Early prediction of metastatic insulinoma using tumor size and biochemical markers: a retrospective cohort study

Insulinomas are rare functional pancreatic neuroendocrine tumors. Metastasis is found in up to 15% of patients and is difficult to predict at diagnosis.

We retrospectively analyzed 64 patients (56.2% female, median age 49 years) with biochemically confirmed insulinoma treated at a European Neuroendocrine Tumor Society Center of Excellence between 2010 and 2024, assessing clinical characteristics, tumor biology, diagnostics, treatments and survival (median follow-up 23 months) .

Metastasis was present in 27% of cases (17/64). Metastatic insulinomas had significantly larger primary tumors (median 32 vs. 15 mm, p < 0.001), higher Ki-67 indices (median 11% vs. 2%, p < 0.001), shorter overall survival (median 51 months vs. not reached, p < 0.001) and a shorter interval from symptom onset to diagnosis (median 2 vs. 8 months, p < 0.05). Surgery was performed in 79% of patients (39/45 non-metastatic vs. 10/17 metastatic insulinomas). All metastatic cases required further multimodal therapy, including pharmacological and local-ablative treatments. A logistic prediction model based on tumor size, fasting glucose and the insulin-to-C-peptide ratio was developed to predict metastasis in insulinoma patients and showed excellent performance (AUC 0.96, sensitivity 90.9%, specificity 93.8%) in the insulinoma cohort. ROC analysis identified clinically relevant size thresholds, with a tumor size cut-off at 19 mm capturing all metastatic insulinomas with available size data and a cut-off at 24 mm improving specificity for metastasis to 84.6%. Applied to 24 published insulinoma cases, the model achieved a classification accuracy of 75%. Stratification into three risk categories ("low," "medium," and "high") further enhanced clinical interpretability. In line with molecular studies, metastatic insulinomas in our cohort exhibited disproportionate proinsulin secretion in the available cases, supporting the concept of β-cell dedifferentiation with impaired prohormone processing as a feature of aggressive disease.

Our analysis highlights relevant clinical and biological distinctions between non-metastatic and metastatic insulinomas. Based on three diagnostic parameters, we propose a predictive model for early risk stratification to guide staging intensity and follow-up in insulinoma management.