MyGeneRisk Colon: A Web-Based Tool for Personalized Colorectal Cancer Risk Prediction Based on Genetics and Lifestyle

This paper introduces MyGeneRisk Colon, a publicly accessible web-based tool that enhances colorectal cancer risk prediction by integrating polygenic risk scores with demographic, family history, and lifestyle factors to provide personalized risk profiles and prevention education.

The Gist

Imagine you are trying to predict the weather for your hometown. You could just look at the calendar and guess, "It's usually rainy in April." That's like the old way of assessing cancer risk: looking at general statistics and family history. But what if you had a super-accurate weather app that combined your specific location, the current wind patterns, and a detailed forecast of your personal habits?

That is exactly what this paper introduces: MyGeneRisk Colon. It is a free, secure website that acts like a "personalized weather app" for your risk of developing colorectal cancer (cancer of the colon or rectum).

Here is the story of how they built it, broken down into simple concepts:

1. The Problem: The "One-Size-Fits-All" Map Was Failing

For a long time, doctors told people, "If you are over 45, get screened." But cancer is starting to happen in younger people, and not everyone has the same risk.

• The Old Way: Some tools only looked at your lifestyle (do you smoke? do you eat red meat?). Others only looked at your genes.
• The Gap: Most tools were built using data mostly from white people, so they didn't work well for everyone. Also, they were often too complicated for regular people to use.

2. The Solution: A "Super-Recipe" for Risk

The researchers wanted to build a tool that mixes two main ingredients to get the most accurate prediction possible: Genetics and Lifestyle.

• The Genetic Ingredient (The Polygenic Risk Score): Think of your DNA as a massive library of tiny instruction books. Scientists found over 200 specific "typos" in these books that make some people more likely to get colon cancer. They created a score (like a credit score) that adds up all these tiny risks.
  • The Innovation: This tool can take your raw genetic data (which you might have from a home test like 23andMe or Ancestry) and instantly calculate this score, even if you aren't part of a big research study.
• The Lifestyle Ingredient: This includes things you can change: how much red meat you eat, if you smoke, your weight, and whether you've had a colonoscopy.
• The Secret Sauce: They didn't just guess these numbers. They combined data from 16 different studies involving over 670,000 people from all over the world (including Black, Hispanic, Asian, and White populations). This makes the "recipe" much more accurate for everyone, not just one group.

3. How It Works: The "Privacy-First" Kitchen

One of the biggest fears people have is, "If I upload my DNA, will someone steal it?" The creators of MyGeneRisk Colon built a "kitchen" designed so that no one ever tastes your food except you.

• The Process: You upload your genetic file and answer a few questions about your life.
• The Magic Trick: The computer does the heavy lifting in the cloud (on Amazon's servers), but it does it in a way that it never saves your data. It's like a chef who cooks your meal in a kitchen that is demolished the second the meal is served.
• The Result: The final calculation happens right on your computer screen. The website never keeps a copy of your DNA or your answers. It's a "ghost kitchen" for your health data.

4. Building It with the Community: The "Town Hall"

The scientists didn't just build this in a lab and hope people liked it. They held a "Town Hall" with a Community Advisory Panel.

• They asked regular people: "Is this too confusing?" "Is the language too scary?" "How long does it take?"
• The Feedback: The community said, "Make it take less than 10 minutes," "Give us clear steps on what to do next," and "Tell us if we need to see a doctor."
• The final tool reflects this feedback, offering clear, actionable advice like "Talk to your doctor about screening" or "Eat more fruit," rather than just giving a scary number.

5. What You Get: Your Personal Report

When you finish, the tool gives you a report that looks at your risk in three ways:

1. Based on your genes alone.
2. Based on your lifestyle alone.
3. The combined score (The Real Deal).

It tells you your risk over the next 10 years and for your whole life, comparing it to the "average person."

• Example: Maybe your genes say you are at high risk, but your healthy lifestyle says you are at low risk. The tool combines them to give you a realistic picture and tells you exactly what steps to take to lower that risk.

Why This Matters

This paper isn't just about a new website; it's about democratizing health.

• It's Fair: It works better for diverse racial and ethnic groups than previous tools.
• It's Private: It respects your data more than almost any other tool out there.
• It's Actionable: It doesn't just say "You are at risk." It says, "Here is your risk, and here is how you can lower it."

In short, MyGeneRisk Colon is a bridge. It takes complex, high-tech science (genomics) and turns it into a simple, private, and helpful conversation starter between you and your doctor, helping everyone stay ahead of colorectal cancer.

Technical Summary

1. Problem Statement

Colorectal cancer (CRC) is a leading cause of cancer-related mortality, with a concerning rise in incidence among individuals under 50. While screening (e.g., colonoscopy) is effective, traditional risk stratification often relies on family history, which is absent in over 80% of cases.

• Limitations of Current Tools: Existing risk models often rely solely on lifestyle factors or, if they include genetics, are limited to Polygenic Risk Scores (PRS) for European ancestry populations. There is a lack of accessible, user-friendly clinical tools that integrate genome-wide PRS with lifestyle factors across diverse racial and ethnic groups.
• The Gap: A significant disconnect exists between advanced scientific risk modeling and its translation into real-world, privacy-preserving, public-facing tools that can utilize Direct-to-Consumer (DTC) genetic data.

2. Methodology

The development of MyGeneRisk Colon involved three core pillars: rigorous statistical modeling, community engagement, and secure technical infrastructure.

