%0 Journal Article %A Martin Lodge %A Richard Wahl %T Characterizing the repeatability of oncology PET standardized uptake values %D 2013 %J Journal of Nuclear Medicine %P 335-335 %V 54 %N supplement 2 %X 335 Objectives To apply essential refinements to the Bland-Altman method for measuring the repeatability of FDG PET standardized uptake values (SUVs) so as to allow for the case where repeatability varies as a function of the underlying SUV (de Langen 2012). Methods The method was applied to previously published test-retest tumor SUVmax data (n=152, Velasquez 2009, Nahmias 2008, Hoekstra 2002, Minn 1995). In order to reflect tumor response monitoring protocols involving serial PET studies we define the relative difference D with respect to a single baseline measurement (SUV1), as opposed to the mean of two test-retest measurements (common statistical practice): D = 100 × d / SUV1 where d = SUV2 - SUV1. Two clinically-relevant situations were separately considered: SUV1>SUV2 (possible tumor response) and SUV1<SUV2 (possible tumor progression). For each situation, a regression technique was used to estimate the SD of D as a function of SUV1 under the assumption of normally distributed data (half-normal method). From these data, 95% limits of repeatability were determined as a function of the baseline SUV. Results Conventional Bland-Altman analysis was not appropriate because the magnitude of the difference data was a function of the underlying SUV. This was the case when differences were calculated on the original scale (d), after log transformation and also on the relative scale (D). D was not normally distributed over the full SUV range, although within smaller SUV sub-ranges (SUV1<5, 5<SUV1<7.5, 7.5<SUV1<10, etc) D was normally distributed (Shapiro-Wilk p >> 0.05). The half-normal method produced 95% limits of repeatability that were a function of SUV1 and differed for the cases of tumor response and progression, e.g. (+42%, -29%), (+27%, -29%) and (+19%, -26%) at baseline SUVs of 5, 10 and 15. Conclusions The limits of repeatability for SUVmax can be conveniently expressed in relative terms, although these limits are a function of the baseline SUV. When SUV change is defined with respect to a single baseline measurement, the limits of repeatability are different for suspected tumor response and progression. %U