D1.266 - Phenotypes of IgE-mediated Cow’s Milk Protein Allergy in an Irish Cohort Using Latent Class Analysis

The milk ladder is the first-line management for IgE-mediated Cow’s milk protein allergy (CMPA). Traditional phenotypes of CMPA refer to the degree of baking with which they react to, namely baked milk tolerant or baked milk allergic phenotypes. In Ireland, patients are not categorised into these phenotypes, as all patients regardless of allergic status to baked milk, are commenced on the milk ladder.

Artificial intelligence has been used in the field of allergy for phenotype discovery, especially in asthma and allergic rhinitis, however, this has not been applied to food allergy. Therefore, this study aims to apply unsupervised machine learning to identify phenotypes of IgE-mediated CMPA in an Irish retrospective cohort and compare their outcomes during dietary advancement therapy using the milk ladder.

Patients diagnosed with IgE­-mediated CMPA between 2011 and 2020 at Cork University Hospital, Ireland who were managed using IMAP milk ladder were included. 

The unsupervised learning model utilised was latent class analysis (LCA). Numerical parameters were converted to categorical variables based on whether the value was above or below the 75^th percentile of the median. Models that would identify 2 to 4 classes were explored, assessing each model’s performance using entropy and compared models using the Bayesian Information Criterion (BIC) and Akaike’s Information Criterion (AIC). The clinical outcomes and characteristics of each cluster or class were compared using the chi-square test. 

The 3-class model was chosen as the LCA model for milk allergy phenotypes.

Three distinct phenotypes were identified among the milk allergy cohort (n=171). The first class identified was thermed ‘Older children, highly sensitised, comorbidities, anaphylaxis’ (32.2%, n=55), Class 2 was termed ‘Moderate Comorbidity, Strong family history, more female’ (24.6%, n=42),and Class 3 was termed ‘Low-risk, lowest sensitisation/comorbidity’, with the highest proportion of the total cohort (43.2%, n=74).

Children in Class 3 were significantly less likely to have to carry an AAI for any reason (% (6.8% Class 3) vs 40% (Class 1) vs 16.7% (Class 2), p=0.0001). Those in Class 1 successfully completed the milk ladder at a lower rate, but this was not statistically significant (80% (Class 1) vs 83% (Class 2) vs 85% (Class 3), p-0.74). The three classes of CMPA phenotypes had comparable rates of reported allergic symptoms during treatment and treatment duration. 

Latent class analysis provides a novel categorisation of phenotypes of CMPA which does not rely on the status of allergic reaction of baked milk. Despite the distinction of clinical phenotypes of CMPA in our cohort, there was no significant difference in rates of completing the milk ladder or time taken to complete the ladder. 

This study shows that the treatment outcome of the milk ladder does not differ among CMPA phenotypes. Further external validation international cohort needs to be conducted to reliably inform future clinical guidelines.