An algorithm for diagnosing IgE-mediated food allergy in study participants who do not undergo food challenge

Background: Food allergy diagnosis in clinical studies can be challenging. Oral food challenges (OFC) are time-consuming, carry some risk and may, therefore, not be ac-ceptable


| INTRODUC TI ON
Food allergy affects approximately 5% of young children and has a significant quality of life and economic impact. [1][2][3] The gold standard for diagnosing food allergy in clinical studies is a double-blind, placebo-controlled oral food challenge (OFC). 4 However, open (nonblinded) OFC are widely used for detection of food allergy in young children, where placebo-controlled challenges may be impractical and the reporting of subjective symptoms less likely. 5 Recent guidance from the European iFAAM (Integrated approaches to Food Allergen and Allergy risk Management) project provides a framework for when to undertake OFC in clinical studies where food allergy is an end-point. 6 This guidance suggests that if a food has been eaten without symptoms of allergy, or if a food has not been eaten, but there is no evidence of allergic sensitization, then OFC is not required. However, the guidance does not specify more precisely the frequency, form and quantity of food ingestion that is sufficient to exclude allergy, or which symptoms related to ingestion indicate an allergic reaction. Importantly, the guidance does not address detection of food allergy where OFC is not completed. Hence, there is a need for a more detailed and complete algorithm to guide food allergy diagnosis within clinical studies, especially those where uptake of OFC by study participants is low.

The Barrier Enhancement for Eczema Prevention (BEEP)
study is a pragmatic, multi-centre, randomized-controlled trial of daily emollient use for the first year of life for primary prevention of eczema, in infants with a family history of atopic disease (ISRCTN21528841). Evaluation of IgE-mediated food allergy was added to the initial study protocol after recruitment had commenced, but before any 2-year follow-up visits were conducted, once separate funding was secured. 7 It was anticipated that a significant number of participants might decline to attend for OFC in BEEP, as the food allergy testing was presented to parents at the 2-year visit as an optional sub-study in a pragmatic, low-contact trial with eczema prevention as its initial focus. Furthermore, the BEEP trial was a multi-centre trial run across a wide geographical area and study OFCs were limited to two UK centres, Sheffield and London (See map of geographical distribution of recruited individuals in Supinfo 2). In the EuroPrevall UK birth cohort study, based on a similar-aged population, over 30% of parents offered OFC to cow's milk declined attendance. 8 We, therefore, developed an algorithm for use in the BEEP study which aims to reliably diagnose IgE-mediated food allergy. We then validated the algorithm in BEEP study participants who had undergone OFC and separately in participants from the control group (standard introduction group) of another intervention trial, Enquiring About Tolerance (EAT; ISRCTN14254740), who had undergone OFC and had a similar set of information available about allergenic food exposure, reactions and sensitization status.

| ME THODS
We evaluated food allergy outcomes in the BEEP study, which presented an opportunity to develop a new approach to food allergy diagnosis in clinical research studies with expected low uptake of OFC. We then validated our findings in BEEP study participants who had undergone OFC and in participants from the control group of the EAT study who had undergone OFC.

| BEEP study design
The BEEP study is a pragmatic, parallel-group, multi-centre, assessor-blind, randomized-controlled trial, details of which have been described elsewhere. 7 In brief, the BEEP study randomized 1395 participants between November 2014 and November 2016, recruited from 16 study centres across England. Participants were born at ≥37-week gestation and had a first-degree relative with parent-reported eczema, asthma or allergic rhinitis diagnosis. They were randomized within 3 weeks of birth to apply daily emollient for 12 months plus skin care advice, or skin care advice only. Written informed consent was obtained at screening, prior to randomization, either during pregnancy or within 21 days of birth. Separate written consent was obtained at the 2-year visit for skin prick testing (SPT) and additional consent was obtained if OFC was required.