A. Statistical Modeling & Data Sources

• Cohort Data: The underlying risk model was derived from an Individual Participant Data (IPD) meta-analysis of 16 diverse prospective cohorts (including NHS, HPFS, UK Biobank, MEC, etc.), totaling 673,265 individuals (~17,355 CRC cases).
• Lifestyle Factors: The model incorporates 18 risk factors (e.g., BMI, smoking, red meat intake, aspirin use, family history, endoscopy history).
  • Harmonization: A multistep procedure harmonized variables across studies. Continuous variables underwent sex-specific winsorization to handle outliers.
  • Analysis: A two-step meta-analysis was used:
    1. Multiple Imputation: 180 imputed datasets were generated per study to handle missing data using joint modeling.
    2. Random-Effects Meta-Analysis: Study-specific Hazard Ratios (HRs) were synthesized using multivariate random-effects models to account for between-study heterogeneity.
• Polygenic Risk Score (PRS):
  • Utilized an Asian-Euro PRS trained on ~250,000 individuals (Asian and European ancestry).
  • Validated in independent datasets across four groups: Asian, Black/African American, Hispanic, and Non-Hispanic White.
  • To avoid overfitting, the tool uses pre-calculated Odds Ratios (ORs) per standard deviation of PRS from independent validation studies (Thomas et al., 2023) rather than re-estimating them on the training data.
• Absolute Risk Calculation:
  • Attributable Risk (AR): A novel sampling-based approach was developed to estimate ARs for both lifestyle and PRS, accounting for uncertainty in HR estimates.
  • Baseline Hazard: Calculated using SEER (Surveillance, Epidemiology, and End Results) incidence rates (2015–2019), adjusted by ARs and competing risks (non-CRC mortality).
  • Output: The model calculates the probability of developing CRC over 5, 10, or 20 years, and lifetime risk, stratified by sex, age, and race/ethnicity.

B. Community Engagement

• Community Advisory Panel (CAP): An 8-member CAP representing diverse US demographics was formed to guide the tool's design.
• Feedback Integration: CAP members reviewed existing tools and provided specific recommendations, such as:
  • Including clear disclaimers (not for symptomatic individuals).
  • Using Yes/No questions where possible.
  • Ensuring the tool takes <10 minutes.
  • Providing actionable steps (e.g., "talk to a doctor").
  • Ensuring accessibility (ADA compliance) and clear visual risk communication.

C. Technical Architecture & Security

• Real-Time Imputation: A key innovation is the ability to perform single-sample genotype imputation in real-time.
  • Pipeline: Users upload raw DTC-GT files (e.g., 23andMe, Ancestry). The system validates, parses by chromosome, phases (Eagle2), and imputes (Minimac4) against the 1000 Genomes reference panel.
  • Performance: Validated against batch processing, showing high concordance (Pearson correlation > 0.98).
• Privacy-Preserving Design:
  • Serverless Computing: Uses AWS Lambda for ephemeral, stateless processing. Data is processed in temporary storage and deleted immediately after calculation.
  • Client-Side Calculation: Final PRS aggregation, risk estimation, and report rendering occur entirely within the user's browser. No genetic or lifestyle data is stored on the server after the session.
  • Encryption: All data transfer is encrypted; no backend component ever sees the complete set of genotypes.

3. Key Results

• Risk Associations: The meta-analysis confirmed that factors like obesity, smoking, red meat intake, and family history increase risk, while aspirin/NSAID use, calcium, fruit intake, and endoscopy reduce risk. Sex-specific differences were noted (e.g., post-menopausal hormone use reduces risk in females).
• PRS Performance: The PRS showed strong association with CRC across all four major racial/ethnic groups, though with attenuated effect sizes in Black/African American populations due to historical underrepresentation in GWAS and genetic diversity.
• Tool Functionality: The MyGeneRisk Colon portal (https://mygenerisk-colon.fredhutch.org/) successfully demonstrates the feasibility of:
  • Processing DTC data and calculating genome-wide PRS in 2–3 minutes.
  • Integrating this with lifestyle data to generate personalized absolute risk reports.
  • Providing risk estimates for "PRS only," "Lifestyle only," and "Combined" scenarios.

4. Significance and Contributions

• First Integrated Tool: MyGeneRisk Colon is one of the first publicly accessible tools to combine genome-wide PRS with lifestyle factors for CRC risk prediction, offering a more accurate assessment than either factor alone.
• Diversity and Generalizability: Unlike many existing PRS tools limited to European ancestry, this model is validated across Asian, Black, Hispanic, and White populations, addressing a critical equity gap in precision medicine.
• Privacy-First Innovation: The architecture proves that complex genomic computations (imputation of millions of variants) can be performed securely and privately without storing sensitive user data, setting a new standard for consumer-facing genomic tools.
• Community-Centric Design: By integrating CAP feedback, the tool bridges the gap between complex science and health literacy, ensuring the output is actionable and understandable for diverse populations.
• Clinical Translation: The tool serves as a proof-of-concept for integrating genomic risk scores into Electronic Health Records (EHR) and clinical decision support systems, potentially enabling risk-stratified screening (e.g., earlier colonoscopy for high-risk individuals).

5. Limitations

• Self-Reported Data: Lifestyle factors (diet, etc.) are subject to measurement error.
• Ancestry Gaps: While validated in four major groups, PRS accuracy may be lower in underrepresented populations (e.g., Native American, mixed ancestry) due to reference panel limitations.
• External Validation: The combined model requires further validation in broader, real-world clinical settings to confirm its impact on screening uptake and health outcomes.

In summary, MyGeneRisk Colon represents a significant step forward in translating complex polygenic and epidemiological research into a secure, equitable, and actionable public health tool for colorectal cancer prevention.