Page 131 - 89Zr-Immuno-PET:Towards a Clinical Tool to Guide Antibody-based Therapy in Cancer

Interobserver reproducibility of 89Zr-immuno-PET
For a group of n VOI, the interobserver variability is given as the median (interquartile range).
Reliability. A reliability parameter was used to assess whether differences in tumor uptake between lesions can be distinguished, despite measurement error due to interobserver variability. A two-way random model with absolute agreement (single measure) was used to obtain the ICC and 95% confidence interval. This means that the 3 observers in our study were considered as a random sample of all possible observers, and the systematic differences between the observers were included in the measurement error as we were interested in absolute agreement between the observers.
Reliability, expressed as ICC, was calculated as:
ICC=σ2lesion / (σ2 lesion + σ2obs + σ2error ), [3]
where σ2obs is the systematic part, and σ2 error is the random part of the measurement
error, while σ2 lesion is the true variance between tumor lesions. ICC calculations 6 were performed in SPSS, version 22.
Statistical analysis
For comparison of interobserver variability between two groups, Wilcoxon matched-pairs signed rank test was used for paired data (e.g. SUVmean on D3 and D6 for the same tumor lesions). For comparison of median CoV between multiple groups, a one-way ANOVA (non-parametric) was performed, using Friedman test with Dunn’s multiple correction to compare median CoV for paired data (SUVmean, SUVmax and SUVpeak for the same tumor lesions). For all statistical tests, a p-value <0.05 was considered statistically significant. Statistical tests were performed in GraphPad Prism, version 6.02.
RESULTS
VOI delineation
In total, 103 VOI were manually delineated by each observer. The number of VOI was not evenly distributed over the patients (Table 1). The range in interobserver variability (SEM for SUVpeak) for all VOI combined was 0 to 2.3 (median 0.4,
129