Improving “may contain” labels: A call to team up and share data

Food production is increasingly a global enterprise, involving international supply chains and multiple facilities. Thus, there is potential for unintended allergen presence (UAP) at multiple points during manufacture. Many food businesses mitigate against this through the use of precautionary allergen labelling (PAL), although this does not replace the need for compliance with Good Manufacturing Practice and risk management protocols such as Hazard Analysis Critical Control Point (HACCP). The use of PAL, however, is not regulated in the vast majority of countries. Their use is voluntary and in practice, very inconsistent.1 Although the Expert Committee noted that five times more people would react to an ED05 exposure than to an ED01 (the amount needed to trigger reactions in 1% of the allergic population), the recommendation was to use ED05 for RfDs, rather than a more conservative cut-off. This was informed by several factors. First, there is less uncertainty over ED05 values than for ED01. The Expert Committee also noted that while more allergic people would react to ED05, using ED05 as the RfD would not be expected to result in an increased rate of severe anaphylaxis events. Current analytical capabilities are generally able to measure ED05 for most priority food allergens, although there are other concerns which remain unaddressed, such as sampling strategies and the effect of food processing on allergen detection.3 In contrast, analysis of allergen levels below ED05 is currently not possible for most priority allergens. Thus, using RfDs based on ED05 facilitates laboratory validation, something which may be desired by regulators and industry to inform risk assessment. Indeed, many food businesses use internal levels that are below ED05 for implementation to ensure compliance, something that is feasible with ED05 but not with ED01. Therefore, using ED01 is less verifiable and more difficult to “police”. It is likely that many food businesses would not feel confident implementing a risk management plan based on ED01; the potential costs of frequent food recalls makes it very likely that they will continue to err on the side of caution and apply PAL even when there is no real risk posed to the consumer. Using more conservative RfDs based on ED01 might therefore paradoxically increase the use of PAL. The datasets informing ED05 consist of thousands of patients with food allergy, but there is always a need to further refine ED05 estimates and validate them in other populations.2 In this respect, the publication by Mortz et al. is important.4 The authors report on a large dataset of 2612 reactive oral food challenges (OFC, 91.5% open) to a variety of food allergens and across different ages, including 20% adults. In comparison to published datasets (the largest of which is maintained by TNO/FARRP and is the underlying dataset for both the Australian VITAL scheme and the recent FAO/WHO Codex discussions),2 some specifics need to be considered. Overall, the ED05 estimates published by Mortz et al. are around three to five times higher than those in the TNO/FARRP dataset (Table 1).5 This discrepancy is even greater for ED01 values, but less pronounced for those allergens with more data available such as egg and peanut. Possible explanations may be different patient populations, subtle differences in regards to indication for OFC, different criteria to define the eliciting dose and differences in how the data are fitted. TNO/FARRP data used a “Stacked Model Averaging” approach which results in a better fit than the interval censored log-normal distribution used by Mortz et al. which tends to result in higher estimates of ED01 and ED05.5 For some foods, such as cow's milk and egg, the initial dose of the challenge was higher in the publication from the Danish group (11 mg egg; 1 mL milk ~20 mg assuming 2% milk; 1 mg or 3 mg for peanut and tree nuts) than used in other studies. This may explain why a number of participants reacted to the first dose, a phenomenon referred to as “left-censoring” which occurred in 2.7% of the datapoints for egg and 8.4% for cow's milk. These differences affect the precision of the curve fit, particularly at the lower end of the dose-distribution curves (i.e., ED01/05). It is also for this reason that the estimates for proportion of allergic individuals who might react to very low doses (0.5/1/5 mg food protein) presented in table 3 in the manuscript by Mortz et al.4 must be interpreted with caution. The fitting of data at these very low protein levels depends on the availability of actual data points at that end of the curve to improve precision: very few (if any) participants reacted to <5 mg of protein, which results in wide confidence intervals at the lower end of the curve. It is for this reason than any attempt to quantify the proportion of individuals reacting to very low levels of exposure must be validated with alternative approaches—for example, using single-dose OFC as has been published for peanut and cow's milk.6, 7 At the same time, such an approach to allergen management—where avoidance advice can be based on a patient's individual threshold—can be transformative.8 Removing anxiety over the likelihood of a reaction to a very low level exposure can improve health-related quality of life measures.6 Indeed, perhaps immunotherapy should be targeted to increasing an individual's reaction threshold, rather than targeting complete desensitisation and “remission” which is a less achievable outcome. An outstanding question is whether data derived from OFC are applicable to “real world” exposures. Administering incremental doses during challenge is different to ingestion of (typically a single dose) during real-world exposures, and could theoretically induce transient desensitisation and over-estimate reaction thresholds.9 However, evidence suggests this is not a concern in most individuals.10 Furthermore, ED01/05 levels of exposure tend to be equivalent to the first dose(s) during OFC, and therefore not impacted by dose titration. A second concern is how food processing, and inclusion of allergen in a complex food “matrix”, impacts allergenicity and allergen absorption. However, most challenge protocols use unprocessed allergen, so this is not of concern because processing usually results in reduced allergenicity (an important exception being peanut, which has been addressed by using roasted rather than raw peanut for OFC). Indeed, data suggests that incorporation of egg or milk into baked matrices has little impact on ED01/05 thresholds.11 The data from Mortz et al.4 are important and reassuring, in that reference doses proposed by the recent FAO/WHO Expert Panel are likely to be protective for the vast majority of food-allergic individuals. To move things forward and provide further reassurance as to discussions now being held by the Codex Committee on Food Labelling (CCFL), it is important that clinicians with appropriate and well-procured OFC datasets (standardized protocols with well-defined amounts of food protein) make their data available for analysis. The establishment of RfDs has a significant advantage in that it uses human data which is directly translatable (often, toxicology assessments are performed in animal models and then have a “safety factor” applied to mitigate for extrapolation to humans). Let's work together to maximise routinely-collected data in our clinics, to help inform allergen risk management for the benefit of our food allergic patients. P.J. Turner reports grants from UK Medical Research Council, NIHR/Imperial BRC and JM Charitable Foundation; personal fees from UK Food Standards Agency, Aimmune Therapeutics, Allergenis, Aquestive Therapeutics and Novartis, outside of the submitted work; is co-lead of the Resuscitation Council UK Working Group on Anaphylaxis, and current Chairperson of the World Allergy Committee Anaphylaxis Committee. Dr. Eiwegger reports grants from CIHR, FWF, Food Allergy and Anaphylaxis Program Sickkids, ALK. He is site PI of company sponsored trials by DBV, Novartis and Stallergenes Greer, EFSA, FARE. Personal fees from Danone/Nutricia/Milupa, ThermoFisher, Aimmune, Stallergenes Greer, ALK, MADX and Non-financial support from Novartis and MADX, all outside the submitted work; He is associate editor in Allergy. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.