Analysis of Clinical Data Using Imprecise Prior Probabilities

A new method based on the theory of imprecise probabilities, for analysing clinical data in the form of a contingency table, is applied to a well-known set of statistical data from randomized clinical trials of two treatments for severe cardiorespiratory failure in newborn babies.

Abstract

This paper describes a new method based on the theory of imprecise probabilities, for analysing clinical data in the form of a contingency table. The method is applied to a well-known set of statistical data from randomized clinical trials of two treatments for severe cardiorespiratory failure in newborn babies. Two problems are distinguished. The inference problem is to draw conclusions about which treatment is more effective. The decision problem is to determine whether one treatment should be preferred to another for the next patient, or whether it is ethical to select the treatment by randomization. The two problems are analysed using three possible models for prior ignorance about the statistical parameters, and one of the models is modified to take account of earlier clinical data. In this example the four models produce essentially the same conclusions.