Just because an association has been found to exist between two factors does not necessarily imply there is causality. For example, you could find an association between the prevalence of a disease like lung cancer in relation to exposure to a risk factors like smoking. In order to examine causality in more detail, firstly, it is useful to assess if the relationship is valid? There are three main questions to address related to chance, bias and confounding. If you then consider the relationship valid, there are numerous other factors to consider in relation to starting to make the case for causality such as temporality (exposure must precede the disease or factor being examined), strength of relationship (the stronger the association the more likely it is causal), plausibility (proper scientific explanation whey exposure might cause the disease), experimental evidence (stopping the exposure should stop the disease), biological gradient (higher exposure levels leads to a higher likelihood of disease), consistency with other research and evidence, specificity (causality argument helped if exposure is associated with a specific disease as opposed to a wide range of diseases), coherence (consistent with national history of disease, e.g. lung cancer rates higher in countries where more people smoke), and analogy (causality argument also helped if scientific mechanisms can be examined as further evidence). Thus is an association or relationship is found between two or more factors, it does not necessary mean it is causal.
For instance in relation to temporality, ex-smokers often have higher mortality rates compared to current smokers after surgery. Does this mean that quitting smoking is bad? No, it just means that their exposure occurred before they quit. The ones that quit had the poorest health prior to surgery and were more likely to quit prior to surgery following medical advice. Current smokers may have smoked less and have been in better health prior to surgery.
Also see: Bias, Confounding, Confidence Intervals, Effect Modification, Interaction and Small Numbers.