Long-term reliability in reporting of childhood pets by adults interviewed twice, 9 years apart. Results from the European Community Respiratory Health Survey I and II.

UNLABELLED: Investigation of long-term effects of childhood pet exposure is usually based on retrospective information provided by adults, while there is little knowledge about the reliability in adult reporting of childhood events. We analyzed 8287 adults interviewed about childhood pets twice, on average nine years apart, in the European Community Respiratory Health Survey. Agreement between the surveys in reporting of childhood cats, dogs and birds were investigated with kappa statistics, and potential effects of disease status on agreement were analyzed with kappa statistics and multiple logistic regressions. Cats, dogs and birds in childhood were reported by 44, 41 and 38%, respectively. Cohen's kappa for agreement in adult reporting of childhood pets was 0.714 (95% CI=0.698-0.729) for cat, 0.709 (0.691-0.722) for dog, and 0.606 (0.591-0.626) for bird. Thus, agreement was significantly higher for reporting of cat and dog than for bird. Adult wheeze, asthma or atopy did not influence agreement. Neither did adult cat sensitization influence agreement in adult reporting of childhood cat. Childhood factors such as moving house <5 years, or growing up as a single child, in a large family or in a rural area, were associated with poorer agreement, while adult factors were unrelated to agreement. PRACTICAL IMPLICATIONS: Long-term reliability in adult reporting of childhood pets was substantial, and not influenced by disease status. Thus, collection of information about childhood pets from adults appears to be reliable for the purpose of studying adult allergic disease. Future studies should consider that the reliability was higher for a more important childhood event and influenced by childhood rather than adult characteristics. Imperfect reliability contributed to underestimation of the effects of pets on adult allergy; i.e. with a kappa of 0.71, a true odds ratio (OR) of 0.80 would be attenuated to 0.86. Future studies should account for non-differential misclassification error.