Metastatic patterns vary by primary site in rare neuroendocrine carcinoma

In a retrospective population-based study of 6,044 adults with extrapulmonary poorly differentiated neuroendocrine carcinoma, recorded metastatic patterns at diagnosis differed markedly according to the tumor’s primary site. Pancreatic primaries had an adjusted prevalence of liver metastasis nearly four times that of genitourinary primaries and an adjusted prevalence of brain metastasis almost three times as high. By contrast, bone involvement was more frequently recorded for genitourinary tumors.

The analysis of Surveillance, Epidemiology, and End Results data for diagnoses from 2010 through 2023 also found that patients with more recorded metastatic organs had shorter overall survival. Despite these anatomical differences, adding complete primary-site-by-organ interactions improved prediction in later patients only negligibly, suggesting that most predictive information came from the number of involved organs rather than the complexity of their interaction. The eligible cohort included 4,826 tumors with small-cell morphology and 1,218 with large-cell morphology.

Median age was 67 years, with an interquartile range of 57 to 75 years, and 57.0% of patients were male. Distant-stage disease was recorded in 52.0% of the cohort, and first-course chemotherapy was recorded as administered in 67.8%. Information on bone, brain, liver, and lung involvement was complete for 5,820 patients, or 96.3%.

Of these, 2,356 had at least one of the four organs recorded as involved and formed the core cohort used to examine metastatic combinations and survival associations. Liver involvement produced the largest contrast among primary sites. It was recorded in 71.1% of pancreatic primaries, 52.9% of colorectal primaries, and 43.9% of upper gastrointestinal primaries, compared with 19.3% of genitourinary primaries.

After adjustment for age, sex, race or ethnicity, diagnosis year, and morphology, pancreatic tumors had an adjusted liver-metastasis prevalence ratio of 3.80 relative to genitourinary tumors, with a 95% confidence interval of 3.39–4.25. Corresponding ratios were 2.88, with a 95% confidence interval of 2.56–3.23, for colorectal primaries and 2.31, with a 95% confidence interval of 2.03–2.63, for upper gastrointestinal primaries. Brain metastases were less common overall, but pancreatic primaries again had the highest recorded prevalence, at 6.5%, and an adjusted prevalence ratio of 2.86, with a 95% confidence interval of 1.75–4.66, compared with genitourinary tumors.

Bone involvement was recorded in 22.1% of genitourinary primaries, while adjusted prevalence in every other primary-site group was approximately 35%–48% lower. Sensitivity analyses broadly supported the principal prevalence estimates. For pancreatic versus genitourinary liver metastasis, the adjusted prevalence ratio was 3.80 when unknown status was excluded, 3.77 when unknown values were treated as negative, and 3.79 after multiple imputation.

Comparable results were obtained for colorectal and upper gastrointestinal primaries, and extending the imputation procedure from 10 to 50 updates changed the site-specific estimates very little. These analyses strengthen the reported contrasts, but they cannot eliminate uncertainty introduced by registry recording practices. SEER documented involvement of bone, brain, liver, and lung only at diagnosis; peritoneal, distant nodal, adrenal, other-organ, and later metastases were incompletely represented.

Imaging modality, assessment intensity, and indications for testing were unavailable, so the patterns may reflect differences in both disease distribution and clinical ascertainment. Among the 2,356 patients in the core metastatic-spectrum cohort, median overall survival declined as the recorded number of involved organs increased. Median survival was 6.70 months for one organ, 5.32 months for two, 4.63 months for three, and 4.40 months for four.

Twelve-month survival was 29.7%, 20.9%, and 8.8% for patients with one, two, and three involved organs, respectively. A 12-month estimate was not reported for the four-organ group because its longest observed follow-up was 9.72 months. Liver involvement alone was the most frequent individual pattern, recorded in 1,038 patients, and was associated with median survival of 5.82 months.

Lung involvement alone had a median of 8.18 months, whereas the combination of liver and lung involvement had a median of 3.45 months. These observational associations describe the recorded disease burden; they do not show that additional organ involvement caused the shorter survival. Evidence that some organ-specific survival associations varied by primary site came from Cox models containing interactions between primary site and each metastatic organ.

The global test favored the interaction model over a main-effects model, with a likelihood-ratio chi-square statistic of 43.49 for 20 degrees of freedom and P=0.00176. In the full cohort, liver metastasis was associated with overall survival in genitourinary primaries at a hazard ratio of 1.65, with a 95% confidence interval of 1.40–1.95. Corresponding liver-metastasis hazard ratios were 2.35, with a 95% confidence interval of 1.59–3.47, for colorectal primaries and 3.02, with a 95% confidence interval of 1.64–5.58, for pancreatic primaries.