| BEEP study food allergy evaluation
Formal evaluation for the point prevalence of IgE-mediated food allergy, using a combination of food allergy history, SPT and OFC, in accordance with iFAAM guidance, was conducted at the primary outcome assessment visit at age 2 years. All BEEP participants were offered SPT at these visits, usually conducted in the home unless parents preferred a clinic visit. Peanut extract (Inmunotek), fresh whole cow's milk, and fresh raw hen's egg white were tested, with positive (1% histamine) and negative (0. atopic dermatitis, diagnosis, food allergy, oral food challenge, paediatrics peanut flour (Golden peanut company) were used for OFC. 9 All study staff remained blinded to treatment allocation. The presence of a clinical reaction during OFC was determined using modified PRACTALL criteria (See Supinfo 1). 4

| Elaboration of iFAAM guidance
iFAAM guidance suggests OFC is not indicated in clinical trials if (A) the food is eaten without out symptoms or (B) the food has not been eaten but the participant is not sensitized. OFC is indicated if (C) the food is not eaten but the participant is sensitized or (D) the food is eaten and a reaction has occurred. 6 We first elaborated on these criteria in order to more precisely define "sensitized", "reaction", "food" and "eating without reaction".
For the definition of "sensitized", we considered a participant to be "sensitized" when the skin prick test was any response (>0 mm), in order to maximize the sensitivity of our procedure. 10 Though correlation between skin test response size and OFC outcome is dependent on the population studied, in general, SPT weal size correlates with probability of reacting at OFC, and in 2-year olds, a 7 mm response is considered strongly predictive for milk, egg or peanut allergy. Hence, we categorized ≥7 mm as "strongly sensitized". 11 Values between 1 and 6 mm were considered "intermediate sensitization". For those without SPT, we considered SpIgE to be negative if it had been done and <0.35 kUa/L.
For the definition of "reaction", we considered any parent-reported reactions to a food within 2 hours of exposure to be an "immediate" reaction.
For the definition of "food", we used the most allergenic form of food consumed in the local population. We considered milk and egg allergy as allergy to any form of milk or egg, even if other forms such as baked or processed milk/egg were tolerated. We considered peanut ingestion in any form to be relevant. We used fresh milk, raw egg white and commercial peanut extract for SPT.

F I G U R E 1 Selection of study participants for invitation to oral food challenge
For the definition of "eating without reaction", we recorded parent-reported ingestion. We adapted food frequency questionnaires from the EAT study, to identify frequent and recent ingestion of each food allergen. 12 "Recent" ingestion was the food being ingested within 3 months of the 2-year interview for milk and egg; within 1 month for peanut. The shorter duration for peanut is due to the possibility of new-onset peanut allergy in the second year. "Frequent consumption" was defined as three or more separate reported ingestions of two or more grams of relevant food protein, at any age.

| Food allergy diagnosis algorithm
We expected some participants would decline to attend OFC, due to an established diagnosis or unwillingness to travel to an OFC centre.
To establish accurate food allergy status for these participants we convened an expert panel of experienced paediatric allergists (MK, NJ, MRP, RJB) to determine food allergy status whilst remaining blinded to treatment allocation. We developed the algorithm using an iterative process of discussion and consensus-building, involving the panel and the wider trial management group and trial steering committee. We did not use a formal Delphi or nominal group process.
We referred to relevant literature and recent cohort studies or food allergy prevention trials in developing the algorithm and modified some data collection materials and processes from the BASELINE cohort study and EAT prevention trial. 12,13 The panel used all available information from the study procedures to guide their decisionmaking (Table 1). Through repeated revisiting of all cases of possible food allergy within BEEP, we developed a consensus-based approach to diagnosing IgE-mediated food allergy to milk, egg or peanut in a setting where participants did not undergo OFC.

| Validation of food allergy diagnosis algorithm
Having developed the algorithm using information from BEEP study participants who did not have an OFC, we then compared algorithmderived expert panel decision-making with OFC outcome in (i). BEEP study participants who underwent OFC; and (ii). EAT study, standard introduction group, participants who underwent OFC. Validation was conducted with panel blinding to food challenge outcomes and was conducted separately for each dataset.
EAT is a randomized trial of 1303 exclusively breastfed 3-month old infants that compared introduction of six allergenic foods from age 3 months with advice to continue exclusive breastfeeding to age 6 months. 12 The EAT data set (ITN900AD) is available through

| Algorithm for diagnosis of food allergy where OFC cannot be completed
The final algorithm is shown in Figure 2. For classification of overall food allergy status, the food with the highest hierarchical determined food allergy outcome is used; that is, if a participant has OFC confirmed egg allergy but has "probable" peanut allergy they are classed as having OFC confirmed food allergy.
We include a demographic table of those children who did not undergo oral food challenge and who had their food allergy status classified through this algorithm in the Supinfo 1.

