Sujin Kim

Ph.D.

University of Kentucky

Associate Professor

NLST (Learn more about this study)

NLST-1200

Feb 29, 2024

Feasibility of colorectal and lung cancer screening using machine learning towards effective screening and early detection

This project focuses on applying inverse reinforcement learning (IRL) algorithms to optimize screening and early detection protocols for colorectal cancer (CRC) and lung cancer (LC), leveraging data from the Truven MarketScan and the National Lung Screening Trial (NLST), respectively. The initiative is driven by the critical need to improve screening strategies, enhancing early detection rates and subsequently reducing cancer mortality. For lung cancer, the project utilizes data from the NLST, which demonstrated the efficacy of low-dose computed tomography (LDCT) in reducing lung cancer mortality among high-risk populations. This dataset provides a solid foundation for modeling lung cancer screening protocols within an RL framework, aiming to refine decision-making processes to pinpoint the most effective screening intervals and methodologies. In the realm of CRC, the Truven MarketScan database offers extensive real-world patient data, including screening practices and outcomes. This information is crucial for identifying prevailing patterns and optimizing screening guidelines through RL algorithms. The objective is to simulate various screening scenarios to determine the most beneficial strategies that encourage early detection and treatment. The methodological backbone of the project is the partially observable Markov decision process (POMDP), a sophisticated RL model adept at navigating the complexities and uncertainties inherent in cancer screening. This approach allows for the simulation of patient pathways and the iterative learning of optimal screening strategies, informed by comprehensive data from both the NLST and Truven MarketScan datasets. By integrating these datasets within our RL models, we aim to address the unique challenges posed by CRC and lung cancer screening. For lung cancer, leveraging NLST data enables the algorithm to learn from established clinical trial outcomes, guiding the model towards strategies that have a proven impact on mortality reduction. Meanwhile, the CRC screening model benefits from the diverse, real-world insights provided by the Truven MarketScan data, facilitating the development of tailored screening recommendations that can adapt to varying patient profiles and risk factors. The ultimate goal of this project is to harness the potential of RL to revolutionize cancer screening protocols. By developing models that can dynamically adjust screening recommendations based on evolving data and patient-specific factors, we aim to significantly improve the efficacy of early cancer detection. This not only has the potential to enhance patient outcomes but also to contribute to the broader public health goal of reducing the burden of these prevalent cancers. In summary, this project represents a pioneering effort to apply advanced RL techniques to the critical field of cancer screening. Through the strategic use of key datasets like the NLST for lung cancer and the Truven MarketScan for CRC, we are poised to develop innovative screening models that promise to improve early detection rates, optimize healthcare resources, and ultimately save lives.

The project aims to achieve the following specific outcomes:

1. Develop tailored RL models for CRC and lung cancer screening that accurately simulate the decision-making process, integrating comprehensive data from the Truven MarketScan and NLST datasets, respectively.
2. Identify optimal screening strategies that are personalized based on patient demographics, history, and risk factors, thereby enhancing early detection rates and reducing mortality.
3. Create a scalable and adaptable framework that can be applied to other cancers and screening processes, demonstrating the broad potential of RL in healthcare.

By integrating detailed clinical trial data and real-world patient interactions within an RL framework, this project aspires to not only advance the field of cancer screening but also to provide a tangible tool that healthcare providers can use to make informed, data-driven screening decisions. Ultimately, the goal is to significantly improve patient outcomes through the early detection and treatment of CRC and lung cancer, showcasing the transformative potential of applying advanced machine learning techniques to complex healthcare challenges.