Epidemiology, treatment and outcomes of gastroenteropancreatic neuroendocrine neoplasms Authors: Johannes Uhlig, James Nie, Joanna Gibson, Michael Cecchini, Stacey Stein, Jill Lacy, Pamela Kunz, Hyun S Kim Published: 2024 ABSTRACT AbstractTo investigate incidence, treatment patterns and outcomes of gastroenteropancreatic neuroendocrine neoplasms (GEP-NEN) in the United States. The 2019 National Cancer Database was searched for adult GEP-NEN patients. Main outcomes included overall and site-specific incidence, treatment patterns, and overall survival (OS). Overall survival was evaluated using averaged Cox regression. 86,324 GEP-NEN patients were included (6.33% of all GEP malignancies). From 2004 to 2016, annual GEP-NEN cases increased (n = 4,010 to n = 9,379), largely driven by low-stage, low-grade disease. Most patients received surgery, either alone (72.9%) or in combination with systemic therapy (4.9%). Longest overall survival (OS) was evident in patients with low stage and low grade GEP-NEN of the small intestine and rectum (p < 0.001). Patients undergoing surgical resection demonstrated longest OS. The addition of systemic therapy was most effective in high stage G3 NEN. Having higher income (≥$63,333) and private insurance or Medicare, but not Medicaid, was associated with improved survival. GEP-NEN incidence increases, likely due to improved detection and diagnosis. Treatment patterns have evolved to follow the latest international guidelines and site-specific improvement in survival is noted. In addition to disease specific factors, insurance access and socioeconomic factors emerged as potential targets for improving outcomes.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-024-81518-4.Keywords: Neuroendocrine tumors, Incidence, SurvivalSubject terms: Cancer epidemiology, Gastrointestinal cancer, Surgical oncology Introduction Neuroendocrine neoplasms (NEN) are a heterogenous cancer entity derived from neoplastic transformation of neuroendocrine cells throughout the body with approximately two-thirds arising from the gastroenteropancreatic system1–3. Gastroenteropancreatic neuroendocrine neoplasms (GEP-NEN) are predominantly indolent, but there is considerable heterogeneity in their clinical presentation based on site of origin, hormone secretion, disease stage, and histological grade4–6. Since their discovery in 1907, GEP-NEN classification has undergone numerous revisions as data informing their biology has emerged4. As of 2019, the World Health Organization (WHO) classifies GEP-NEN by histologic differentiation pattern as well-differentiated neuroendocrine tumor (WDNET) or poorly-differentiated neuroendocrine carcinoma (PDNEC) (Supplemental Table 1; representative pathology in Supplemental Figs. 1 and 2)3,7. The vast majority of GEP-WDNET tumors are G1 or G2, while GEP-NEC are G37. The histologic differentiation and WHO classification of grade mirror the distinct molecular differences between GEP-WDNET and GEP-NEC. For instance, pancreatic WDNETs have mutations in DAXX/ATRX, MEN1 or the mTOR pathway, whereas pancreatic NECs have mutations in RB1, TP53 and CDKN2A8–11. Recent publications and the WHO have reported on the new category of GEP-WDNET G3 tumors, which are distinct histologically and molecularly from poorly differentiated NEC/G3 cancers7,12,13. Although GEP-NEN are considered rare, several large population-based studies have reported an increasing incidence over the past four decades, possibly attributable to increased use of cross-sectional imaging and endoscopic procedures and more widely adopted classification schemes1,5,14–16. In comparison to incidence, only in recent years has survival in GEP-NEN been shown to improve4,5,16. Recent approvals of everolimus, sunitinib, and peptide receptor radionuclide therapy (PRRT) may further bolster this trend1. To date, there is scarce literature reporting on the treatment and associated outcomes for GEP-NEN. Improved understanding of GEP-NEN is vital as their prevalence increases due to their indolent behavior17. In this study, we comprehensively evaluate GEP-NEN regarding contemporary incidence in the United States, current treatment approaches and associated