Developing and evaluating polygenic risk prediction models for stratified disease prevention

Abstract

Knowledge of genetics and its implications for human health is rapidly evolving in accordance with recent events, such as discoveries of large numbers of disease susceptibility loci from genome-wide association studies, the US Supreme Court ruling of the non-patentability of human genes, and the development of a regulatory framework for commercial genetic tests. In anticipation of the increasing relevance of genetic testing for the assessment of disease risks, this Review provides a summary of the methodologies used for building, evaluating and applying risk prediction models that include information from genetic testing and environmental risk factors. Potential applications of models for primary and secondary disease prevention are illustrated through several case studies, and future challenges and opportunities are discussed.

Conflict of interest statement

Competing interests statement

The authors declare no competing interests.

Figures

Figure 1. Steps for building and evaluating absolute risk models for the general population

The flowchart shows the different steps involved in building and evaluating models for the estimation of disease risks of individuals in the general population based on polygenic risk associated with common single-nucleotide polymorphisms (SNPs) and environmental risk factors. Adapted with permission from David Check, US National Institutes of Health.

Figure 2. Hypothetical distribution of absolute risk for breast cancer

Risk stratification of the population based on a hypothetical distribution of the lifetime risk of breast cancer — that is, the probability that a woman in the population is diagnosed with breast cancer between the ages of 30 and 80 years. A comprehensive model including genetic and environmental risk factors can be used to obtain estimates of the absolute risk of individuals in the population. Women may make different lifestyle choices or decisions about possible preventive interventions depending on their level of risk and their personal values. The more spread the model-based distribution of risk in the population is, the larger the number of individuals the model will be able to assign to risk categories for which the risk–benefit implications of potential interventions could be different. BRCA1, breast cancer 1; TP53, tumour suppressor p53. Adapted with permission from David Check, US National Institutes of Health.

Figure 3. Effect-size distribution for susceptibility markers and implications for risk prediction

True effect-size distribution of individual single-nucleotide polymorphisms (SNPs) and predictive power of polygenic risk scores (PRSs) under two distinct models (model 1 (panel A) and model 2 (panel B)) for the genetic architecture of breast cancer. The total heritability explained by the additive effect of SNPs from genome-wide association studies (GWAS), termed narrow-sense GWAS heritability, is assumed to be the same (sibling relative risk ~1.4) between the two models, but the number of underlying susceptibility SNPs over which the heritability is dispersed is allowed to be different. The estimates of GWAS heritability and the value of M = 4,241 as the number of underlying, independent susceptibility SNPs are obtained empirically based on an analysis of effect-size distribution using summary-level results available from the DRIVE (discovery, biology, and risk of inherited variants in breast cancer) project of the Genetic Associations and Mechanisms in Oncology (GAME-ON) Consortium. Under this model for effect-size distribution (panel A), a single-stage GWAS study including 59,000 cases and an equal number of controls is expected to lead to the discovery of the same number of susceptibility SNPs for breast cancer as has been reported to date. The value of M = 1,000 is chosen to represent a hypothetical effect-size distribution where the same degree of GWAS heritability is explained by a smaller number of SNPs (panel B). In both models, it is assumed that the PRS is defined by the additive effects of SNPs reaching genome-wide significance (P < 5 × 10−8). The different coloured lines in panel Aa and panel Ba represent the power curve for the detection of SNPs at a genome-wide significance level as a function of effect size for studies of different sample sizes (numbers of cases/number of controls; K = 1,000). The different coloured lines in panel Ab and panel Bb show the expected receiver operating characteristic curves for PRSs that were built based on studies of different sample sizes and a PRS that can be built based on infinite sample size, thus explaining GWAS heritability. Comparison of panel Ab with panel Bb illustrates that when the number of underlying susceptibility SNPs is larger, the effect sizes are smaller, the average power of detecting individual susceptibility SNPs is lower, and the discriminatory ability of PRSs improves at a slower rate with sample size. AUC, area under the curve. Adapted with permission from David Check, US National Institutes of Health.

Figure 4. Role of polygenic risk in determining absolute risk reduction for coronary heart disease and bladder cancer achievable by modification of environmental risk factors

Ten-year risk of coronary heart disease associated with statin therapy (panel a) and 30-year risk of bladder cancer associated with smoking status (panel b), across genetic risk categories defined by the polygenic risk score (PRS) distribution. Brackets indicate the absolute risk reduction (ARR) between treatment or exposure groups for subjects in different PRS categories. The tables show the ARR and relative risk reduction (RRR) between treatment or exposure groups (panel a, statin versus control group; panel b, former versus current smokers), across PRS categories. The studies illustrate that subjects at higher polygenic risk may benefit more (that is, have a greater reduction in absolute risk) from risk-reducing interventions, such as statin therapy or smoking cessation. Data in panel a from REF. . Data in panel b from REF. , American Association for Cancer Research.

Figure 5. Role of polygenic risk in determining the optimal age of initiation for screening of breast and colon cancers

Age at which the risk of developing breast cancer reaches 2.4% (panel a) or the risk of developing colon cancer reaches 0.68% (panel b) over the next 10 years, for women at different levels of polygenic risk, with and without a family history of the disease. The risk levels of 2.4% and 0.68% correspond to the average population 10-year risk of developing each disease for women at the currently recommended starting ages for screening in the countries where the original studies were conducted (that is, 47 years old for breast cancer in the United Kingdom, and 50 years old for colorectal cancer in the United States). The studies illustrate that the risk threshold for screening is reached at earlier ages for subjects with higher genetic risk, defined by the polygenic risk score (PRS) and a family history of the disease. Data in panel a from REF. . Panel b adapted with permission from REF. , Elsevier.