A carregar...
Publicação
International validation of prognostic tissue biomarkers for cholangiocarcinoma after tumor resection
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.57 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Colangiocarcinoma (CCA) é uma neoplasia agressiva das vias biliares com mau prognóstico. Mesmo após cirurgia com intenção curativa, a sobrevivência livre de recorrência (RFS) mediana é de ~2 anos e a taxa de sobrevivência a 5 anos é 7–20%. A integração de dados de expressão tumoral com variáveis clínicas surge como estratégia promissora para prever o prognóstico. Este estudo desenvolveu e validou modelos de RFS e sobrevivência global (OS) em tumores de CCA ressecados. Entre 496 genes candidatos identificados na literatura, selecionaram-se 52 por análise de sobrevivência em cinco coortes. Numa coorte multicêntrica (n=221), esses genes foram avaliados por RT-qPCR/OpenArray juntamente com variáveis clínicas. Seis modelos de RFS foram treinados; o Random Survival Forest (RSF) apresentou melhores parâmetros, contendo 13 variáveis —incluindo FSCN1, SLC2A1, LDHA, KRT19, estadio linfo-nodal TNM, idade, sexo e subtipo de CCA— e obteve C-indices de 0,75 (treino) e 0,735 (teste), estratificando doentes em três grupos de risco. Para OS, o RSF também foi superior (0,732/0,562). A análise SHAP destacou FSCN1, LDHA e SLC2A1 como preditores-chave. Single-cell RNA-sequencing confirmou a expressão destes genes em colangiócitos malignos e no microambiente tumoral. Este enquadramento reprodutível e clinicamente acessível apoia a estratificação pós-operatória e a validação prospetiva no CCA.
Cholangiocarcinoma (CCA) is an aggressive malignancy of the bile ducts with poor outcomes due to late diagnosis and limited therapeutic options. Even after curative-intent surgery, median recurrence-free survival (RFS) is ~2 years, and 5-year survival remains only 7–20%. Recent advancements suggest that integrating tumor gene expression with clinical variables may improve prognostic modeling. This study developed and validated survival models for RFS and overall survival (OS) in resected CCA. From 496 literature-based candidates, 52 genes were retained by survival analysis of transcriptomic data from five cohorts. In a multicentric cohort (n=221), RT-qPCR/OpenArray assessed these genes alongside clinical variables. Six RFS models were trained; the random survival forest (RSF) showed the best discrimination, calibration, and generalization. The optimized RSF, incorporating 13 features —such as FSCN1, SLC2A1, LDHA, KRT19, TNM lymph node stage, age, sex, and CCA subtype— achieved IPCW C-indices of 0.75 (training) and 0.735 (testing), stratifying patients into three risk groups. For OS, RSF again performed best (C-indices 0.732/0.562). SHAP analysis highlighted FSCN1, LDHA, and SLC2A1 as key predictors. Single-cell RNA-sequencing confirmed that model genes were enriched in malignant cholangiocytes and tumor microenvironment. This reproducible, clinically accessible framework supports postoperative risk stratification and prospective validation in CCA.
Cholangiocarcinoma (CCA) is an aggressive malignancy of the bile ducts with poor outcomes due to late diagnosis and limited therapeutic options. Even after curative-intent surgery, median recurrence-free survival (RFS) is ~2 years, and 5-year survival remains only 7–20%. Recent advancements suggest that integrating tumor gene expression with clinical variables may improve prognostic modeling. This study developed and validated survival models for RFS and overall survival (OS) in resected CCA. From 496 literature-based candidates, 52 genes were retained by survival analysis of transcriptomic data from five cohorts. In a multicentric cohort (n=221), RT-qPCR/OpenArray assessed these genes alongside clinical variables. Six RFS models were trained; the random survival forest (RSF) showed the best discrimination, calibration, and generalization. The optimized RSF, incorporating 13 features —such as FSCN1, SLC2A1, LDHA, KRT19, TNM lymph node stage, age, sex, and CCA subtype— achieved IPCW C-indices of 0.75 (training) and 0.735 (testing), stratifying patients into three risk groups. For OS, RSF again performed best (C-indices 0.732/0.562). SHAP analysis highlighted FSCN1, LDHA, and SLC2A1 as key predictors. Single-cell RNA-sequencing confirmed that model genes were enriched in malignant cholangiocytes and tumor microenvironment. This reproducible, clinically accessible framework supports postoperative risk stratification and prospective validation in CCA.
Descrição
Palavras-chave
Cholangiocarcinoma biomarkers machine learning overall survival prognosis recurrence-free survival