Purpose of review Oral immunotherapy (OIT) can have a major positive impact on patients with IgE-mediated food allergies, increasing reaction thresholds and reducing the need for dietary and lifestyle limitations. However, patients experience more frequent allergic reactions during OIT than when following dietary avoidance, and 10-75% of patients on OIT may experience anaphylaxis to treatment doses. Our ability to identify patients at higher risk of more severe or frequent reactions during OIT is limited. We review the current data available and highlight the gaps in knowledge which impede our ability to predict response to treatment, occurrence of dose-related adverse events, and thus acceptance of OIT into wider clinical practice. Recent findings Our ability to predict the risk of severe reactions in food-allergic patients is limited, due to the multitude of allergen and host-related factors which influence this. While OIT is thought to reduce this risk, little is known about the immunomodulatory effect of OIT on these factors, and the resulting risk of allergic events during OIT. Several factors have been associated with reaction severity during OIT and treatment withdrawals, including high allergen-specific IgE levels and certain IgE epitope binding patterns. Other factors proposed include the degree of sensitisation on skin testing, initial reaction threshold, prior reaction severity, age, and concomitant allergic disease including allergic rhinitis and asthma. These have also been associated with more severe events in foodallergic patients not undergoing OIT, and suggest a specific patient phenotype prone to more severe and persistent food allergy, which also impact on poorer outcomes during OIT.Ironically, it is this patient phenotype that arguably has most to gain from OIT. Summary Our understanding of the constellation of factors contributing to reaction threshold, nature, and severity in food allergy is improving, and this helps understand the complexity of OIT safety. Potential predictors of OIT safety are becoming available. However, accurate prediction cannot be done at individual level at present, and significant concerns around OIT safety remain, arguably making this treatment unsuitable for routine practice. Predictors are likely to be study-and population-specific, given the variation in OIT protocols, patient characteristics, and adverse event reporting. Consensus is needed on detailed adverse event reporting, and pooled analysis of multiple series is likely to yield more useful predictors of OIT safety.