Background:Lung cancer is the leading cause of cancer-related mortality,and while low-dose computed tomography screening may reduce mortality,emerging prognostic models show superior discriminative efficacy compared t...Background:Lung cancer is the leading cause of cancer-related mortality,and while low-dose computed tomography screening may reduce mortality,emerging prognostic models show superior discriminative efficacy compared to age-and smoking history-based screening.However,further research is needed to assess their reliability in predicting lung cancer risk in high-risk patients.Methods:This study evaluated the predictive performance and quality of existing lung cancer prognostic models through a systematic review and meta-analysis.A comprehensive search was conducted in PubMed,Cochrane,Web of Science,CNKI,and Wanfang for articles published between January 1,2000,and February 13,2025,identifying population-basedmodels incorporating all available modeling data.Results:Among 72 analyzed studies,models were developed from Asian(28 studies,including 23 Chinese cohorts)and European/American(48 studies)populations,with only 6 focusing on nonsmokers.Twenty-one models included genetic markers,15 used clinical factors,and 40 integrated epidemiological predictors.Although 37 models underwent external validation,only 4 demonstrated minimal bias and clinical applicability.A meta-analysis of 11 repeatedly validated models revealed calibration and discrimination,though some lacked calibration data.Conclusions:Few lung cancer prognostic models exist for nonsmokers.Most models exhibit poor predictive performance in external validations,with significant bias and limited application scope.Widespread external validation,standardized model development,and reporting techniques are needed to accurately identify high-risk individuals and ensure applicability across diverse populations.展开更多
基金funded by theGuangzhou Municipal Science and Tech-nology Bureau(No.2023A03J0507).
文摘Background:Lung cancer is the leading cause of cancer-related mortality,and while low-dose computed tomography screening may reduce mortality,emerging prognostic models show superior discriminative efficacy compared to age-and smoking history-based screening.However,further research is needed to assess their reliability in predicting lung cancer risk in high-risk patients.Methods:This study evaluated the predictive performance and quality of existing lung cancer prognostic models through a systematic review and meta-analysis.A comprehensive search was conducted in PubMed,Cochrane,Web of Science,CNKI,and Wanfang for articles published between January 1,2000,and February 13,2025,identifying population-basedmodels incorporating all available modeling data.Results:Among 72 analyzed studies,models were developed from Asian(28 studies,including 23 Chinese cohorts)and European/American(48 studies)populations,with only 6 focusing on nonsmokers.Twenty-one models included genetic markers,15 used clinical factors,and 40 integrated epidemiological predictors.Although 37 models underwent external validation,only 4 demonstrated minimal bias and clinical applicability.A meta-analysis of 11 repeatedly validated models revealed calibration and discrimination,though some lacked calibration data.Conclusions:Few lung cancer prognostic models exist for nonsmokers.Most models exhibit poor predictive performance in external validations,with significant bias and limited application scope.Widespread external validation,standardized model development,and reporting techniques are needed to accurately identify high-risk individuals and ensure applicability across diverse populations.
