Small Cell and Large Cell Neuroendocrine Carcinoma of the Colon and Rectum: Population-Based Analysis of Incidence, Survival, and Site-Specific Outcomes
Background: Colorectal neuroendocrine carcinomas (NECs), including small cell (SCNEC) and large cell (LCNEC) subtypes, are rare but aggressive malignancies with limited population-level data. Moreover, prior studies have limited stratification by histologic subtype and anatomic location, despite evidence of disease heterogeneity in colorectal malignancies. This study aimed to characterize the clinical characteristics, treatment patterns, and outcomes of colorectal NEC by histologic subtype and tumor location, and to compare them with neuroendocrine tumors (NETs) and adenocarcinomas (ACs). Methods: A retrospective cohort review was conducted utilizing the SEER database from 2000 to 2022 to identify patients with colorectal SCNEC, LCNEC, NET, and AC. Demographic, clinical, therapeutic and survival data were analyzed and compared using Kaplan–Meier methods, log-rank tests, and Cox proportional hazards regression. Results: A total of 790 SCNEC, 498 LCNEC, 31,200 NET, and 638,898 AC cases were identified. SCNEC and LCNEC were associated with advanced stage at diagnosis (66% and 55% stage IV, respectively) and inferior survival outcomes compared to NET and AC (median OS (mOS) of 7 and 8 months for SCNEC and LCNEC, respectively, p = 0.0002). Stratified by location, colonic SCNEC had the poorest survival (mOS 5 months), worse than rectal SCNEC (mOS 9 months); this pattern was not observed in LCNEC. Surgical resection was associated with improved OS in both NEC subtypes (HR range 0.25–0.56, all p < 0.012). Radiation therapy was associated with improved survival in NEC overall (SCNEC HR 0.58; LCNEC HR 0.71, both p < 0.05), with the observed benefit appearing greatest in rectal NEC. Demographic factors were associated with survival in NET and AC but had no significant impact in NEC. Conclusions: Colorectal SCNEC and LCNEC are highly aggressive malignancies with poor outcomes, with SCNEC demonstrating modestly inferior outcomes compared to LCNEC. This study demonstrates that both histologic subtype and anatomic site of disease are important determinants of prognosis and treatment response. This supports the concept that colorectal NEC represents a biologically heterogeneous disease, with colonic SCNEC showing the worst outcomes and rectal NEC associated with a potential enhanced response to radiation. These findings support the need for further prospective and molecular studies to better define treatment approaches and targeted therapies for these rare malignancies.
Performance Modeling of Lightweight Retrieval-Augmented Large Language Models for Low-Resource Plastic Surgery Settings
Background: Large language models (LLMs) are being used by surgeons for education and reference yet concerns about hallucinations and reliability limit safe adoption. Retrieval-augmented generation (RAG) can offer a potential solution by grounding responses in a high-quality external database (e.g., medical textbooks) to enhance accuracy. However, performance tradeoffs across different RAG configurations—many of which exponentially increase computational cost—remain poorly characterized. Methods: In total, 120 lightweight, open-source RAG configurations were evaluated across 40 plastic surgery-focused question-answering tasks (20 single-hop, 20 multi-hop), spanning multiple subspecialties (4800 total evaluations). Configurations varied by base LLM (Phi-3-mini-128k-instruct vs. BioMistral-7B), embedding model, database size, chunk size, and query hop type. Performance was assessed using semantic similarity (Ragas) to physician-validated reference answers. Performance was analyzed using linear mixed-effects regression with query as a random effect and fixed and interaction effects selected via likelihood testing and AIC. Results: High performance was achievable using lightweight, open-source models. While BioMistral-7B had high mean sematic similarity under specific configurations (mean semantic similarity up to 0.786), Phi-3-mini-128k-instruct demonstrated more consistent performance across query complexity. Larger database sizes significantly improved semantic similarity, with the largest gain at intermediate sizes (e.g., size 5: +0.043, p = 0.001). Embedding choice had a strong effect, with bge-large-en-v1.5 improving performance (p = 0.0016) and Bio_ClinicalBERT markedly reducing it (p < 0.001). Multi-hop queries substantially reduced performance (p < 0.001), though this effect was attenuated for Phi-3-mini-128k-instruct via a strong model × hop-type interaction (p < 0.001). Conclusions: RAG systems for plastic surgery do not require large proprietary models, as performance depends on configuration choices and interaction effects rather than isolated components. With advancements, predictive modeling may enable resource-efficient, safe deployment of clinical RAG systems.