outcomes. Materials and methods Data source The 2019 participant user file of the National Cancer Database (NCDB) was queried in March 2020 comprehensively evaluate incidence, treatment patterns, and outcomes of GEP-NEN. The NCDB was established in 1989 as a joint quality improvement project between the Commission on Cancer of the American College of Surgeons and the American Cancer Society. It contains data on over 34 million records from more than 1500 Commission on Cancer–accredited facilities in the United States, making it approximately 3 times larger than the Surveillance, Epidemiology, and End Results (SEER) database18. Approximately 70% of newly diagnosed cancer patients are covered by the NCDB annually. Still, the NCDB data are not representative to the general US-population. This retrospective study was approved by the Yale University School of Medicine ethics committee. Due to the retrospective nature of the study and the evaluated anonymized NCDB data, the Yale University School of Medicine ethics committee waived the need of obtaining informed consent. This study was conducted in accordance to the Declaration of Helsinki in its most recent version. Study cohort Adult patients diagnosed between 2004 and 2016 with GEP-NEN were identified using site codes (160–169, 170–179, 180, 182–189, 199 and 209, and 250–259) and ICD-03 histologic codes (8013, 8150–8153, 8155–8157, 8240–8242, 8244–8246, 8249, and 9091). The following ICD-03 histological codes were excluded: 8243 (goblet cell carcinoid), 8154 (mixed islet-cell/ exocrine adenocarcinoma), and 8574 (adenocarcinoma with neuroendocrine features). We also excluded primary diagnoses of lung cancer, medullary carcinoma of the thyroid, pheochromocytoma, and paraganglioma as well as patients with secondary cancer diagnoses. NEN were further divided in pancreatic NEN and gastrointestinal NEN (GI-NEN). GEP cancers of non-neuroendocrine histologies from the same site codes were queried for epidemiological comparison. The follow-up period ended on January 1st, 2017, after which patients were right-censored. Exclusion criteria were age < 18, AJCC stage 0 disease, and missing data on survival status, metastatic sites, and follow-up time. Socioeconomic variables and tumor factors were evaluated as potential predictors of outcomes. The primary objective of this study was to assess overall survival, defined as the number of months from diagnosis to date of death or censoring. Variables Data collected included patient-level variables (age, sex, race, insurance status, education, residency area, region, comorbidities measured by Charlson-Deyo Comorbidity index), hospital-level variables (academic vs. non-academic), and tumor-level variables (year of diagnosis, tumor location, time from diagnosis to treatment, stage, grade, tumor size, presence and location of metastases. T and N categories were classified by the higher rating when comparing clinical and pathological assessment). To ensure patient anonymity, the NCDB suppressed facility type and location for patients aged under 39 years. First line treatment was stratified by surgical resection of the primary tumor with or without addition of systemic therapy, or systemic therapy without surgical resection. In the NCDB, “systemic therapy” refers to, but does not distinguish between various cytotoxic chemotherapies, sunitinib, or everolimus. Other therapies, such as somatostatin analogues, were either not captured or recorded in limited capacity and were thus not included in treatment analysis. NCBD does not provide Ki-67 proliferative index or mitotic rates, but does report tumor grades as which for this study were summarized as “G1, G2” (well differentiated; moderately differentiated; intermediate differentiation) and “G3” (poorly differentiated; undifferentiated; anaplastic). For the majority of statistical analyses, G1 and G2 tumors were analyzed together, as these could not be reliably distinguished according to the current WHO classification system. Statistical analyses Data were analyzed in March 2020. Continuous variables were compared using the Wilcoxon rank sum test and categorical variables were compared using the χ2 test. Overall survival