Research on Risk Transfer Pathways for Lung Cancer Among Middle-Aged and Older Individuals Using Deep Reinforcement Learning: Retrospective Cohort Study

Background: Nowadays, lung cancer has a significantly high incidence rate worldwide. The mortality rate of lung cancer continues to rise; it is more common in middle-aged and older individuals and poses a great threat to human health. Objective: This study aimed to assess the lung cancer risk among middle-aged and older individuals in a timely manner and to establish an efficient pathway for the risk transfer. Methods: We proposed a deep reinforcement learning model based on deep Q-network (DQN) to explore the risk transfer pathway for lung cancer among middle-aged and older individuals. Risk stratification of lung cancer occurrence was deduced through deep neural network. The DQN model was developed using the Health and Retirement Study cohort for model training and internal validation. We also used the China Health and Retirement Longitudinal Study cohort for model external validation. Transfer simulation of multiple pathways in different cycles was calculated in a stratified risk groups-leveraged DQN model. Results: We developed and evaluated the DQN method for optimizing the risk transfer pathway among middle-aged and older individuals, with accuracy ranging from 0.917 (95% CI 0.896-0.928) to 0.949 (95% CI 0.909-0.961) and area under curve ranging from 0.906 (95% CI 0.887-0.933) to 0.927 (95% CI 0.893-0.938). External validation was conducted to assess the model's effectiveness and availability. A total of 8780 and 3763 samples from the Health and Retirement Study were used for model training and internal testing, respectively, and 16,442 samples from the China Health and Retirement Longitudinal Study were used for external validation. The results showed that DQN models illuminated the optimal risk transfer pathways for stratified risk groups. Lung cancer incidence in the high risk group had declined by 68.2% through risk transfer of model-based simulation, which had declined by 56.9% in the medium risk group. Through simulative intervention and risk transition deduced from the DQN model, lung cancer incidences of the high risk, medium risk, and low risk groups were obviously decreased. Conclusions: A DQN-based deep reinforcement learning model was proposed and validated to develop and simulate a risk transfer pathway of lung cancer among middle-aged and older individuals. Risk stratification supplied an effective foundation for lung cancer risk transition.