The article by Miquel Porta and his colleagues1indicating that pancreatic cancer cases without activating mutations in the K-ras gene drank significantly less coffee than cases with a mutation and the accompanying editorial by Paolo Vineis2 are both important. The study by Portaet al epitomises the way epidemiology is likely to be practised in the future, and the editorial by Vineis highlights the issues that will confront investigators when molecular epidemiology becomes more mainstream. The two papers also bring forward a crucial question: “How do we assess causality in epidemiology?”.

The study by Porta and his colleagues has a considerable advantage over conventional case-control studies—it is less likely to have suffered from substantial selection or information bias.3 Unless chance has operated in a sinister way, …