was initially assessed using the Kaplan-Meier methods and compared using log-rank analysis. Covariates with p < 0.10 in univariate analysis were considered for inclusion into the multivariable Cox proportional regression model. The proportional hazards assumption was visually assessed plotting Schoenfeld residuals. Due to evident violations of the proportional hazards assumption, weighted Cox regression models were used for all survival models, implementing Prentice weights, censoring correction and robust variance estimation19. For statistical analyses, GEP-NEN tumor locations were partially combined. A priori, a statistical multiplicative interaction test between tumor grade, tumor stage, and treatment was planned to evaluate varying treatment effectiveness with two-way and three-way interaction terms in the multivariable model. All analyses were performed using R, version 3.4.3 (R Core Development Team, Vienna, Austria), and RStudio, version 1.1.414 (R Studio Inc, Boston, MA). An α level of 0.05 was chosen to indicate statistical significance and all provided P values are 2-sided. Hazard ratios are reported with 95% confidence intervals (95% CI). Results Patient population A total of n = 86,324 patients with GEP-NEN fulfilled the inclusion criteria (Table 1). Patients were diagnosed at a median age of 58 years (interquartile range [IQR], 50–68 years), with n = 42,026 male (48.7%) and n = 44,298 female (51.3%) patients. GEP-NEN comprised 6.3% of all cancers of the GEP system diagnosed during this time period (total n = 1,363,843). The absolute number of annual new cases of GEP-NEN increased by 133.9% from 2004 to 2016, compared to 9.3% for GEP cancers of non-neuroendocrine histology (Fig. 1). In 2004, GEP-NEN contributed 4.3% of all GEP cancers, increasing to 8.8% by 2016. Compared to gastrointestinal neoplasms of non-neuroendocrine histology, GEP-NEN patients were more likely to be diagnosed at a younger age (median age of 58 vs. 66 years; p < 0.0001), be female (51.3% vs. 47.5%; p < 0.0001), be African American (17.3% vs. 12.5%; p < 0.0001), have private insurance (52.8% vs. 36.8%; p < 0.001), belong to higher income brackets (p < 0.001), and have fewer comorbidities (p < 0.0001). Patient characteristics are summarized in Table 1. Table 1Baseline characteristics of included patients. Provided p-values assess overall differences in variable distribution in the GEP-NEN and Non-GEP group.GEP-NENNo. 86,324GEP Non-NEN histologyNo. 1,277,519P-valueAge58.0 (50.0–68.0)66.0 (56.0–77.0)< 0.0001Gender< 0.0001 Female44,298 (51.3%)607,407 (47.5%) Male42,026 (48.7%)670,112 (52.5%)Race< 0.0001 African American14,910 (17.3%)159,452 (12.5%) Caucasian66,296 (76.8%)1,044,744 (81.8%) Other race5,118 (5.9%)73,323 (5.7%)Insurance status< 0.0001 Not Insured2,963 (3.4%)49,021 (3.8%) Private Insurance/Managed Care45,610 (52.8%)470,504 (36.8%) Medicaid5,783 (6.7%)76,406 (6.0%) Medicare29,021 (33.6%)642,023 (50.3%) Other Government1,101 (1.3%)13,838 (1.1%) Insurance Status Unknown1,846 (2.1%)25,727 (2.0%)Median household income for residence area< 0.0001 < $40,22714,755 (17.1%)224,072 (17.5%) $40,227 − 50,35316,407 (19.0%)254,979 (20.0%) $50,354 − 63,33218,129 (21.0%)262,621 (20.6%) >=$63,33329,009 (33.6%)387,078 (30.3%) Missing8,024 (9.3%)148,769 (11.6%)Residency area< 0.0001 Metropolitan area72,630 (84.1%)1,047,748 (82.0%) Urban area8,407 (9.7%)140,476 (11.0%) Rural area/NA5,287 (6.1%)89,295 (7.0%)Comorbidities [Charlson Score]< 0.0001 065,969 (76.4%)902,392 (70.6%) 114,850 (17.2%)266,289 (20.8%) 23,627 (4.2%)73,341 (5.7%) >=31,878 (2.2%)35,497 (2.8%)Time diagnosis to treatment< 0.0001 Median (IQR)0.0 (0.0–34.0)15.0 (1.0–31.0) Missing13,087 (15.2%)206,357 (16.2%)Tumor location< 0.0001 Appendix6,890 (8%)14,202 (1.1%) Ascending colon1,318 (1.5%)121,981 (9.5%) Cecum3,384 (3.9%)132,012 (10.3%) Descending colon186 (0.2%)35,385 (2.8%) Pancreas18,774 (21.7%)258,220 (20.2%) Rectosigmoid junction916 (1.1%)71,385 (5.6%) Rectum16,370 (19.0%)194,173 (15.2%) Sigmoid colon1,388 (1.6%)166,258 (13.0%) Small intestine28,750 (33.3%)20,578 (1.6%) Stomach6,877 (8.0%)130,309 (10.2%) Transverse colon588 (0.7%)100,331 (7.9%) Other site793 (0.9%)32,685 (2.6%)Stage< 0.0001 Stage