Patient characteristics
Fifty-one patients with hepatic mCRC and TARE were included [median age, 61 years (range: 40–83); n = 35 male, n = 16 female]. Prior to TARE, all patients received systemic therapy, with either stable (n = 9, 17.6%) or progressive disease (n = 42, 82.4). Detailed patient characteristics are shown in Table 1.
Response to TARE
At the 3-month follow-up after TARE, n = 24 (47.1%) of the patients showed treatment response [complete response: n = 2 (3.9%), partial response: n = 11 (21.6%), stable disease: n = 11 (21.6%)]; and n = 27 (52.9%), progressive disease, as presented in Table 2.
Outcome prediction model
Binary logistic regression identified none of the analyzed common clinical parameters such as age, sex, pre-therapeutic laboratory data, primary tumor side, metastatic tumor burden of the liver, and KRAS status (all p-values > 0.05). Hence, there are no common baseline patient characteristics that can be used to predict response to TARE. Therefore, radiomic analysis of the data was performed subsequently.
Two independent radiomic features [Energy, Maximal Correlation Coefficient (MCC)] differentiated well between responders and non-responders (Figure 2). Thereby, “Energy” is a measure of the magnitude of voxel values in an image, while “MCC” is a measure of complexity of the texture. Larger values of Energy and MCC imply a higher tumor heterogeneity, resulting in a higher probability of not responding to TARE.
For predicting treatment response in the unseen test dataset, the radiomic-based model yielded an AUC in ROC of 0.75 (95% CI, 0.48-1) with a sensitivity of 83 % and a specificity of 62 % (Figure 3). The optimal model derived cut-off (best operating point = 0.4, Figure 3) was applied to the radiomics signature of the test dataset to perform risk stratification for disease progression after TARE. The high-risk group (Y ≥ 0.4) revealed shorter overall survival than the low-risk group (Y < 0.4; median 3.4 vs. 6.4 months, p < 0.001; Figure 4). The median overall survival was 4.9 months.