Study design and patient acquisition
The CONKO-007 trial examines the effectiveness of chemoradiation compared to chemotherapy alone after induction chemotherapy with 3 cycles of gemcitabine or 6 cycles of FOLFIRINOX (Fig.
1a).
Eligible patients were aged 18 years or older with histologically-confirmed non-resectable adenocarcinoma of the pancreas without distant metastasis according to CT-imaging of thorax and abdomen. ECOG-performance status was equal or less than two. Each patient provided written informed consent before participating in the study.
The trial was conducted following ICH-GCP guidelines, approved by the central Ethics Committee (University Hospital Erlangen, 322_12AZ) and by the Federal Institute for Drugs and Medical Devices (BfArM, 4,038,763). The trial registration (EudraCT: 2009–014476-21, NCT01827553) was obtained prior to recruitment. Patients were recruited at 52 German cancer centers.
The data concerning pretherapeutic radiographs collected prospectively from the first 200 patients included in the trial were evaluated in this analysis. 50% of the patients were consecutively enrolled at 6 cancer centers: Erlangen, Magdeburg, Göttingen, Dresden, Freiburg, Bochum (> 10 patients per center) and 50% at the remaining centers (1–9 patients per center) between 04/2013 and 07/2016. The sequence of evaluation and documentation is displayed in Fig.
1a.
The CONKO-007 study network uses a GCP-certified commercially available clinical trials management system SecuTrial® [
18]. After patient’s enrollment, the participating centers uploaded CT or MR scans performed prior to therapy in the globally standardized DICOM format. One hundred ninety scans were sent as scheduled in the study protocol. The missing 10 scans were handed over at the monitoring visits. SecuTrial® pseudonymized the DICOM images during upload. Once the upload was completed, 5 leading pancreatic surgeons automatically got an email containing the invitation to view the images.
The initial radiographs taken prior to enrollment were either reviewed in a web based fashion online or downloaded and analyzed in a DICOM viewer. In 4 high volume centers for pancreatic surgery, experienced pancreatic surgeons evaluated the uploaded pictures blinded except for age, sex and study site. In one case, the CT or MR images were presented to the interdisciplinary tumor board and simultaneously evaluated by a radiologist and a surgeon. The observers were aware of the judgement of the enrolling center.
The following items were documented: Suspected liver metastasis, suspected peritoneal metastasis, distance between the tumor and vascular structure of more than 1 mm, tumor contact of less than 180°, tumor contact of more than 180° to the vascular structures. Furthermore, a category “cannot be defined” was available. The vascular structures evaluated were celiac trunk, common hepatic artery, superior mesenteric artery, branches of the jejunal artery, superior mesenteric vein, and portal vein. Finally, resectability was evaluated either as locally advanced, complete R0 resection possible, and R0 resection undetermined. The statement of the panel concerning resectability (R0 possible, not possible or undetermined) did not influence the sequence of treatment but was provided automatically to the participating centers by email, usually within 3 working days.
Statistical analysis
Analysis of parameters associated with local resectability was performed using SPSS (IBM Version 23.0, IBM Armonk, NY, USA) in conjunction with Excel (Microsoft, Redmond, WA). Significance was determined by Chi-square test with post-hoc analysis by cellwise adjusted residual analysis in two- way contingency tables according to Garcia-Perez [
19]. Multiple comparisons were accounted for by Bonferroni correction. Interobserver agreement was calculated by the estimation of Fleiss-kappa. A κ of below 0.199 indicates poor agreement, 0.200–0.399 indicates fair agreement, 0.400–0.599 indicates moderate agreement, 0.600–0.799 indicates strong agreement, and more than 0.800 indicates a very strong and almost perfect agreement. The dispersion index was calculated according to Loether and MacTavish [
20] and compared by Kruskal-Wallis-Test for independent samples followed by a Bonferroni post hoc test. A dispersion index of 0 indicates a perfect match of all 5 examiners.