I22,879 (26.5%)235,430 (18.4%) Stage II8,233 (9.5%)302,688 (23.7%) Stage III11,746 (13.6%)289,876 (22.7%) Stage IV15,651 (18.1%)349,759 (27.4%) Unknown stage27,815 (32.2%)99,766 (7.8%)Tumor grade< 0.0001 Grade I37,750 (43.7%)99,509 (7.8%) Grade II7,726 (9.0%)613,904 (48.1%) Grade III6,502 (7.5%)251,212 (19.7%) Grade IV0 (0.0%)21,820 (1.7%) Grade unknown34,346 (39.8%)291,074 (22.8%)Tumor size [mm]< 0.0001 Median (IQR)17.0 (8.0–32.0)40.0 (29.0–58.0) Missing26,063 (30.2%)368,155 (28.8%)Metastases< 0.0001 Brain metastasis103 (0.1%)2,449 (0.2%) Bone metastasis761 (0.9%)12,402 (1.0%) Liver metastasis8,348 (9.7%)126,344 (9.9%) Lung metastasis722 (0.8%)37,159 (2.9%) Missing39,002 (45.2%)653,607 (51.2%)Treatment< 0.0001 Systemic alone5,478 (6.3%)217,071 (17.0%) No surgery or systemic13,707 (15.9%)195,336 (15.3%) Surgery w/o systemic62,945 (72.9%)458,791 (35.9%) Surgery + systemic4,194 (4.9%)406,321 (31.8%)Treatment facility type< 0.0001 Academic center31,687 (36.7%)416,946 (32.6%) Non-academic center46,529 (53.9%)823,750 (64.5%) Facility type suppressed for age 0–39 years8,108 (9.4%)36,823 (2.9%)Treatment facility location< 0.0001 East North-Central14,320 (16.6%)221,923 (17.4%) East South-Central5,313 (6.2%)86,810 (6.8%) Middle Atlantic11,765 (13.6%)196,914 (15.4%) Mountain3,323 (3.8%)50,141 (3.9%) New England4,508 (5.2%)70,528 (5.5%) Pacific8,620 (10.0%)140,164 (11.0%) South Atlantic17,183 (19.9%)267,197 (20.9%) West North-Central6,298 (7.3%)94,024 (7.4%) West South-Central6,886 (8.0%)112,995 (8.8%) Facility location suppressed for age 0–39 years8,108 (9.4%)36,823 (2.9%) Baseline characteristics of included patients. Provided p-values assess overall differences in variable distribution in the GEP-NEN and Non-GEP group. Annual case numbers for gastroenteropancreatic neuroendocrine neoplasms and gastroenteropancreatic non-endocrine neoplasms stratified by location. Overall mean increase provided as black line. Disease characteristics The highest proportion of GEP-NEN originated in the small intestine (33.3%) and lowest in the descending colon (Fig. 2). GEP-NEN patients more commonly were smaller in size (median size 17.0 mm vs. 40.0 mm) and of lower grade compared to GEP non-NEN histologies (p < 0.001): 87.5% of GEP-NEN with known grades were G1/G2 (60.2% patients with known grade). 18.1% GEP-NEN patients had stage IV disease at diagnosis, compared to 27.4% for GEP non-neuroendocrine histologies (p < 0.001). Among patients with known GEP-NEN grade and stage (n = 41,766), a cross tabulation is provided in Supplemental Table 2. The increase in GEP-NEN cases over the study period was driven by low-stage, low-grade disease across most primary sites; pancreatic and small intestine NEN exhibited an increase in high-stage disease (Supplemental Figs. 3 & 4). Notably, the number of NEN with unknown AJCC stage abruptly decreased in 2010 across all strata. Supplemental Fig. 5 summarizes geographic differences in annual GEP-NEN case numbers and stage distribution in the United States. Neuroendocrine neoplasm incidence according to primary location of the gastroenteropancreatic system. Treatment trends Patients with GEP-NENs were more likely to be treated at an academic center than non-neuroendocrine GI patients (36.7% vs. 32.6%). Most GEP-NEN patients were treated with surgery, either alone (72.9%) or in combination with systemic therapy (4.9%). Only 6.3% of GEP-NEN patients received systemic therapy without surgery and 15.9% of GEP-NEN patients received best supportive care. Stratification by cancer site, stage, and grade revealed that during the study period, utilization of systemic therapy increased for G3 NEN irrespective of site or stage (Supplemental Fig. 6). For GEP-NEN, the mainstay of treatment was surgical resection across all disease stage/strata with additional use of systemic therapy for G3 disease. Use of systemic therapy in G1/2 GI-NEN disease declined during the study period. For high stage, G3 GI-NEN, use of surgery without systemic therapy declined 49% in favor of systemic therapy alone. For pancreatic NEN, surgical resection alone was most often utilized in low-stage G1/2 disease, and alone or with systemic therapy in low stage, G3 