Colorectal cancer is now the leading cause of cancer death in men under 50 in the US. That alone should force a reset in how the field thinks about cancer risk.

A new Perspective in Cell from Shi, Cao, and colleagues argues that rising early-onset cancers aren’t just a trend — they’re exposing the limits of the research infrastructure used to understand cancer causation. Across 42 countries, 75% reported rising incidence in 6 cancer types among adults under 50, with average annual increases of 0.8% to 3.6% between 2003 and 2017. At the same ages, Generation X and Millennials face higher risks than earlier-born cohorts.

That pattern is consistent with life-course environmental and behavioral exposures accumulating from early life. But the authors are careful: for some cancers, rising incidence may also reflect period effects such as earlier detection. The signal is real — but the causes remain only partially understood.

What the paper makes clear is that the current epidemiologic toolkit was not built for this question. Most major cohorts begin in midlife. Health records rarely capture childhood or adolescent exposures. And as the authors note, the first several decades of life remain “largely invisible” in most datasets.

The underappreciated argument here isn’t just that we don’t know enough — it’s that we’re not measuring the right things in the right way. Only about 30% to 45% of cancers are currently attributable to established modifiable causes, even though 75% to 80% may be preventable in theory. That gap reflects not only undiscovered causes, but also exposures that are poorly captured: single time-point BMI, retrospective alcohol recall, and other blunt instruments that miss timing, intensity, and trajectory across the life course.

The authors outline three complementary frameworks to move the field forward:

• a tissue ecosystem–anchored approach to cause discovery,

• a biological state–based model for risk prediction, and

• a dynamic framework for estimating preventability.

But they are equally clear about the constraints. Cancer incidence is low at young ages, making new birth cohorts slow and expensive to yield answers. The near-term path is more pragmatic: link and harmonize what already exists — cohorts, electronic health records, and biobanks — into a connected system capable of capturing life-course exposures.

That is easier said than done. Federated data systems, privacy governance, and fragmented health records remain major barriers. And while AI-driven risk models show promise, the paper emphasizes a key limitation: in low-incidence populations, even high-performing models can have modest positive predictive value — a calibration problem that matters clinically.

“This roadmap aims to stimulate conceptual, resource, and methodological advances to accelerate cancer etiology research and prevention in the era of rising early-onset cancers,” the authors write.

The takeaway is not that early-onset cancer is a completely distinct disease — evidence for fundamentally different biology remains limited and inconsistent. It’s that it cannot be understood as simply a younger version of midlife cancer. The exposures, timing, and trajectories that shape risk are different — and the tools we’ve relied on were not designed to capture them.

If the field is going to keep up, it will need to treat early-onset cancer as a life-course epidemiologic problem — and build the infrastructure to match.

The authors declared no competing interests.

Source: Cell