The cost of dichotomising continuous variables

Altman and Royston have done a great service in discouraging the
categorisation of continuous data in health research. A particularly bad
case of this is the tendency of researchers (and indeed public statistical
agencies) to use often broad age categories for data that originally had
single year of age. In addition to the serious statistical disadvantages
of this practice, there has come to be a habit of believing that certain
age cut-off points matter. The evidence is clear that being over 65 has
little relevance in terms of disease or health care costs, and the
patterns found are of continuous twice differentiable curves. Perhaps we
should encourage new terminology, such as 'evidence destruction', in place
of the more neutral 'categorisation'.

Competing interests:
None declared