disease. For high-stage, G3 pancreatic NEN, most patients received systemic therapy alone by 2016 – the proportion receiving surgery alone and best supportive care declined. Survival analyses Overall survival was assessed in a cohort of n = 43,516 GEP-NEN patients with known grade, follow-up time and survival status. Median follow in this cohort was 44.1 months (IQR: 23.7–71.9 months). For statistical modelling, tumor size was not considered due to missing data. On multivariable analyses, OS was associated with grade, stage, location, and treatment (Table 2). Shorter OS was evident for patients with G3 vs. G1/G2 GEP-NEN (HR = 3.15 (95% CI: 2.81–3.52), p < 0.001) (Fig. 3). Efficacy of surgical resection and systemic therapy varied by stage and grade (Fig. 4): across all strata, patients undergoing surgical resection demonstrated favorable outcomes, with longest OS reported in patients with stage 1/2 G1/2 GI-NEN (median OS > 156 months). In G3 disease, the combination of surgery and systemic therapy demonstrated longer OS versus surgery alone. Factors associated with improved survival on multivariate analysis included having higher income (≥$63,333) having private insurance, Medicare, or other government insurance, and receiving treatment at an academic center. Of note, having Medicaid was not associated with improved survival over being uninsured. Table 2Univariate and multivariable Cox regression models evaluating overall survival, using weighted regression with Prentice weights, censoring correction and robust variance estimation to account for violations of the proportional hazards assumption. Variables with p < 0.1 in univariate models were considered for inclusion in the final multivariable model, and retained if multivariable p < 0.05. * facility location not included in the multivariable model due to collinearity with facility type.VariableLevelsUnivariate HR (95% CI)Multivariable HR (95% CI)Age[18.0,90.0]1.04 (1.02–1.05), p < 0.0011.04 (1.03–1.04), p < 0.001GenderFemale1 (reference)1 (reference)Male1.07 (0.97–1.18), p = 0.1491.09 (1-1.19), p = 0.042RaceAfrican American1 (reference)-Caucasian1.14 (0.99–1.32), p = 0.066-others0.73 (0.59–0.9), p = 0.004-Proportion of patients without highschool diploma>=17.6%1 (reference)-10.9–17.5%0.7 (0.44–1.09), p = 0.113-6.3–10.8%1.14 (0.86–1.5), p = 0.356-< 6.3%0.74 (0.62–0.89), p = 0.001-Urbanmetropolitan area1 (reference)-urban area1.99 (1.2–3.3), p = 0.007-rural area/NA1.18 (0.92–1.51), p = 0.185-InsuranceNot Insured1 (reference)1 (reference)Private Insurance/Managed Care0.58 (0.49–0.69), p < 0.0010.61 (0.51–0.72), p < 0.001Medicaid0.85 (0.68–1.06), p = 0.1540.84 (0.67–1.05), p = 0.132Medicare1.62 (1.37–1.9), p < 0.0010.79 (0.66–0.95), p = 0.013Other Government0.67 (0.47–0.94), p = 0.0220.65 (0.47–0.89), p = 0.008Insurance Status Unknown0.99 (0.73–1.34), p = 0.9530.9 (0.67–1.21), p = 0.487Income< $40,2271 (reference)1 (reference)$40,227 − 50,3530.94 (0.78–1.13), p = 0.510.95 (0.82–1.11), p = 0.513$50,354 − 63,3320.88 (0.72–1.07), p = 0.1960.86 (0.73–1.02), p = 0.075>=$63,3330.73 (0.61–0.87), p < 0.0010.79 (0.68–0.91), p = 0.002Comorbidities (Charlson Deyo Comorbidity Index)01 (reference)1 (reference)11.6 (1.35–1.89), p < 0.0011.34 (1.17–1.54), p < 0.00122.19 (1.89–2.53), p < 0.0011.59 (1.38–1.85), p < 0.001>=32.95 (2.4–3.63), p < 0.0012.44 (2.01–2.95), p < 0.001Cancer gradeG1/G21 (reference)1 (reference)G35.53 (4.96–6.16), p < 0.0013.18 (2.85–3.55), p < 0.001StageStage I1 (reference)1 (reference)Stage II2.17 (1.87–2.51), p < 0.0011.45 (1.23–1.69), p < 0.001Stage III2.11 (1.82–2.45), p < 0.0011.43 (1.21–1.68), p < 0.001Stage IV7.48 (6.6–8.47), p < 0.0013.98 (3.43–4.62), p < 0.001Unknown stage2.18 (1.88–2.52), p < 0.0011.69 (1.44–1.99), p < 0.001Cancer siteAppendix1 (reference)1 (reference)Colon3.7 (2.92–4.69), p < 0.0011.22 (0.97–1.52), p = 0.084Other site0.16 (0.02–1.04), p = 0.0550.05 (0.01–0.31), p = 0.001Pancreas2.02 (1.64–2.48), p < 0.0011.11 (0.89–1.37), p = 0.362Rectum0.98 (0.74–1.29), p = 0.8810.77 (0.59–1.01), p = 0.059Small intestine1.61 (1.31–1.99), p < 0.0010.95 (0.77–1.18), p = 0.651TreatmentChemo alone1 (reference)1 (reference)No surgery or chemo0.44 (0.37–0.52), p < 0.0010.82 (0.7–0.96), p = 0.011Surgery w/o chemo0.14 (0.12–0.17), p < 0.0010.4 (0.34–0.46), p < 0.001Surgery + chemo0.43 (0.35–0.52), p < 0.0010.5 (0.42–0.59), p < 0.001Year of cancer diagnosis[2004.0,2015.0]0.94 (0.93–0.95), p < 0.0010.97 (0.96–0.98), p < 0.001Facility typeAcademic center1 (reference)1 (reference)Facility type suppressed for age 0–39 years0.45 (0.37–0.54), p < 0.0011.59 (1.22–2.06), p < 0.001Non-academic center1.23 (1.11–1.36), p < 0.0011.19 (1.09–1.3), p < 0.001Facility locationEast North Central1 (reference)-East South Central1.15 (0.74–1.77), p = 0.541-facility location suppressed for age 0–39 years0.38 (0.31–0.47), p < 0.001-Middle Atlantic0.9 (0.78–1.03), p = 0.117-Mountain0.8 (0.65–0.98), p = 0.033-New England0.82 (0.61–1.1), p = 0.192-Pacific0.82 (0.69–0.98), p = 0.026-South Atlantic1.08 (0.94–1.25), p = 0.271-West North Central1.01 (0.83–1.22), p = 0.948-West South Central0.93 (0.8–1.08), p = 0.343- Univariate and multivariable Cox regression models evaluating overall survival, using weighted regression with Prentice weights, censoring correction and robust variance estimation to account for violations of the proportional hazards assumption. Variables with p < 0.1 in univariate models were considered for inclusion in the final multivariable model, and retained if multivariable p < 0.05. * facility location not included in the multivariable model due to collinearity with facility type. Kaplan-Meier plots showing shorter OS for Grade 3 disease across GI and pancreatic NEN. Grade 1 and Grade 2 tumors were not classified using current WHO criteria (mitotic count and Ki67 proliferative index) given NCBD limitations. Plots were limit to 132 months due to low subsequent event rates. Overall survival of GEP-NEN patients stratified by stage and treatment. Across all strata, patients undergoing surgical resection demonstrated favorable outcomes, with longest OS reported in patients with low-stage and low-grade GEP-NEN. Patients with GEP-NEN benefited from systemic therapy over best supportive care only in Grade 3 disease. Discussion To our knowledge, this is the first study to examine epidemiology, treatment, and outcomes of GEP-NEN in a national cohort. Using contemporary US data, we demonstrated that between 2004 and 2016, the annual number of GEP-NEN cases more than doubled, growing disproportionately in comparison to non-neuroendocrine GEP histologies. These findings mirror the magnitude of increasing incidence of GEP-NENs reported in recent North American population-based studies5,14,16,20. The greatest increases seen in GEP-NEN sites were the cecum/appendix (5.4-fold), pancreas (3.9-fold), and stomach (2.4-fold). For non-neuroendocrine GEP neoplasms, only the rates of small intestine (2.2-fold) and pancreatic diagnoses (1.6-fold) increased substantially. While increased use of cross-sectional imaging and endoscopy has significantly outpaced growth of non-neuroendocrine GEP cancers, it has significantly increased detection of earlier stage disease GEP-NEN5,17,21. Within our cohort, low-grade and non-metastatic disease contributed to the majority of the increase in annual GEP-NEN diagnoses (Supplemental Figs. 1 & 2). A portion of the increase in low-grade disease can be attributed to wider adoption of classification systems, reflected in the decline in cases with unknown grade and unknown stage after 201022. Interestingly, our sample showed increased detection of both low and high-stage pancreatic NENs. Increased incidental detection has made it such that non-functioning pancreatic NENs now make up the majority of new pancreatic NEN diagnoses23. Earlier diagnosis of these non-functioning tumors, of which > 50% have been found to exhibit malignant potential, may have contributed to the improved 5-year survival noted in our sample compared to a SEER analysis from a decade ago (65% vs. < 40%).16 The advent of new targeted therapies such as sunitinib or everolimus in advanced disease may have also contributed to this trend. The 5-year survival of colonic and gastric primaries in our cohort are similar to a SEER study from a decade ago, possibly due to the lack of increase in low grade or low stage disease at both sites16. In our