Comparison of axillary, tympanic and rectal body temperatures using a covariate-adjusted receiver operating characteristic approach

ABSTRACT

Accurate temperature measurement is crucial in pediatric population. Before diagnostic tests are implemented in practice, it is suggested that their accuracy or ability to discriminate to be studied. The accuracy of a diagnostic test can be summarized in a Receiver Operating Characteristic [ROC] curve. This study was carried out to compare the accuracy of tympanic and axillary methods with rectal measurement in children less than 6 years old. A total of 220 pair of ears, axillaries, and rectal sites were used to determine the body temperature in patients aged between 3 months and 6 years, who referred to Emergency Department of Ali Asghar Hospital affiliated to Bushehr University of Medical Sciences. Rectal temperature [RT] was considered as gold standard. Fever was defined as RT >/= 38°C. RT, axillary, right and left tympanic temperature were measured. Measure agreement was assessed by covariate-adjusted ROC regression. By comparing the area under the curves in Hanely method and the results from ROC regression analysis, we found out a significant agreement among the three measuring techniques and none of them was more accurate than the others. None of these techniques [axillary, right and left tympanic] was more accurate than the others and it is better to use a technique that is more convenient, painless, and safer than rectal temperature. We also propose using a modified parametric distribution-free ROC estimator which is conceptually easy and is simple to implement with the existing softwares for comparing the accuracy of medical tests