The statistical analysis of immunohaematological data.

Abstract

Almost 30 years ago, I submitted one of my first research papers on immunohaematology for publication in a prestigious journal. After a few weeks, I received back the referee's comments, with many criticisms, including the fact that I had used tests. The referee had been disappointed because statistical tests are not usually performed in immunohaematological studies. For that time, the referee was actually right. Even in more recent years, however, immunohaematologists continued to apply statistics very sparingly. For example, consider a table such as table I, which is frequently encountered in immunohaematological articles. It contains rows representing antibody specificities, columns representing two or more techniques for antibody detection, which are being compared, while data are titration end-points. When there are many rows (say, more than six), a fleeting reader would greatly appreciate a summary statistic. However, I remember just a few papers in which such a statistic was presented1–3, and in only one of them1 was it the most appropriate (the geometric mean)4. Table I Summary data (reciprocals of titre end-points) from a comparison of three techniques (Aggl: standard tube test; ELAT-W: enzyme-linked antiglobulin test; ELAT-G: like ELAT-W but with gradient centrifugation instead of washing) for the indirect antiglobulin ... The analysis of immunohaematological studies is, however, straightforward, provided adequate consideration is given to the type of measurement and the distribution of the data. In most cases, non-parametric tests are appropriate. In this paper, I shall briefly review the modern software that is freely available to researchers and then show the use of tools in the most common situations. First of all, however, I shall discuss the contribution of statistics to the presentation and analysis of immunohaematological data.