cohort, treatment effectiveness varied by stage and grade: across all strata, patients treated with surgical resection demonstrated longest overall survival. This is in line with international guidelines that recommend surgical resection as the mainstay of treatment for low grade local or locoregional GEP-NEN24. Present guidelines differ in their recommendations for cytotoxic chemotherapy: the European Society for Medical Oncology recommends chemotherapy for advanced pancreatic NENs and high grade NENs of any site, whereas the National Comprehensive Cancer Network (NCCN) recommends chemotherapy for advanced pancreatic NENs only24,25. Within NCCN, there is disagreement on whether to offer chemotherapy for advanced GI-NEN given the low response rates and lack of PFS benefit25. Our data show that addition of systemic therapy, which largely corresponds to chemotherapy in NCDB (but not somatostatin analogues), was most effective in G3 stage III and IV disease, with use increasing across all G3 disease over time. Use of systemic therapy in low grade GI-NEN declined during the study period, but approximately 5% of patients with low grade, high stage disease still received systemic therapy in 2016. For a subset of aggressive, low-grade GI-NENs, some NCCN panelists make provisions for using cytotoxic chemotherapy when no other options are available24–26. Use of cytotoxic chemotherapy in this subset of patients may explain the worse outcomes in G1/G2 disease for systemic therapy across all stages of GEP-NEN (Fig. 4). While the combination of surgery and systemic therapy was associated with the best OS for high-stage, G3 GEP-NENs, studies in recent years suggest differential response based on histology27,28. Response to platinum-based chemotherapy has been noted to be substantially greater for patients with Ki-67 proliferation index of > 55%, which most often corresponds to NEC, whereas surgery has been associated with better survival for G3 GEP-NETs than for GEP-NEC27–29. In our cohort, privately insured patients made up a higher proportion of GEP-NEN compared to other GI cancers, perhaps reflective of the younger age at diagnosis and improved access to diagnostic imaging for detection of disease. We found treatment at an academic center to be associated with better outcomes, suggestive of the importance of integrating multi-disciplinary expertise in managing GEP-NENs30. Given the delays in obtaining a GEP-NEN diagnosis, difficulty in finding a treatment center with sufficient expertise, and the chronic course of most GEP-NENs, having proper access to care and sufficient financial resources can drastically alter disease prognosis30–32. Insured GEP-NEN patients have been shown to be significantly more likely to receive somatostatin analogues and less likely to undergo emergency surgery for tumor resection31. While we report survival benefit for privately insured and Medicare patients, the surprising finding that Medicaid patients have worse outcomes than uninsured patients in multivariate analysis shows that socioeconomic factors beyond access to care are important for survival. This corroborates findings from a universal healthcare system study: a Canadian study found that NEN patients with low socioeconomic status had worse outcomes despite similar stage at diagnosis and similar treatment33. In multivariable analysis, a recent SEER study that suggests that the racial disparities in GEP-NEN survival are decreasing and that the differences now seen are driven largely by socioeconomic and psychosocial factors33,34. We similarly did not find racial differences in survival in multivariate analysis nor did we find survival differences for patients living in rural versus metropolitan areas, in line with another SEER study35. Geographically, the West North-Central and West South-Central United States were associated with higher rates of stage III/IV disease, yet survival differences across regions were marginal in univariate analysis. The strength of this study lies in the detailed assessment of demographic, diagnostic, and treatment trends from a large patient population representative of typical patients treated in the United States. Still, this study