulted in unreliable variance estimates when we stratified by BMI category and presence of ACIP-recognized chronic medical conditions due to zero observations in primary sampling units of the NHANES survey.
For hospitalized patients and NHANES participants we used BMI [height (m)/weight (kg)
2] as an index of relative weight. For individuals aged 2–19 years, we calculated age- and sex-specific BMI z-scores and used CDC 2000 growth charts to define BMI category, where underweight was <5
th percentile, normal weight 5–84
th percentile, overweight 85–94
th percentile, and obese was ≥95
th percentile.
[10],
[11],
[12] For individuals ≥20 years we defined underweight as BMI<18.5, normal weight as 18.5–24.9, overweight as 25.0–29.9, obese as 30.0–39.9, and morbidly obese ≥40.0.
[13],
[14] For deaths, no data were available on measured height and weight, so we used mention of obesity or morbid obesity within the medical condition section of the reporting form. For all three datasets we excluded individuals <2 years old and pregnant women because obesity definitions are not applicable to them.
We included information on 516 (10%) of 5009 patients hospitalized with 2009 pandemic influenza A(H1N1) in the United States between May 1–July 12, 2009. These patients were a convenience sample of all hospitalizations reported to CDC from 24 states as part of prospective surveillance project, as described previously,
[15] for which state health departments abstracted information from medical charts using a standard case form. We were provided with additional reports of hospitalizations from New York City, Wisconsin, and Texas, that were not included in the surveillance project, but had been collected by health departments using the same or similar abstraction forms during the same time period. If height and weight were not reported in data collected for the prospective study, we attempted to contact patients for this information. For a minority of patients for whom we had weight but not height data, we estimated their obesity category by assuming that their height was not less than the 5
th percentile or greater than the 95
th percentile of the population (see
Supplement S1). We included information on deaths reported to CDC from April 27 to July 23, 2009. State health departments collected information from medical charts, medical examiner reports, and death certificates, using a standard data collection form. All hospitalizations and deaths included in our analyses were confirmed to have pandemic H1N1 infection by real-time reverse transcription polymerase chain reaction (rRT-PCR).
[16] We estimated the proportion of the U.S. population in different BMI categories from NHANES 2003–4 and 2005–6. NHANES methodology has been described previously.
[17]For each dataset, we calculated the proportion of individuals in each BMI category by age group (2–19 years and ≥20 years). We stratified our analysis by presence or absence of an ACIP-recognized chronic medical condition, which is likely to confound the relationship between BMI category and hospitalization or death due to 2009 pandemic A(H1N1) influenza. We calculated the ratio of odds (odds ratio, OR) of being in each BMI category, where the normal weight category was the reference category, for hospitalizations and deaths (cases) compared to the U.S. population (cohort). Because we compared cases to the cohort of the U.S. population selected before the time period during which cases could occur (i.e., a case-cohort design), the OR estimates the risk ratio of hospitalization or death by BMI category.
[18] To estimate 95% confidence intervals (95%CI) and p-values of the odds ratios we used the Taylor series linearization method to combine the variance estimate of the log-odds of exposure among cases (hospitalizations or deaths) with a variance of the log-odds of exposure among the population as estimated using NHANES.
[19],
[20] For hospitalized patients, we conducted a sensitivity analysis to consider how robust our analysis was with regard to adjustment of age among ≥20 year olds, by creating a logistic regression model for each BMI category with Normal weight as the reference category, where hospitalization was the dependent variable and age group (20.0–35.4, 35.5–45.9, 46.0–54.9, ≥55.0 years) was the independent variable (see
Supplement S1). We conducted a further sensitivity analysis of our treatment of missing BMI data (see
Supplement S1). We also repeated the analysis for hospitalized patients after changing the comparison group to a cohort of the U.S. population that had been admitted to hospital for any reason in a 12 month period, as estimated by NHANES. The rationale for this additional analysis was to assess whether any observed association between BMI category and hospitalization was due to more frequent hospitalization among obese and morbidly obese individuals. We used SAS 9.2 for all analyses.
The investigation of 2009 pandemic influenza A(H1N1) infections was deemed public health surveillance, not human subjects research and therefore did not require Institutional Review Board review.
This article has been 
=