Statistical analysis

EFHL data was managed with Stata 12.1. Data analyses were undertaken using R [13]. Chi-square tests were used to compare associations between parental-reported allergy and potential risk factors. The presentation rate to ED of children with allergic disorders was calculated based on the whole EFHL cohort (n = 3404, Table ​Table1).1). A p-value of < 0.05 was considered statistically significant. Prevalence rates and 95% confidence intervals were computed using the Mid-p exact test for person-time rates. Person-years (PYs) for prevalence rates of parental-report allergy were calculated by adding the number of years each child had contributed to the study. Person-years for the prevalence of presentation to ED with allergy was calculated for each child based on the last date of ED data available (31 December 2013) minus the child’s date of birth, and summed across the cohort. Logistic regression analyses were conducted to identify variables independently associated with parental report of allergy as well as ED diagnosis of allergy (including and excluding ‘rash’). For the parental report model, variables significant at a p < 0.10 level in the univariate analysis and with complete data on at least 90% of children were considered in a forward stepwise conditional regression model with entry and exit criteria of p < 0.10. The same regression technique was used for models determining variables associated with ED presentation for allergy (including and excluding “rash” lead to two separate models). However, because some of the key univariate predictors had > 10% missing data for this model, various models were built on the full sample as well as for subsets with complete data on key variables found to be significant in univariate analysis. Interactions between variables were assessed; significant interactions (p < .05) were accounted for in all models. In determining the best model, the number of records, the strength of the associations, the persistence of covariates across models, and the Negelkerke’s R square value were considered.