The adjusted liver hazard ratio for upper gastrointestinal primaries was 1.03, indicating little estimated association in that group, although the estimate was imprecise. A focused analysis using five liver-by-site interaction terms found a similar pattern: the pancreatic estimate was 2.80, with a 95% confidence interval of 1.59–4.95, in the full cohort, 3.75, with a 95% confidence interval of 1.73–8.14, in 2010–2017, and 3.23, with a 95% confidence interval of 1.16–9.02, in 2018–2022. Some survival associations also varied over time, complicating interpretation of a single hazard ratio.

Residual tests indicated nonproportional effects for age, liver involvement, selected primary-site indicators, and the gynecologic- and pancreatic-by-liver interactions. An exploratory piecewise model divided follow-up at six months and allowed liver associations to differ between the two intervals. The genitourinary liver hazard ratio was 1.99, with a 95% confidence interval of 1.59–2.50, before six months and 1.59, with a 95% confidence interval of 1.27–1.98, thereafter.

Pancreatic estimates were 1.79, with a 95% confidence interval of 0.87–3.68, and 3.71, with a 95% confidence interval of 1.50–9.20, respectively. The piecewise extension did not remove all evidence of nonproportionality, so the reported hazard ratios summarize associations that could vary during follow-up rather than constant effects throughout the disease course. Prediction models indicated that metastatic information was useful mainly when summarized as burden rather than as a detailed anatomical spectrum.

Four prespecified ridge-penalized Cox models were developed in 1,297 distant-stage patients diagnosed from 2010 through 2017 and evaluated without refitting in 1,374 later patients diagnosed from 2018 through 2022. The basic model included age, sex, and morphology; subsequent models added primary site and then either the recorded four-organ count or separate indicators for bone, brain, liver, and lung. In the later cohort, the basic model achieved a Harrell C-index of 0.545, compared with 0.573 after adding primary site, 0.598 after adding organ count, and 0.607 after adding organ identity.

The corresponding 12-month areas under the curve were 0.550, 0.614, 0.660, and 0.672, while the 12-month Brier scores were 0.212, 0.206, 0.198, and 0.196, with lower scores indicating less prediction error. Separate organ identity nevertheless produced a small statistical improvement over organ count. Paired bootstrap comparisons estimated a C-index difference of 0.0085, with a 95% confidence interval of 0.0035–0.0139, a 12-month AUC difference of 0.0119, with a 95% confidence interval of 0.0033–0.0231, and a Brier-score difference of −0.0020, with a 95% confidence interval of −0.0038 to −0.0004, in favor of the spectrum model.

By contrast, adding the full primary-site-by-organ interaction matrix to the adjusted survival analysis changed the later-period C-index by only 0.00169 and reduced the 12-month AUC by 0.0131 in a separate mortality analysis. Prediction remained modest overall. The spectrum model also showed calibration weakness at 24 months: its calibration slope was 0.670, and mean predicted mortality was 0.887 compared with an observed risk of 0.856.

Performance varied by primary site, with 12-month AUC ranging from 0.532 for upper gastrointestinal tumors to 0.725 for other primaries, cautioning against treating pooled performance as universal. The findings have important boundaries. Although the study included more than 6,000 patients from a common registry, it was constrained by the rarity and anatomical diversity of the disease, the pooling of uncommon sites into an “other” category, and sparse brain-positive and four-organ strata.

Pathological classifications came from registry coding without central review, Ki-67 index, mitotic count, immunohistochemical, or molecular data. Available clinical information was also incomplete, including performance status, laboratory values, quantitative tumor burden, systemic regimens, treatment response, and recurrence. Registry identifiers unavailable in the research export prevented adjustment for within-registry clustering.

Both development and temporal assessment cohorts came from SEER, model selection used later-period performance, and the study did not test whether metastatic patterns modify treatment benefit. The results therefore show that primary site was associated with substantially different recorded metastatic patterns in extrapulmonary poorly differentiated neuroendocrine carcinoma, but that the organ count captured most of the limited predictive information available from those patterns.

Subject of Research: Primary-site variation in recorded metastatic patterns and survival associations in extrapulmonary poorly differentiated neuroendocrine carcinoma

Article Title: Primary-site variation in recorded metastatic patterns and survival associations in extrapulmonary poorly differentiated neuroendocrine carcinoma

Article References: Liang, H., Wu, X., Li, X., Wu, Q., Qiu, R., & Zheng, F. (2026). Primary-site variation in recorded metastatic patterns and survival associations in extrapulmonary poorly differentiated neuroendocrine carcinoma. Frontiers in Endocrinology, 17, Article 1975853. https://doi.org/10.3389/fendo.2026.1975853

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1975853

Keywords: extrapulmonary neuroendocrine carcinoma, metastatic patterns, primary tumor site, liver metastasis, brain metastasis, bone metastasis, overall survival, SEER, cancer epidemiology, prognostic modeling, Primary-site, variation