Developing a risk prediction model for future colorectal neoplasia
Principal Investigator
Name
AASMA SHAUKAT
Degrees
MD MPH
Institution
NYU Langone Health
Position Title
Professor
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCOI-2062
Initial CDAS Request Approval
Jul 14, 2026
Title
Developing a risk prediction model for future colorectal neoplasia
Summary
Colorectal cancer (CRC) is the second leading cause of cancer death in the United States. Early identification and removal of advanced colorectal neoplasia (ACRN)—including advanced adenomas, sessile serrated lesions with dysplasia, and traditional serrated adenomas—can reduce CRC incidence and mortality. To further decrease CRC risk after removing ACRN, surveillance colonoscopy is recommended at three years. However, the majority of post-polypectomy CRCs occur within three years, even though 60% of individuals have no lesions at 3 years. Thus, it is clear that the recommended surveillance interval is not appropriate for early detection of CRC in ACRN patients and that many patients undergo screening colonoscopies unnecessarily. Thus, improving methods to more accurately predict which patients will progress to CRC and assign appropriate surveillance intervals represents an important unmet need in CRC prevention.
Digital pathomics is self-supervised artificial intelligence (AI)/machine learning approaches that can construct an unbiased ‘atlas’ of histopathological changes present in precursors lesions that are associated with future risk of ACRN. Digital pathomics approaches can increase the accuracy of tumor classification and predicting clinical outcomes. In prior work, we have shown that (AI)/machine learning (ML)-based analysis of histology images can identify biologically-relevant features associated with clinical outcomes, including risk of progression and response to therapy in non-small cell lung cancer. In the context of ACRN, our AI/ML approaches have shown the ability to predict which ACRN patients will progress to CRC. Our overarching hypothesis is that AI/ML approaches can extract digital pathomic features that can predict the risk of ACRN progression. We propose to construct AI/ML models that can predict risk of ACRN progression to CRC. Demographic and lifestyle risk factors, such as smoking and high BMI are known risk factors for metachronous ACRN, but have not been combined with digital pathomics information to improve the prediction of indivduals at risk for future ACRN. We will investigate these hypotheses by evaluating ACRN from a cohort of 50-75 year old patients diagnosed with ACRN in the last 3 years, and developing and validating a risk prediction model for risk of ACRN in the future.
Aims
SA1. Characterize digital pathomic features predictive of ACRN progression in individuals with prior advanced colorectal neoplasia. We aim to use AI/ML approaches to extract features in digital images of ACRN that can predict progression to ACRN. We will study the role of lifestyle and other risk factors in pathomic features, and risk of future ACRN.
SA2. Study the differences in digital pathomic features predictive of ACRN progression in younger (50-60 years) and older individuals (61 and older). We aim to compare digital pathomics features and lifestyle risk factor differences in younger vs older individuals to characterize differences in risk prediction.
Collaborators
AASMA SHAUKAT NYU Langone Health
Sean Hacking NYU Langone Health
Aris Tsirigos NYU Langone Health