Age dependency of the polygenic risk score for colorectal cancer

Main text

To the Editor: In a recent paper, Thomas and colleagues¹ used several approaches to develop polygenic risk scores (PRSs) for colorectal cancer (CRC) from large case-control studies and measured performance when applied to a cohort study under the assumption that the risk gradient is a constant. They found that a PRS based on ∼1.2 million genetic variants had the best performance in terms of differentiating cases from controls, and there was an area under the receiver operating characteristic curve (AUC) of 0.654 (95% confidence interval [CI]: 0.639–0.669). They did not report the risk gradient for the continuous PRS. Under the assumption that the PRS is a normally distributed multiplicative risk factor, we calculated the odds ratio (OR) per standard deviation (SD) in the PRS to be 1.75 (95% CI: 1.65–1.86) (see e.g., “supplemental material” to Schmidt et al.²). This is greater (p = 0.01) than the OR per SD of 1.61 (95% CI: 1.57–1.65) for breast cancer PRS when also estimated as a constant.³

Familial risk of CRC, however, does not appear to be a constant, at least not with age. For monozygotic twin pairs, the familial risk is 6-fold if the co-twin is diagnosed at age 60 years and decreases to 3-fold if diagnosed at age 80 years; for dizygotic twin pairs, the corresponding familial risk goes from 4-fold to <2-fold, respectively.⁴ From a segregation analysis of population-based family data in which familial aggregation excluding known DNA mis-match repair genes is assumed to be attributed solely to polygenic causes, the polygenic variance decreased from 3.74 at ages <40 years to 0.8 at ages ≥70 years.⁵

There is evidence that, for CRC, the PRS risk gradient has a negative dependency on age as does the familial risk. Archambault and colleagues⁶ reported specifically that a CRC PRS based on 95 genome-wide significant variants had a greater association with CRC diagnosed before 50 years of age than when diagnosed at a later age. Without commenting in the main text, they also reported in their supplemental material that the OR per SD went progressively from 1.75 to 1.60 to 1.52 to 1.44 as age increased from the 40s to 50s to 60s to 70s and that this trend was highly significant (p = 10⁻¹⁰). On the log-OR scale, these risk gradients are 0.56, 0.47, 0.42, and 0.37, which is an almost linear decline of about 0.06 per decade.

The observation of an age-dependent risk gradient for the 95-variant PRS is important because it suggests that this PRS might be better able to rank people in terms of their CRC risk when applied at a younger age than when screening is recommended, and this has major implications for targeted screening and prevention; there appears to be no clear age threshold. On the other hand, using a PRS with an assumed age-constant risk gradient would lead to underestimates of the risk at younger ages and overestimates at older ages, which could compromise the calibration and therefore the value of the PRS for guiding screening.

Quantifying how the strengths of the measured polygenic association depends on age is also important etiologically. For breast cancer, where the familial risk is also highly dependent on age^7,8 (even after taking into account the major susceptibility genes BRCA1 and BRCA2⁹), the current best PRS is not discernibly dependent on age (p = 0.4).³ For both CRC and breast cancer, however, their familial risks not only depend on age but there is also evidence that the familial associations are not entirely due to genetic factors, at least not at all ages. Large population-complete Nordic twin studies suggest that there are non-genetic familial effects, especially at younger ages than when screening usually starts.^4,7

Missing heritability is the extent to which familial aggregation of a disease has yet to be explained by known genetic risk factors, such as mutations in susceptibility genes and PRS. The variance of the 95-variant CRC PRS goes from about 0.31 in one’s 40s to 0.14 in one’s 70s. Therefore, about one fifth of the unknown genetic variance excluding DNA mis-match repair genes⁵ is explained by that PRS, independent of age. For breast cancer, however, the proportion of unknown genetic variance excluding BRCA1 and BRCA2 explained by the PRS depends on age, increasing from 10% at in one’s 40s to 60% in one’s 70s.^3,9 How much of the missing heritability of CRC is explained by the new PRS developed by Thomas and colleagues, and at what ages, has yet to be quantified.

Therefore, clarification of age dependency of PRS for CRC is an important issue not only for clinical and population health but also etiologically. It would be of great interest to know whether the CRC PRS developed by Thomas and colleagues has age-dependent risk gradients and, if so, at what rate (e.g., is it linear on the log-OR scale?). Using the individual-level data, this could be done for females and males, separately and combined, so that the age- and sex-specific absolute risk of the PRS would be more precisely quantified and used for “prioritizing those at high risk for targeted screening or intervention and to design clinical trials to test prevention strategies in the higher-risk groups,” as discussed by Thomas and colleagues.

Declaration of interests

The authors declare no competing interests.