A recent publication, “From Algorithms to Clinical Utility: A Systematic Review of Individualized Risk Prediction Models for Colorectal Cancer,” in Gastrointestinal Disorders, highlights the promise and challenges of risk prediction models in advancing colorectal cancer (CRC) screening. Led by Deborah Jael Herrera as the first author, along with the ORIENT Project Team aims to bridge the gap between algorithmic precision and clinical usability.
Key Insights from the Review
- Promise of Risk Models: The review evaluated 44 risk models, identifying several with strong discriminatory accuracy (AUCs ranging from 0.57 to 0.90) and robust calibration.
- Challenges in Clinical Utility: Despite these advancements, many models lacked essential external validation or decision curve analysis. Issues like the exclusion of key populations, insufficient validation methods, and arbitrary risk thresholds were common barriers to real-world implementation.
- Recommendations for Improvement: The study advocates for inclusive model design, rigorous validation techniques, and transparent performance metrics reporting. It highlights the need for decision curve analysis and the periodic recalibration of models using local data to enhance clinical integration.
Common CRC Risk Factors by Population
1. Asian Populations: Studies from countries such as China, Korea, Japan, and Thailand identified these most common predictors:
- Age: A universal risk factor in all models.
- Smoking History: Associated with CRC in 89% of studies.
- Sex: Differences in risk between males and females were considered in most models.
- Body Mass Index (BMI): Included in 74% of studies.
- Family History of CRC: Highlighted as a strong hereditary risk.
- Alcohol Intake and Diabetes Mellitus: Emerging predictors in select studies.
2. Caucasian Populations: Models derived from the United States, Canada, and Europe revealed a similar yet narrower set of predictors:
- Age and Sex: Found in nearly all models.
- Smoking History and Alcohol Intake: Key lifestyle-related predictors.
- Family History of CRC: Highlighted in about half of the models.
- Less Common Predictors: BMI, NSAID use, physical activity, red meat consumption, and height.
3. Multi-Ethnic or Minority Populations: Research focusing on diverse groups, including Black, Hispanic, Asian, and Afro Caribbean populations, identified these factors:
- Age, Sex, Smoking History, and BMI: Consistent across most models.
- Ethnicity: Recognized as an important variable in specific studies.
- Distinctive Protective Factors: Physical activity, aspirin/NSAID use, and vegetable consumption emerged as unique protective elements in models derived from U.S. veteran cohorts.
Why This Matters
CRC remains a leading cause of cancer-related deaths globally. Personalized risk prediction models hold significant potential to refine CRC screening, making it more targeted and accessible. This study emphasizes the importance of aligning algorithmic development with practical clinical needs to unlock their full potential for patient care.
Acknowledgment and Support
This study is part of the ORIENT project, funded by the Kom op Tegen Kanker (Fight Against Cancer) initiative. Read the full article in Gastrointestinal Disorders