TL;DR
Researchers have developed two novel disease risk scores (DRS and DRSA) for skin cancers, combining 32 genetic and non-genetic factors. These scores can identify individuals at significantly higher risk (up to 13-fold) and, importantly, the age-independent DRSA can predict earlier diagnosis and more severe disease, offering potential for personalized early detection strategies.
What Problem Does This Paper Address?
Skin cancer is the most common cancer globally, and while largely preventable and treatable with early detection, current screening methods often lack personalized risk assessment. Existing predictive models have limitations, including incomplete risk factor catalogs and heavy reliance on age, which hinders their utility for early detection of asymptomatic individuals. The authors aim to address this by developing comprehensive risk scores that integrate a wide array of genetic and non-genetic factors, with a particular focus on creating an age-independent score for better early detection applications.
Methods
The study employed a multi-stage approach to model development:
- Initial Model Training: Logistic regression models were trained for each of the three major skin cancers (BCC, SCC, melanoma) using demographic factors (age, sex, ancestry) and 27 previously identified risk factors from the literature.
- Incorporating Novel Phenotypes: An additional 608 phenotypic risk factors from the 23andMe database were explored. Models were built by combining the initial 20-factor model with each of these additional factors, selecting new significant factors based on stringent p-value thresholds. This resulted in models with 31 risk factors.
- Adding Polygenic Risk Score (PRS): A polygenic risk score, derived from genome-wide association studies (GWAS) conducted on the training set, was incorporated, leading to final models with 32 risk factors.
- Risk Score Construction: Two primary disease risk scores were developed:
- DRS (Disease Risk Score): This score incorporates all 32 identified risk factors.
- DRSA (Disease Risk Score, age-independent): This score excludes age as a factor, aiming to provide an age-independent risk assessment.
- Validation: The models and risk scores were validated on a separate, independent dataset.
Datasets
- Training Set: 103,008 participants from the 23andMe cohort, collected between May and September 2016. This set included individuals aged 30-90 with European ancestry.
- Validation Set: 88,924 participants from the 23andMe cohort, collected between October 2016 and June 2017. This set also included individuals aged 30-90 with European ancestry.
- Longitudinal Data: Follow-up surveys from 49,501 participants in the validation set over two years were used to assess incident cases and cancer-free status.
The study utilized a large, self-reported dataset from the 23andMe customer base, which differs from traditional clinical datasets by its scale and the nature of self-reported information.
Key Insights and Ideas
- The Power of Combined Risk Factors: The study highlights that integrating a broad spectrum of genetic and non-genetic factors significantly enhances skin cancer risk prediction, far beyond what individual factors can achieve.
- The "Long Tail" Phenomenon: Similar to PRS, the developed risk scores exhibit a "long tail" behavior, meaning a small percentage of individuals with the highest scores have a disproportionately elevated risk. This is crucial for identifying high-risk individuals for targeted interventions.
- Age-Independence is Key for Early Detection: The development of DRSA, an age-independent score, is a significant advancement. It allows for risk stratification that is not confounded by age, making it more suitable for identifying asymptomatic individuals who may benefit from earlier or more frequent screening.
- Beyond Incidence: Predicting Severity: High DRSA scores were not only associated with a higher likelihood of developing skin cancer but also with earlier diagnosis and more severe, recurrent forms of the disease. This suggests these scores could inform prognosis and treatment intensity.
- Unexpected Risk Factors: The inclusion of factors like BMI (with an inverse relationship for BCC/SCC) and a "clean desk" factor (with a slight inverse correlation) as significant predictors adds nuance to our understanding of skin cancer risk, suggesting complex interplay of lifestyle and physiological factors.
Key Results
- The primary DRS, combining 32 risk factors, achieved good predictive performance with AUCs of 0.79, 0.80, and 0.78 for BCC, SCC, and melanoma, respectively.
- Individuals in the top DRS percentile showed up to a 13-fold increase in skin cancer risk compared to those in the middle percentile.
- The DRSA, which excludes age, demonstrated strong risk enrichment in its upper tails, with a 3.7x, 5.3x, and 9.8x increase in risk for BCC, SCC, and melanoma, respectively, in the top percentile.
- High DRSA scores were associated with earlier disease diagnosis by up to 14 years.
- The PRS alone showed more modest predictive performance, with odds ratios per standard deviation increase ranging from 1.33 to 1.57.
- The study found that men have approximately 1.6x higher risk of developing skin cancer than women.
- Dysplastic moles were a strong predictor, increasing melanoma risk by 6.7x.
- BMI showed an inverse relationship with BCC and SCC risk, with higher BMI associated with lower risk.
Limitations and Future Work
- Self-Reported Data: The study relies on self-reported diagnoses, which can be subject to misclassification. While the authors performed some validation against external data, this remains a potential limitation.
- European Ancestry Focus: The models were primarily developed and validated on individuals of European ancestry. The authors acknowledge that portability to non-European ancestries needs further investigation, as genetic and non-genetic risk factors may behave differently across populations.
- PRS Limitations: The PRS variants were selected from the training set, which may have limited statistical power for detecting associations. Larger GWAS are needed to expand the set of associated variants and potentially improve PRS performance.
- Future Work: The authors suggest refining risk scores by better characterizing genetics and individual risk behaviors related to sun exposure. Further research is needed to understand prediction performance across different European subpopulations and to assess portability to non-European ancestries. Expanding the set of associated variants for PRS and investigating known pathogenic variants for skin cancers not identified by GWAS are also recommended.
Summary updated