Variance of age-specific log incidence decomposition (VALID): a unifying model of measured and unmeasured genetic and non-genetic risks

John L. Hopper, James G. Dowty, Tuong L. Nguyen, Shuai Li, Gillian S. Dite, Robert J. MacInnis, Enes Makalic, Daniel F. Schmidt, Minh Bui, Jennifer Stone, Joohon Sung, Mark A. Jenkins, Graham G. Giles, Melissa C. Southey, John D. MathewsInternational Journal of Epidemiology 2023 · DOI

TL;DR

This paper introduces VALID, a novel framework for decomposing the variance in disease risk into its genetic and non-genetic components, accounting for both measured and unmeasured factors. By unifying concepts from epidemiology and genetics, VALID provides a more comprehensive understanding of disease etiology, particularly for complex conditions like breast cancer.

What Problem Does This Paper Address?

Epidemiology has long sought to understand why individuals of the same age exhibit different risks for developing diseases. While genetic and environmental factors are known contributors, their relative importance and interplay, especially for complex traits, remain challenging to quantify. Existing models often struggle to integrate familial aggregation (risk clustering in families) with individual-level risk factors, and to disentangle the contributions of measured versus unmeasured genetic and non-genetic influences. This paper tackles this by proposing a unifying model that can partition disease risk variance into these distinct components, offering a more holistic view of disease causation.

Methods

The core of the VALID model is the concept of variance in risk, defined as the age-specific log(incidence) or logit(cumulative incidence). The key metric for quantifying a risk factor's discriminatory power is Δ = log(OPERA), where OPERA is the odds ratio per adjusted standard deviation of a risk score. This Δ represents the difference in mean risk score between cases and controls. The square of this value, Δ², then represents the variance in risk attributed to that specific risk score.

VALID builds upon Fisher's seminal work on decomposing variance in quantitative traits. Instead of relying on correlations, VALID converts familial odds ratios into variance components. A crucial aspect is the introduction of the "Unifying Equation," which relates familial risk ratios (FRR) to the risk gradient (Δ) and the correlation between relatives (r). This allows for the estimation of variance components for both genetic and non-genetic factors, including those that are unmeasured. The model also considers the "equal environments assumption" in twin studies, which can maximize the proportion of familial variance attributed to genetic factors.

Datasets

The paper utilizes data from the Nordic Twin Study of Breast Cancer for analyzing familial aggregation and estimating variance components. Additionally, it references findings from studies on breast cancer risk prediction using polygenic risk scores (PRS), mammographic density, and major susceptibility genes (e.g., BRCA1, BRCA2). No new datasets were introduced in this paper.

Key Insights and Ideas

  • Unifying Framework: VALID offers a single, coherent framework to integrate diverse risk factors (genetic, non-genetic, measured, unmeasured) and their contributions to disease risk variance.
  • Familial Risk as a Bridge: By leveraging familial odds ratios, VALID provides a direct link between observable family aggregation patterns and underlying genetic and environmental variance components.
  • Finite Genetic Variance: A key insight is that the variance attributable to genetic factors has a natural upper limit, determined by the familial odds ratio in genetically identical twin pairs. This contrasts with non-genetic factors, whose variance is theoretically unlimited.
  • Age-Dependent Risk Factors: The study highlights that the contributions of genetic and non-genetic factors to disease risk can change with age, a nuance often overlooked in simpler models.
  • Beyond "Heritability": VALID moves beyond the often-misunderstood concept of heritability (a proportion) to focus on the absolute magnitude of variance components, aligning with Fisher's original emphasis on the "magnitude of the genetic variance."

Key Results

  • The application of VALID to breast cancer suggests that while substantial genetic risk factors have been identified, a significant portion of familial risk, especially for younger women, remains unexplained by current genetic models.
  • For breast cancer, the study found that the additive genetic variance (A) and shared environmental variance (C) decrease with age. On average, about two-thirds of the declining familial variance was attributed to genetic factors.
  • The paper demonstrates that polygenic risk scores (PRS) explain a relatively small proportion of familial breast cancer risk, particularly for women diagnosed before age 50.
  • Mammographic density, when considered as a risk factor, shows a substantial familial component and can explain a significant portion of familial variance, potentially rivaling current PRS.

Limitations and Future Work

  • Unmeasured Factors: While VALID can account for unmeasured factors, precisely quantifying their contribution remains a challenge. The model relies on assumptions about the independence of risk factors when combining them.
  • Model Assumptions: The "equal environments assumption" in twin studies, while useful for maximizing genetic variance estimation, is a simplification. The paper acknowledges that more nuanced modeling of shared environmental variance is possible.
  • Data Limitations: The analysis of familial risk factors is dependent on the availability and quality of family history data and twin registries.
  • Future Directions: The authors suggest that further research could explore the application of VALID to other diseases, investigate the role of epigenetics in disease risk, and refine the modeling of non-genetic factors, particularly those influenced by lifestyle and environment. The potential for AI in discovering new risk factors and improving risk prediction is also highlighted.

Summary updated

Abstract

VALID has shown that, while substantial genetic risk factors have been discovered, much is unknown about genetic and familial aspects of breast cancer risk especially for young women, and little is known about individual-specific variance in risk.
BibTeX
@article{hopper2023variance,
  author = {John L. Hopper and James G. Dowty and Tuong L. Nguyen and Shuai Li and Gillian S. Dite and Robert J. MacInnis and Enes Makalic and Daniel F. Schmidt and Minh Bui and Jennifer Stone and Joohon Sung and Mark A. Jenkins and Graham G. Giles and Melissa C. Southey and John D. Mathews},
  title = {{Variance of age-specific log incidence decomposition (VALID): a unifying model of measured and unmeasured genetic and non-genetic risks}},
  journal = {International Journal of Epidemiology},
  year = {2023},
  doi = {10.1093/ije/dyad086},
}
Figure