A MULTIVARIATE APPROACH TO THE DESCRIPTION OF PATIENT POPULATIONS - AN EXAMPLE OF THE ANALYSIS OF THE HORMONE PROFILES OF PATIENTS WITH ADVANCED PROSTATE-CANCER

In view of the multifactorial nature of the endocrine dysfunctions tha t may develop during prostate cancer and the unsuitability of the most widely used statistical methods to study such dysfunction, we have in the present study examined the relationships among 17 biological vari ables in 26 patients with advanced prostate cancer by two complementar y multivariate methods, correspondence factorial analysis (CFA) and a hierarchical automatic classification procedure. The 17 variables incl uded 14 hormones, their precursors or metabolites [LH, FSH, estradiol (E2), testosterone, dihydrotestosterone (DHT), androstenedione (A), an drostenediol (Ediol), dehydroepiandrosterone (DHA), DHA-sulphate (DS), cortisol (CORT), 17alpha-hydroxyprogesterone (17-OH-PROG), pregnenolo ne (PREG), 17alpha-hydroxy-pregnenolone (17-OH-PREG), and androstanedi ol glucuronide (ADG)], one plasma binding protein, namely, sex-hormone -binding protein (SHBG) and two tumour markers, prostatic acid phospha tase (PAP) and prostate-specific antigen (PSA). The originality of the se multivariate methods is that they do not preselect a dependent vari able nor perform two-by-two correlations as in stepwise multiple regre ssion analysis but describe the patient population by extracting layer s of correlations (from strong to weak) from amid confounding variable s. Compared to principal component analysis which is based on covarian ce, CFA, based on the chi2-metric, enables the licit representation of both tests and patients on the same factorial maps. From an examinati on of proximity among variables, it is possible to deduce which tests are related, which groups of patients have similar hormone profiles, a nd which tests vary most in which patients. The most discriminant fact ors in this particular population of patients were PSA and PAP levels, which were, however, not strongly correlated and were apparently sele ctively associated with certain hormones. PAP seemed the more patholog ical marker; PSA was somewhat anticorrelated to the adrenal androgen ( DHA and DS) and PREG levels. The hormones with the lowest variance wer e A, Ediol and CORT reflecting their key roles in metabolism. A number of patients were hypogonadic. SHBG levels were not closely related to total T levels but anticorrelated with ADG suggesting that, in the pa tients concerned, SHBG decreases the bioavailable T fraction. There wa s no correlation between ADG and precursor hormones (PREG, DHA, DS) bu t a slight anticorrelation between these precursors and DHT. Therefore the source of ADG in these patients does not seem to be increased lev els of precursor hormones nor of DHT but increased peripheral tissue m etabolism of androgens. In future, descriptive multivariate analyses o f large patient cohorts should help to define subpopulations with dist inctive hormone profiles for prospective clinical studies.