Development of a prediction model for a severe reaction in cow's milk challenges

In addition to their conventional purposes, which is for the definitive diagnosis of a food allergy or for the discernment of tolerance to the allergen,1 oral food challenges (OFCs) are performed to determine the threshold dose of food allergens for risk-assessment or minimal avoidance. More severe reactions tend to be provoked in these OFC settings. We have already reported a model for predicting severe allergic reactions provoked in boiled egg challenges.2 In the present study, we developed a model for predicting a severe reaction to a milk OFC. This newmodel was developed via a similar method to that used for the egg challenge.2 An open OFC of milk was performed in accordance with the Japanese Guideline for Food Allergy 20143 throughout the study period. Raw cow's milk was administered in a gradient dose (typically 4 to 5 doses from 0.2-0.5-1-2-5-10-20 ml) every 30e40 min. The challenge was stopped if the patient exhibited an objective allergic reaction corresponding to 5 points of the total score (TS) of Anaphylaxis Scoring Aichi (ASCA).2,4 To quantify the overall severity of the result of OFCs, the TS/Pro2 was applied, which was calculated by dividing the TS by the cumulative protein dose (Pro) of milk (3.3% of whole milk) administered before the appearance of symptoms. We divided patients in the development dataset into two groups (severe cases and non-severe cases) based on the median value of the TS/Pro of the development dataset. The development dataset was obtained from OFCs to cow's milk from April 2012 to May 2013. During this period, 220 OFCs to milk were conducted, and 144 (65.5%) of thosewere positive. The appropriate laboratory data (specific IgE to milk and total IgE) collected within 180 days from OFC were missing in 37 cases, and 10 OFCs performed during the OIT were also excluded. As a result, 97 cases were analyzed. To validate this model in an independent clinical setting (validation dataset), 95 consecutive OFCs to milk (70 positive and 25 negative) with appropriate data and conditions from June 2013 to March 2014 were analyzed in a prospective manner. This study was approved by the institutional review board of Aichi Children's Health and Medical Center. All analyses were performed with the STATA software program (version 12.1 for Mac; STATA Inc., College Station, TX, USA). For all analyses, a 2-sided probability value below 0.05 was considered to indicate statistical significance. The characteristics of the datasets are shown in Supplementary Table 1. Based on the median value of TS/Pro in the development dataset, we defined severe cases as TS/Pro 80 (n 1⁄4 49) and non-