| Algorithm approach for participants who are sensitized with immediate-type reaction
Reported immediate-type reaction and strong sensitization on SPT (≥7 mm) is "probable food allergy". If SPT data are missing, high-level SpIgE results in the same classification. For immediate-type reactions without strong sensitization, reaction symptoms are considered first.
If the reaction is reproducible (occurred twice or more) and typical of an IgE-mediated reaction (eg acute urticaria, angioedema, vomiting, cough/wheeze settling within hours) participants have "probable food allergy". If reaction history is less clear, without strong sensitization, the precise quantity and frequency of allergen previously tolerated is considered. If the participant has evidence of subsequent sufficient exposure without reaction they are labelled "no food allergy", if there is no evidence of sufficient exposure they are labelled "unclear". TA B L E 2 Examples of case classification by algorithm-derived panel consensus partial tolerance is again considered-there is "no food allergy", if partial tolerance criteria are met, and "unclear" food allergy status if not.

| Algorithm approach for participants who are sensitized with no immediate reaction
Examples of panel decision-making using anonymized BEEP study participants are shown in Table 2.

| Classification of participants with "unclear" food allergy status
Participants with "unclear" food allergy status cannot be reliably classified by the panel, but we undertook a sensitivity analysis where "unclear" participants were classified by the panel. They were categorized as "unclear-food allergy possible" where there was a history of reaction, a history of doctor-diagnosed food allergy to another food, or other features to suggest increased probability of food allergy compared with the rest of the study population. "Unclear-food allergy unlikely" status was given to participants where there was no information to suggest any increased risk of food allergy over the rest of the study population, including participants with significant missing data.

| Comparison of algorithm-derived panel decisions with OFC outcomes from BEEP and EAT studies
In 31/69 (45%) BEEP and 44/55 (80%) EAT study control group participants who had an OFC the panel were able to classify participants as "probable food allergy" or "probable no food allergy". For these participants, algorithm-derived panel decisions showed high sensitivity 94% (95%CI 68, 100) in BEEP, 90% (95%CI 72, 97) in EAT; and moderate specificity 67% (95%CI 39, 87) in BEEP, 67% (95%CI 39, 87) in EAT. Other participants could not be confidently classified due to missing information about ingestion or reaction history or sensitization status. The available information was considered, and they were classified as "unclear-food allergy possible" and "unclearfood allergy unlikely" based on available information. Sensitivity and specificity were broadly similar when participants with "unclear" food allergy status were included in the analysis (Table 3). Note: Panel diagnosis of food allergy was established by using the algorithm in Figure 2 by a panel of four allergists blind to result of OFC. Algorithm inputs were in brief; SPT results, ingestion history and reaction history. EAT and BEEP differed by (i). Reaction history, asked in a similar way but EAT did not determine the time course (ii). Food ingestion history was quite similar but somewhat more detail in the EAT questionnaires, (iii). Food used for OFC, EAT used milk and egg powder for DBPCFC. Numbers denote total individuals, not total number of OFCs, some children had more than one OFC. Abbreviations: LR, likelihood ratio; NPV, negative predictive value; PPV, positive predicative value.

| D ISCUSS I ON
for pragmatic trials or those where food allergy is not the primary outcome measure then surrogates for OFC may be required.

ACK N OWLED G EM ENTS
We are grateful to the patient representatives from the BEEP pilot study and main BEEP trial who contributed formal feedback on the planned evaluation of food allergy outcomes within the BEEP trial. We are also grateful to all participants and team members of the BEEP trial. The EAT database is available publicly and we are grateful to the participants, staff and funders of the EAT study for this resource. The BEEP trial was supported by the NIHR HTA programme, and food allergy assessments were supported by a grant from Goldman Sachs Gives. MK is supported by a NIHR Transitional Research Fellowship. The views expressed are those of the authors and not necessarily those of the NHS, NIHR or the Department of Health and Social Care

CO N FLI C T S O F I NTE R E S T
HW is chief investigator of the BEEP study-an independent study that was funded by the NIHR Health Technology Assessment Programme.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.