Utility of clinical and MR imaging parameters for prediction and monitoring of response to capecitabine and temozolomide (CAPTEM) therapy in patients with liver metastases of neuroendocrine tumors

This retrospective study received approval from the local research ethics committee with decision Number 23-0183 and the requirement for written informed patient consent was waived. We consecutively enrolled patients with histologically confirmed, resected or advanced NETs with liver metastases, all of whom received CAPTEM therapy and underwent pretherapeutic MRI at our department. Furthermore, in the sub-analysis focused on therapy monitoring, we incorporated all individuals from this cohort who underwent subsequent MRI examinations (Figure 1). The timeframe for therapy initiation ranged from April 2013 to June 2022. The decision to commence CAPTEM therapy was reached through consensus in an interdisciplinary tumor conference certified for NETs (ENETS Center of Excellence) for each patient.

FIGURE 1.

All patients were positioned supine in a 1.5 T MR system (Siemens Healthcare, Erlangen, Germany). For signal reception a phased-array coil was utilized. Images were acquired in accordance with our standard liver imaging protocol. The following sequences were employed for evaluation:

A single shot T2-weighted sequence (HASTE).

All sequences utilized parallel imaging with an acceleration factor of 2. ADC maps were computed from the acquired DWI-MR images, incorporating all b-values.

FIGURE 2.

Two board-certified radiologists, blinded to the patients’ clinical and follow-up data, reviewed all MRI data in consensus. They randomly identified, on the pretherapeutic MRI, two hepatic metastases per patient that were larger than 1 cm in size, along with the primary tumor if it hadn’t been previously resected. Inclusion criteria for metastases encompassed a homogeneous appearance and absence of artifacts within the lesion across all sequences. The image review took place in two separate sessions, both achieving consensus: 1) pretherapeutic MRI, and for the sub-analysis 2) post-therapeutic MRI, with a three-week interval between each session.

For quantitative analysis, the size of liver metastases and NETs were measured on the hepatobiliary and arterial phases, respectively. ADCmean and ADCmin values of the tumorous lesions were calculated by manually placing circular regions-of-interest (ROIs) on the slice with the largest tumor extent on DWI, excluding structures near the rim to avoid partial volume effects. Signal intensity (SI) values on non-contrast T1-weighted and T2-weighted images were recorded by outlining ROIs of the lesions as large as possible. Percentage of arterial enhancement was visually assessed by the two radiologists in consensus. Additionally, ADC mean and ADC min values, as well as T2-weighted and T1-weighted SI values of the normal liver, pancreas, and spleen, were measured by placing circular ROIs in tumor-free tissue areas. Additionally, SI of the normal liver was measured on the hepatobiliary phase. Tumor-to-organ ratios, including tumor-to-spleen (T/S) ratio and tumor-to-liver (T/L) ratio of SI and ADC, were calculated.

Clinical and surgical records were compiled by a third radiologist. Histopathological confirmed diagnoses of NET, along with their respective Ki-67 indices, were obtained for each patient. Tumor grading adhered to the 2017 WHO Tumor Classification Guideline (G1: Ki-67 Index < 3%, G2: Ki-67 Index 3–20%, and G3 neuroendocrine tumor/neuroendocrine cancer [NET/NEC]: Ki-67 Index > 20%). Given that the primary tumor was resected in 31 out of 44 patients, rendering RECIST 1.1. assessment of treatment response heterogeneous, evaluation of treatment response was conducted through PFS. This was measured in months from the initiation of CAPTEM until progression, as determined by the local interdisciplinary tumor board’s comprehensive assessment of all performed imaging studies (CT, PET/CT, MRI). Responders were defined by PFS ≥ 6 months, while non-responders (NR) were defined by PFS < 6 months, respectively.

FIGURE 3.

Continuous data were summarized by median with interquartile range (IQR) and categorical data by numbers and percentages. Differences between baseline and follow-up parameters were assessed by Wilcoxon signed-rank test for paired samples. Differences of baseline characteristics and parameter changes until follow-up between non-responder and responder were investigated by Wilcoxon rank-sum test for unpaired samples or Fisher’s exact test. The area under the receiver operating characteristic (ROC) curve (AUC) was estimated according to logistic regression models predicting non-responder by selected imaging and clinical parameters. Two AUC values were compared by chi²-test. Sensitivity, specificity, and the Youden-Index were calculated for median-dichotomized parameters. Overall survival (OS) and PFS curves with median survival times were calculated by Kaplan-Meier analysis and compared by log rank-test between individuals separated by the median for selected parameters. Individuals were censored in case of death, progression or end of study. A p-value < 0.05 was considered to indicate statistical significance. All analyses were conducted with Stata 16.1 (Stata Corporation, College Station, TX, U.S.A.).

