Association rule mining of clinical and biomarker data in neuroendocrine tumors: A prospective study on disease progression

There is an unmet need for new methods to predict disease course in patients with neuroendocrine tumors (NET). We investigated 92 putative cancer-related plasma proteins including chromogranin A (CgA) and clinical parameters at the time of diagnosis to identify early factors associated with progressive (PD) or stable disease (SD). Patients with NET grade 1 and 2 of the small intestine (siNET) and pancreas (pNET) were included in this prospective study. Blood samples were obtained at the time of diagnosis before tumor-related therapy was initiated. During 3 years of follow-up, SD or PD was determined according to current clinical practice by each investigator. Association rule mining (ARM) was used to identify combinations of biomarkers and clinical parameters associated with SD or PD. Altogether, 115 patients with siNET and 30 with pNET with complete clinical and biomarker data were included in the analysis representing 3 years of follow-up. Several novel plasma proteins and clinical factors were associated with either PD or SD. In siNET, CgA (>4 upper limits of normal [ULN]) was the most frequent biomarker associated with PD. Females, in contrast to males, with CgA >4 ULN showed a high risk of progression (PPV 100%, NPV 63%). In the siNET cohort, Carboxypeptidase E (CPE) was a discriminating factor between SD and PD. CPE <3.03 was associated with SD, whereas CPE >3.14 was associated with PD (p = 0.003). In the pNET cohort, among clinical variables, only the presence of liver metastasis was associated with PD. CgA was not among the top biomarkers associated with PD. Several parameters, both clinical and biomarker data, as well as combinations of these, were associated with PD or SD 3 years after diagnosis. We identified novel biomarkers improving the association with PD or SD. [Correction added on 28 August 2025, after first online publication: Abstract has been updated.]