Prevalence and Severity of Food Allergies Among US Adults

This survey study provides nationally representative estimates of the distribution, severity, and factors associated with adult food allergy in the United States.


NORC's Probability-based AmeriSpeak Panel
To obtain our target sample size of 40,000 adults, first a target of 7,210 participants was initially recruited from NORC's web-enabled AmeriSpeak ® panel. This target sample size was selected to be consistent with a prior childhood food allergy prevalence study 5 where power analyses determined that such a sample size would provide .90 power (assuming a significance level of .05) to detect: (1) overall and allergen-specific food allergy prevalence rates from 1% to 9%; and (2) variability in prevalence estimates from 1% to 7% among groups as small as 1% of the overall analytic sample. In the present study, a survey completion rate of 51.2% was observed among AmeriSpeak panelists (7,210 completions among 14,095 sampled units). The weighted cumulative AAPOR response rate for the AmeriSpeak sample was 8.8%. This rate is a function of the 18.3% rate of sampled households successfully recruited into the AmeriSpeak panel, the 93.8% rate of successfully recruited households who were also successfully retained into the panel, and the aforementioned 51.2% completion rate among successfully recruited, and retained AmeriSpeak Panelists who were approached for this survey.
The AmeriSpeak panel is a socio-demographically and geographically-representative sample of US households recruited via a probability-based sample design using NORC's National Frame. This allows for derivation of unbiased estimates representing the full population of interest with known margins of error accounting for the sample design, response rates, and survey weighting.

AmeriSpeak Panel Sampling Frame
In order to provide a nationally representative sample, AmeriSpeak leverages the NORC National Sample Frame, constructed by NORC to cover over 97 percent of U.S. households. The 2010 National Frame used a two-stage probability sample design to select a representative sample of households in the United States. The first stage-the sampling unit-is a National Frame Area (NFA), which is either an entire metropolitan area (made up of one or more counties) or a county (some counties were combined so that each NFA contains a population of at least 10,000). The largest NFAs with a population of at least 1,543,728 (0.5 percent of the 2010 Census U.S. population) were selected with certainty; these areas have a high-population density, and are dominated by tracts with street-style addresses. These areas contain 56 percent of the population within 8 percent of the geographic area of the United States. The remaining areas were stratified into areas where street-style addresses predominate, and the remaining areas, which are less likely to have street -style addresses. The latter stratum ("rural" areas) comprises 81 percent of the geographic area, but only 14 percent of the population. Within the selected NFAs, the second stage sampling unit is a segment, defined either in terms of Census tracts or block groups, containing at least 300 housing units according to the 2010 Census. A stratified probability sample of 1,514 segments was selected with probability proportional to size. For most of the 1,514 segments, the USPS DSF provided over 90 percent coverage of the segment in terms of city-style addresses that are geo-codeable.
For the 123 segments where the DSF provided insufficient coverage, we enhanced the DSF address list with in-person listing. The National Sample Frame contains almost 3 million households, including over 80,000 rural households added through the in-person listing. AmeriSpeak incorporated address-based sampling (ABS) addresses in 2016 from the USPS DSF to assure AmeriSpeak sample representation within all US States. As of October 2016, 0.9% of AmeriSpeak Panel recruited adults were sourced from the ABS (99.1% from the National Frame). Proper weights, such as those employed in the present study, allow the full use of the combined sample.

Sample Selection for AmeriSpeak Panel Recruitment
As summarized above, the 2014-2016 AmeriSpeak Panel sample consists of nationally representative housing units drawn from the 2010 NORC National Sample Frame and less than 1% from address-based sampling. The 2010 NORC National Sample Frame is stratified based on segment (Census tract or Census block group) characteristics such as age and race/ethnicity composition of the segment, and then, a stratified simple random sample of housing units is selected.
Specifically, based on Census tract-level data, segments were classified as having a higher concentration of 18-24 year old adults or not, and a higher concentration of Hispanics, non-Hispanic African Americans, and other. Based on these strata definitions, 6 strata (2 based on age times 3 based on race/ethnicity) were used to oversample housing units in segments higher in young adults and/or Hispanics and non-Hispanic African-Americans. This is referred to as the initial sample or first stage of panel recruitment.
In the second stage of panel recruitment, initially sampled but nonresponding housing units are subsampled for a nonresponse follow-up (NRFU). At this stage, consumer vendor data are matched to housing units, and housing units that are flagged (based on consumer vendor data) as having a young adult or minority (Hispanic and non-Hispanic African American) are oversampled for the nonresponse follow-up. Overall, approximately one in five initially nonresponding housing units are subsampled for NRFU. However, as mentioned previously, selection of housing units for NRFU is a stratified simple random sample based on consumer vendor data. Due to NRFU, these initially nonresponding housing units have a much higher selection probability compared to the housing units that were recruited during the first stage of panel recruitment. Note that a small fraction of initially nonresponding housing units are not eligible for NRFU due to these housing units being classified as "hard refusals" or having an appointment for a call back from NORC.
In summary, there are two reasons why the sampling design for AmeriSpeak Panel recruitment deviates from equal probability of selection method sampling: (a) oversampling of housing units in segments with a higher concentration of young adults and minorities results in the sample selection probabilities being higher for housing units in these segments; and (b) the nonresponse follow-up effort results in initially nonresponding housing units having a much higher selection probability. Furthermore, oversampling associated with NRFU results in higher selection probabilities for initially nonresponding housing units that are flagged (based on consumer vendor data) as having a young adult or minority.
Recruitment is a two-stage process: initial recruitment using less expensive methods and then non-response follow-up using personal interviewers. For the initial recruitment, sample units are invited to join AmeriSpeak online by visiting the panel website AmeriSpeak.org or by telephone (in-bound/outbound supported). English and Spanish language are supported for both online and telephone recruitment. Study invitations are communicated via an over-sized pre-notification postcard, a USPS recruitment package in a 9"x12" envelope (containing a cover letter, a summary of the privacy policy, FAQs, and a study brochure), two follow-up post cards, and also follow-up by NORC's telephone research center for matched sample units.
The second-stage non-response follow-up targets a stratified random sub-sample of the non-responders from the initial recruitment. Stratification is based on consumer vendor data and stratification variables from the initial recruitment stage in order to increase sample representation of young adults, non-Hispanic African Americans, and Hispanics. Units participation. NORC field interviewers administer the recruitment survey in-person using CAPI or else encourage the respondents to register at AmeriSpeak.org or call the toll-free AmeriSpeak telephone number to register.

