Assessing significant changes in FDG-PET data from oncology clinical trials

Objectives Serial FDG-PET scans are useful for acquiring evidence of early pharmacodynamic effect of novel anti-cancer drugs. Published test-retest results suggest a log-normal distribution for delta-SUVmax indicating that a fixed cutoff, e.g. 25%, for %change may be adequate to determine significant changes in lesion SUVmax. However, these studies are based on short inter-scan intervals (median 2 days), but for therapy response assessment, such intervals are unrealistic. For longer, on-treatment scan intervals, we develop an approach to calculate more ‘rational’ patient-specific cutoff values for significant changes which considers the statistical nature of the noise and bias effects.

Methods To assess variability in a more realistic context, we analyzed baseline and follow-up FDG-PET scans with a median scan interval of 21 days from 50 advanced stage cancer patients enrolled in a PhI trial. This trial was discontinued and we treat the data as a test-retest study. A simulation-based tool is developed which takes as input baseline lesion SUVmax, the variance of the changes, the noise model and the desired Type I error rate, and outputs lesion and patient based cutoff values. Bias corrections are also included to account for variations in tracer uptake time.

Results We found that changes in SUVmax follow an approximately zero-mean Gaussian distribution with constant variance across levels of the baseline measurements. Our finding contrasts those based on shorter scan intervals. In our setting, because of constant variance, the coefficient of variation is a decreasing function of the magnitude of SUVmax.

Conclusions For therapy monitoring, baseline dependent cut-offs are more appropriate to determine a significant change in SUVmax. For lower baseline values, these cutoff values are notably asymmetric with relatively large changes (e.g. >50%) required to determine significance. Applying this algorithm to FDG-PET data from a new Ph2 trial resulted in a change in metabolic response classification for 22% of patients