I conducted my first analysis of a polygenic risk score for prostate cancer in 2020, as part of my early postdoctoral research. We had just gained access to an updated version of the score, including 269 inherited genetic risk variants, and we were all quite excited about it. The results showed relatively large differences in risk of prostate cancer between categories of polygenic risk, with consistent findings across study populations and ancestry groups.

Our work with the polygenic risk score continued, and only a few years later, a new version of the score was released, now including 451 variants. This score had an even stronger association with prostate cancer risk, and we also observed a strong association with prostate cancer mortality. The absolute risk of dying from prostate cancer was clearly increased among men with a higher polygenic risk score, including before the age of 75 years. I remember looking at the data, and even without any statistical test, it was obvious that the early prostate cancer deaths were concentrated among men with either a strong family history or an elevated polygenic risk score.

However, unlike these analyses based on full study populations, case-only analyses including only prostate cancer cases never showed a strong link between the score and prostate cancer aggressiveness or survival. Men with a high polygenic risk score did not appear to have worse survival, and some analyses were almost suggestive of the opposite, indicating a small protective effect.

At least for me, this was confusing. How can a risk factor influence premature death in the population, yet show little or no association with survival time among diagnosed cases? Intuitively, one might expect that factors pushing fatal events earlier would also shorten survival after diagnosis.

The answer to this is most likely not in biology, but in study design and analytic choices. Most survival studies are conducted among cases only, so-called conditioning on having a cancer diagnosis. When we study risk factors that also affect the probability of becoming a case, such as the polygenic risk score, this conditioning can become problematic and induce bias. Fully understanding this bias is complex; my collaborator simply referred to it as that bias starting with c in our conversations (collider bias), which I think illustrates this. Differences in how prostate cancer is diagnosed and treated add further complexity, which is not always possible to control for in statistical analyses.

So, what did we find in this new analysis? Maybe this was a lucky coincidence, but simple age stratification revealed a pattern that was more consistent with earlier observations from analyses of the full study population. Our analysis showed an association between the polygenic risk score and prostate cancer survival in the middle age group, suggesting worse survival for those men at increased genetic risk.

What does this then mean? At the very least, the results challenge the notion that polygenic risk is irrelevant once prostate cancer has been diagnosed. With some exceptions, it has been historically difficult to show that inherited genetic risk variants influence prostate cancer survival. This difficulty will likely remain in future research but can hopefully at least be recognized and, to the extent possible, be addressed through improved analytical strategies.

This is important because inherited genetic markers and polygenic risk scores offer something that few other biomarkers can compete with, namely an estimate of risk from an early age, long before onset of potential aggressive disease. This gives a long window for intervening and ultimately preventing premature deaths from prostate cancer.

Finally, joint research efforts over the past decades have identified more than 400 genetic risk variants associated with prostate cancer risk. Given that prostate cancer is one of the most heritable cancer types, it seems plausible that at least a proportion of these variants are also linked to disease aggressiveness and progression.