Impact of Non-Response Follow-up on AmeriSpeak Panel Representativeness
The non-response follow-up improves the representativeness of the AmeriSpeak sample with respect to certain demographic segments, including but not limited to rural and/or lower income households, cell-phone only households, persons age 18 to 34, African Americans, Hispanics, and persons without a high school degree or only have a high school degree (no college). Compared to panelists recruited in the initial stage, panelists recruited via the non-response follow-up campaign are more politically conservative, are less knowledgeable about science, report less interest in current events and topics in the news (such as climate change), and are less likely to read a print newspaper.

AmeriSpeak Panel Implementation of Mixed-Mode Data Collection
Panelists may participate in 2 to 3 AmeriSpeak Panel studies per month via online (computer, tablet, or smartphones) or by CATI phone. NORC maintains strict rules to limit respondent burden and reduce the risk of panel fatigue. CATI phone mode respondents represent a population currently under-represented in web panels that exclude non-internet households or "net averse" persons. NORC's telephone interviewers administer the phone mode of survey questionnaires using a data collection system supporting both the CATI phone and web modes of data collection, providing an integrated sample management and data collection platform. For panelists using smartphones for web-mode AmeriSpeak surveys, © 2019 Gupta RS et al. JAMA Network Open. the NORC survey system renders an optimized presentation of the survey questions for these mobile users. For general population client studies, approximately 20% of the completed interviews are completed by the telephone mode.

Survey Sampling International's Non-Probability-based Sample
To augment the probability-based sample recruited from NORC's AmeriSpeak Panel, participants were also recruited from an online non-probability-based sample of US households administered by Survey Sampling International (SSI). A response rate of 5.5% of invited adults was observed. Due to the opt-in nature of the non-probability-based SSI sample, and corresponding lack of a national sampling frame an AAPOR response rate cannot be calculated. Non-probability optin samples do not generally allow for derivation of unbiased estimates due to the lack of randomness in the opt-in sample and the high likelihood that the opt-in panel is skewed toward a non-representative subset of individuals, thus leading to an increased likelihood of biased estimates from the opt-in sample. Nevertheless, SSI employs a variety of approaches to limit bias, including use of invitations of all types to recruit participants with a diversity of motivations. These include e-mail invitations, telephone alerts to mobile devices, banners and messaging on SSI panel community sites. The messages themselves are also varied, including invitations to give your opinion, win a prize, earn cash or prizes or let your voice be heard. A diversity of motivation contributes to high-quality sample. Furthermore, to avoid self-selection bias, specific study details were not included in the invitation. Rather, participants were invited to "take a survey". The details are disclosed later, when the participant initiates the survey.

Complex Survey Weighting Details
Responses from the AmeriSpeak ® panel were used to identify and correct for sampling and non-response biases via calculation and application of probability sampling weights. Following the nonresponse adjustment, iterative proportional fitting (also known as "raking") was conducted in order to rake survey weights to national US population totals associated with age, sex, education, race/ethnicity, and Census Division.
A base weight of one was assigned to each SSI non-probability sample complete, and then iterative proportional fitting was used to rake the non-probability sample weights to national US population totals associated with age, sex, education, race/ethnicity, "early adopter" characteristics, Census Division, and the bias-corrected overall food allergy prevalence estimates calculated using the nationally-representative AmeriSpeak sample. Finally, the probability and non-probability samples are combined through derivation of an optimal composition factor that minimizes mean square error associated with FA prevalence estimates.

Bureau Demographics
The below table demonstrates the high degree to which the 7210 AmeriSpeak respondents in the present study are demographically representative of the general US population. Specifically, the figure compares the demographic distributions of AmeriSpeak respondents before and after our weighting approach was implemented, and then compares the weighted distributions to national demographics reported in the February 2016 Current Population Survey.