Prognostic model of survival for typical bronchial carcinoid tumours: analysis of 1109 patients on behalf of the European Association of Thoracic Surgeons (ESTS) Neuroendocrine Tumours Working Group

Abstract

OBJECTIVES: Typical carcinoids (TCs) are uncommon, slow-growing neoplasms, usually with high 5-year survival rates. As these are rare tumours, their management is still based on small clinical observations and no international guidelines exist. Based on the European Society of Thoracic Surgeon Neuroendocrine Tumours Working Group (NET-WG) Database, we evaluated factors that may influence TCs mortality. METHODS: Using the NET-WG database, an analysis on TC survival was performed. Overall survival (OS) was calculated starting from the date of intervention. Predictors of OS were investigated using the Cox model with shared frailty (accounting for the within-centre correlation). Candidate predictors were: gender, age, smoking habit, tumour location, previous malignancy, Eastern Cooperative Oncology Group (ECOG) performance status (PS), pT, pN, TNM stage and tumour vascular invasion. The final model included predictors with P ≤ 0.15 after a backward selection. Missing data in the evaluated predictors were multiple-imputed and combined estimates were obtained from five imputed data sets. RESULTS: For 58 of 1167 TC patients vital status was unavailable and analyses were therefore performed on 1109 patients from 17 institutions worldwide. During a median follow-up of 50 months, 87 patients died, with a 5-year OS rate of 93.7% (95% confidence interval: 91.7-95.3). Backward selection resulted in a prediction model for mortality containing age, gender, previous malignancies, peripheral tumour, TNM stage and ECOG PS. The final model showed a good discrimination ability with a C-statistic equal to 0.836 (bootstrap optimism-corrected 0.806). CONCLUSIONS: We presented and validated a promising prognostic model for TC survival, showing good calibration and discrimination ability. Further analyses are needed and could be focused on an external validation of this model.