Wherein lies the truth? Assessment of agreement between parent proxy and child respondents

Background, The epidemiological study of diseases of childhood presents a special situation in that the principal respondent is almost always a person other than the index. Use of proxy respondents in paediatric epidemiology has been little studied compared with the use of surrogate respondents for adults. Methods. Agreement between responses from children and their parents to identical questions about the children was assessed in a case-control study of childhood melanoma in Queensland, Australia, Weighted kappa and log-linear modelling techniques were used to measure agreement for ordinal data. Results. Highest agreement was found for reports of unchanging physical characteristics such as eye colour (kappa = 0.88), hair colour (kappa = 0.76), and for history of residing on a farm (kappa = 0.84). Moderate agreement was seen for density of facial freckling (kappa = 0.62), propensity to sunburn (kappa = 0.46) and tanning ability (kappa = 0.47). Variables with low levels of agreement between parent proxy and child respondents included density of freckling on the shoulders (kappa = 0.28) and degree of moliness of the skin at age 5 years (kappa = 0.24). Agreement did not vary according to age or sex of the children. Log-linear modelling was performed to determine the type and magnitude of components of agreement. Conclusions. Agreement between responses from children and parent proxies depends largely upon the type of information sought, rather than characteristics of the respondents.