has several limitations. The NCDB collects hospital-based data, which cannot provide population-wide incidence, with potential limitations to the generalizability of our findings. Although the NCDB covers approximately 70% of newly diagnosed cancer cases in the USA annually, it is not necessarily representative of the US-population, unlike the SEER database (which collects data from representative geographic regions and subpopulations). However, the increasing GEP-NEN cases mirror the findings from other population-based studies5,14–16. Additionally, studies have shown that survival parameters are comparable between the NCDB and SEER database, for example in patients with CRC, while patients demographics may vary slightly35. Further, the comparison of NEN to all non-NEN histologies made in this study fail to reflect on the heterogeneity of histological subtypes of specific cancer sites, for example of the appendix or pancreas. NCDB also does not report tumor specific variables such as Ki-67 or mitotic rate and contains some use of discontinued histological terms, making classification of G1 and G2 disease unreliable given evolving classification systems during the study period. Thus, G1 and G2 disease were combined for this analysis. Nevertheless, we are confident that the G3 classification in our cohort is reliable, as the terms used for classification have been associated with G3 tumors throughout the study period. The NCDB does not report cancer-specific survival nor does it capture specific chemotherapies, targeted molecular agents, or somatostatin analogues so treatment efficacy cannot be compared. Few data exist to compare efficacy of systemic GEP-NEN therapies, but recent meta-analysis suggests that combination of systemic therapies, particularly those involving somatostatin analogues, may be most effective36. In patients with unresectable liver metastases, improvement in patient selection and investigation into the potential of somatostatin receptor analogues and targeted agents to sensitize the effects of locoregional therapy may further improve health-related quality of life for patients with treatment-refractory disease37–39. Conclusion Using the NCDB, we demonstrated that absolute number of GEP-NEN cases has continued to rise from 2004 to 2016, possibly due to increased diagnosis of early stage disease. Survival is dictated not only by disease specific factors, but also access to care and socioeconomic status. Promisingly, real world treatment has evolved to follow evidence driven guidelines. The recent technological improvements in GEP-NEN therapeutics, refinement of disease classification, and growing awareness of the social barriers involved in this disease may help drive improved outcomes in the years to come. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We acknowledge support by the Open Access Publication Funds/transformative agreements of the Göttingen University. Author contributions JU and JN contributed equally to the manuscript. JU, MC, SS, JL and HSK were involved in the concept of the study. JU and HSK acquired and analyzed the data. JU, JN, JG, MC, SS, JL, PK and HSK were involved in the design of the manuscript and revised the manuscript for important intellectual content. JU, JN and JG wrote the original draft of the manuscript and figures. All authors read and approved the final manuscript. Funding information Open Access funding enabled and organized by Projekt DEAL. No funding was acquired for this study. Data availability The data that support the findings of this study are available from the American College of Surgery but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the corresponding authors upon reasonable request and with permission of the American College of Surgery. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Johannes Uhlig and James Nie contributed equally to this work. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data that support the findings of this study are available from the American College of Surgery but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the corresponding authors upon reasonable request and with permission of the American College of Surgery.