OBJECTIVES: Little attention has been paid to determining whether different language versions of the same patient-reported outcome (PRO) measure are equivalent and if their use introduces bias into multinational studies in the form of differential item functioning (DIF). The study was designed to see whether it is possible to overcome such DIF in the development of PRO measures. METHODS: Data collected with a draft PRO - the 12-item Quality of Life Index for Children with Atopic Dermatitis (QoLICAD) - from France, Germany, Italy, the Netherlands and USA were analysed. Rasch analyses were applied to the data from each country to determine item fit and DIF by culture. Where some but not all items display DIF, it is possible to treat them as a different item in each country. This procedure is referred to as ‘splitting' items. The analysis was then re-run on the new item sets. RESULTS: Data were available on 691 (52.4%M; mean age 4.4 years) children in the 5 countries. Rasch analyses applied to the data indicated that, following the removal of misfitting items, there was considerable DIF by country. A European and separate US QoLICAD were derived. Four items were deleted and four were split to create a 20-item European scale that fitted the Rasch model (Chi2 p<.002 after bonferroni correction). Five items were deleted before the US scale fitted the Rasch model (Chi2 p<.004 after bonferonni correction). The European and US data were linked by six common items allowing pooling of data from an international study. CONCLUSIONS: Language adaptations should be conceptually equivalent to the original. However, DIF by culture may remain. If data are to be pooled such DIF needs to be assessed and accounted for. The method illustrated is a suitable way of achieving this but requires further testing in a clinical trial setting.