Ridit scores for analysis and interpretation of ordinal pain data.

Pain studies frequently generate outcome data in the form of ordered categorical responses. Measurement properties of such ordinal variables do not permit straightforward numeric treatment, yet purely categorical (nominal) procedures fail to incorporate the ordered nature of the pain response. This paper describes analysis of ordered categorical data by means of ridit analysis, a simple, yet statistically valid and efficient, technique that yields a highly intuitive interpretation of a treatment effect. The mean ridit comparison of two treatments estimates the probability that a randomly-selected patient from one will have higher pain than a randomly-selected patient from the other. Ridit analysis is illustrated with two examples from a clinical trial involving patients with painful bony metastases subsequent to hormone-refractory prostate cancer.