Prognostic score in high-grade gastrointestinal neuroendocrine tumours (GI-NETS)

Background: Outcomes of patients with high-grade GI-NETs are poor; prognostic markers for risk-stratification are needed.Methods: Consecutive patients, diagnosed with high-grade GI-NETs between 1997-2014, were included. Prognostic factors were identified by the Log-rank test, Cox regression and logistic regression and ROC curve comparisons performed for prediction accuracy. Internal validation of the score by Bootstrap-corrected Harrell Concordance Index (C-index) and Resampling Model Calibration were performed.Results: One-hundred and nine patients were eligible for analysis. Median follow-up time was 9.7 months (1.3-102.9). Median age: 67.7 years (16.3-84.1); 62% male, 84% metastatic; 19% foregut, 5% midgut, 19% pancreas, 28% hindgut and 29% unknown primary. Median ki67: 70% (20-100); ECOG PS 0: 26%, 1: 52%; 70% received chemotherapy. Baseline median alkaline phosphatase (ALK) and LDH were 109 IU/l (45-2035) and 70 IU/l (258-11069), respectively. The maximum model included stage, PS, LDH, Na, ALK, ki67, number of metastatic sites, presence of liver and presence of lung metastases. The score, selected by the lowest Akaike Index Criterion, included liver metastases, PS, ki67, LDH and ALK with 0-6 points assigned to each, resulting in 4 risk groups (A-D) with predicted risk of death, detailed in Table. There was no difference in the survival prediction accuracy between the maximum model and the score. On multivariable analysis, the score was prognostic for overall survival (HR 1.95, 95%CI 1.55-2.47; p <0.001) and had good discrimination (C-index, 0.76) and calibration (mean error, 0.021; percentile 90, 0.037).Conclusions: This simple score identified high-grade GI-NET patients with meaningful differences in survival and may inform clinical decision-making and trial design.