[Prognostic value of a continuous variable and an optimal cutoff point]

To study the prognostic value of a continuous variable, cathepsin D in breast cancer, for example, it has been suggested to consider all the divisions of the population in two groups obtained by changing the cutoff level for cathepsin D, and to select the division leading to the largest difference in survival between the two groups. This method, called optimized cutoff point is statistically incorrect as currently used, and requires corrections. Moreover, the optimised cutoff point will be different in different groups of patients, and has no interpretation either biologically or statistically. Continuous prognostic factors should be studied by dividing the population in more than two groups, in order to examine the shape of the relation between the factor and survival. The limits used to define the groups should not depend on the observed survival in the groups.