Related Experiment Video
Updated: Jun 5, 2026
Multidimensional Coculture System to Model Lung Squamous Carcinoma Progression
Published on: March 17, 2020
Development and Validation of Lung Cancer Risk Prediction Models in the Singapore Chinese Population
Yah Ru Juang1, Wenjia Chen1, Alvin Jing Hui Ng1
1Saw Swee Hock School of Public Health, National University of Singapore and National University Health System, Singapore, Singapore.
Background:
Despite effective screening technologies for smokers, lung cancer is often diagnosed late, especially among never-smokers who fall outside current screening guidelines. We developed 10-year risk prediction models for sex and smoking subgroups to support more precise identification of high-risk individuals in clinical settings.
Patients And Methods:
Using the Singapore Chinese Health Study (n = 63,187; recruitment period: 1993-1998), we developed a robust Cox regression model to predict 10-year lung cancer risk. Essential predictors were identified from sociodemographic, smoking-related, environmental exposure, reproductive, and dietary factors using Elastic Net regularization with 10-fold cross-validation. External validation was conducted using the Singapore Multi-Ethnic Cohort (n = 12,944; recruitment period: 2004-2010). Sex- and smoking-specific risk models were developed using the same approach.
Results:
The overall risk model demonstrated strong discrimination and calibration on external validation (C-index = 0.821; 95% CI, 0.738-0.895). Ranked by relative importance, key predictors included age, smoking exposure (pack-years, smoking status), body mass index, education, sex, environmental tobacco smoke exposure, and dietary factors (tea and fruit consumption). The ever-smoker risk model included age, body mass index, pack-years, education, and weekly fruit consumption (C-index = 0.734), while the never-smoker risk model included age, sex, environmental tobacco smoke exposure, personal cancer history, and allergic rhinitis (C-index = 0.702). Sex-specific risk models showed good discrimination (male C-index = 0.785; female C-index = 0.769), with the female risk model influenced by reproductive and medical history factors.
Conclusion:
These subgroup-specific risk models provide additional insights into lung cancer risk assessment beyond smoking history and, with further real-world evaluation, may support more tailored risk stratification in Singapore's clinical settings.