Weak associations showing odds ratios of less than (say) three to one are often due to chance, bias, or confounding; but if they are causal and relate to common diseases and agents that are prevalent in the community they may be important for the public health and more important than strong associations with agents that occur only rarely. Their establishment requires large studies, meta-analysis of many small ones, or, better, collaborative reanalysis of many studies on a common basis. In the absence of experimental evidence, bias and confounding have to be excluded and positive evidence for causality sought using Hill's guidelines. Examples are given of weak associations that have been shown to be of practical importance (stomach cancer and smoking, lung cancer and environmental tobacco smoke, breast cancer and oral contraceptives) or are still of uncertain significance (liver cancer and smoking, and childhood leukaemia and brain cancer and exposure to extremely low frequency electromagnetic fields). Weak associations that reflect causality can often be revealed only by epidemiological investigation. They may be socially important and their establishment is a challenge that epidemiologists should accept.
J Epidemiol, 1996 ; 6 : S11-S20.