A total of 44 patients, comprising 86 neuroendocrine liver metastases (NELM) and 14 primary pancreatic NETs were included for the evaluation of prognostic factors for PFS. A subset of 33 patients, with corresponding 66 NELM and 12 pNETs, was identified for the sub-analysis of therapy monitoring. Baseline MRI scans were obtained 19d (IQR 1; 61) prior to CAPTEM initiation, and the time interval between baseline MRI and follow-up MRI was 130 days (IQR 113; 161). Most patients were male (75%), had G2 tumors (76%), and the primary tumor originated in the pancreas (84%). Detailed patient characteristics are presented in Table 1.

In the baseline cohort, the overall median PFS was 5.7 months (IQR 3.6; 15.0), and median OS was 25.0 months (interquartile range [IQR] 16.3; 45.3). Responder in the baseline group tended to have a slightly longer median OS 35.0 m (IQR 19.4; 53.4) compared to non-responders, with a median OS 21.4 month (IQR 15.0; 38.3). According to RECIST 1.1,21 patients were rated as stable disease (SD), 3 patients were rated as partial response, and 9 patients were graded as progressive disease.

When comparing baseline and follow-up parameters, no differences were observed, except for arterial vascularization of NELM, which was significantly lower at follow-up time.

The comparison of baseline clinical and imaging parameters between the two response groups revealed that NR had a significantly higher Ki-67 of the primary tumor (16.5% vs. 10.0%, p = 0.01) with three patients graded as neuroendocrine cancer (NEC) in the NR group (none in the R group). Responders showed a significantly higher hepatic tumor burden (20% vs. 5%, p = 0.007). There were no differences in imaging parameters of the NELM, while for the pNETs size varied significantly between response groups with greater diameters of the baseline pNET in R compared to NR (76 mm vs. 38 mm, p = 0.02). However, the statistical evaluation of pNET was limited by the small number of patients with non-resected pNET in our cohort (14 and 12 respectively).

After treatment initiation there was a significant difference in the change of chromogranin A (CgA) between response groups, with an increase in NR compared to a mild decrease in R (61% vs. −2%, p < 0.04). Regarding imaging parameters, there were significant differences in the changes of the size of both NELM (20% vs. −8%, p = 0.038) and pNET (2% vs. −55% p < 0.013) between the two response groups.

Additionally, changes of ADC in NELM differed significantly between response groups, with a decrease in both ADCmin (−23%) and the liver adjusted ADCmean / ADCmean liver ratio (−16%) in NR, compared to an increase in R of both ADCmin (50%) and ADCmean / ADCmean liver (30%). Notably there were no differences in changes in arterial vascularization and signal intensity (SI) on T1w and T2w images between response groups.

ROC analysis of the previously selected imaging and clinical parameters revealed AUC values differing from 0.71 (∆ Size NELM and ∆ ADCmin) to 0.76 (∆ ADC mean/Liver ADCmean) for classifying non-responders vs. responders. The highest AUC for a single parameter was found for ∆ ADC mean/Liver ADCmean, with a median cut-off of < 6.9 which yielded a sensitivity of 76% and a specificity of 75%. The combination of ∆ Size NELM and ∆ CgA or ∆ ADC mean/Liver ADCmean could each slightly, though not significantly, improve AUC (0.79 and 0.77 respectively), while the combination of ∆ Size NELM and ∆ ADCmin yielded the best balance for sensitivity and specificity with 88% and 60% compared to 69% and 65% respectively for ∆ Size NELM alone. Subsequent Kaplan-Meier survival analysis, utilizing the respective median cut-off values (Table 4 and Figure 4) for the parameters, revealed significantly longer PFS times for ∆ ADCmean/ADCmean liver < 6.9 (p = 0.024) and the combination of ∆ Size NELM > 0% + ∆ ADCmin < −2.9% (p = 0.021).

FIGURE 4.