"Oh, don't be such a crybaby! Walk it off!" (says the harsh coach. And he's right!)

If this story doesn't make the point about the elusive nature of genetic causation, what will?

About 10,000 cases of breast and bowel cancer could be prevented each year in the UK if people did more brisk walking, claim experts.

Cancer is a disease of misbehaving cells.  But 'walking' can reduce the risk.  How can that be?

A tumor and its metatastic dissemination is a clone of cells, often if not usually descended from a single misbehaving cell whose misbehavior is inherited by the daughter cells it produces when it divides.  The initiating event may be a mutation in the usual sense (change in DNA sequence in some body cell, such as in the lung or intestine), or it may be due to some other trigger. But it is generally a clonal trigger.

So how can an organism-level exposure, such as to walking or other exercise, have anything to do with what happens to one your billions of cells?  The answer is unclear, but what is clear is that it works through confounding.  That is, it's not exercise itself, but something correlated with it.  In this particular case, if the story is correct, the confounder is obesity.  That leads to chemical agents circulating (related to normal body steroids, perhaps) that can modify cell behavior.  The less obese you are, goes the argument, the less of these mutagens are flying around your bod and bumping into DNA here and there, raising the odds the wrong gene in the wrong cell will be damaged.  Or something like that.

If this explanation is right, obesity is itself a correlate or confounder of the circulating molecules.  Thus walking is correlated with obesity which is correlated with the actual causative agent.  If so, it is easy to see why it is so damned difficult to find a 'cause' of cancer.  Genetic susceptibility variants that are measured are confounded by these other factors.  If undetected, the variant itself becomes an additional confounder.

The confounders are not perfectly associated with each other.  Not all who exercise are slim, not all slim people exercise, not all of either need have any given genetic variant, countless variants across the genome could individually contribute, and what about smoking, diet, and other factors, measured and unmeasured?

If there are too many such factors, even if all are measured, sample sizes for detecting their individual or combinational effects may be prohibitively large.

Welcome to the world of complexity!