Caren van Roekel

195 Dose-effect relationships of holmium-166 radioembolization in colorectal cancer FIGURE 3. Relationship between mean tumor-absorbed dose per patient and metabolic response to treatment at three-months follow-up. The bullets show the mean tumor- absorbed dose per patient. Black vertical lines are the 95%CIs of the mean doses per response category, with the white dot in the middle indicating the mean tumor- absorbed dose per response category. This figure is based on the unadjusted linear mixed-effects regression model as described in Table 3. Based on ROC analysis, the ability of tumor-absorbed dose to discriminate between patients with and without metabolic response was 0.671 (95%CI: 0.54- 0.80) and the ability of mean tumor-absorbed dose per patient to differentiate between responders and non-responders was 0.698 (95%CI: 0.45-0.95) (Figure 4a,b). At a mean tumor-absorbed dose threshold with 100% sensitivity (95%CI: 48-100%) for CR/PR at a patient level (90 Gy), specificity was 38% (95%CI: 21- 56%). At a tumor-level, without accounting for clustered data, sensitivity was 100% using a tumor-absorbed dose threshold of 80 Gy (95%CI: 74-100%) and specificity was 41% (95%CI: 31-51%). Agreement between PERCIST and RECIST was minimal, with κ=0.345 (95%CI 0.14-0.55). Anatomic response was lower than metabolic response in 15 cases (40.5%) and higher in seven cases (18.9%). 7