Statistical methods for the identification and use of prognostic factors.

摘要: This is an expository paper which reviews the rationale for determining prognostic factors and statistical methods finding allowing such in design analysis of clinical studies. The delineation studies useful: possibly providing insight into mechanism disease; determination stratifications patients planning trials; facilitating comparison between outcomes disease different groups patients; assisting allocation treatment to individual patient permitting remedial action. A response variable a measure future health or illness its value usually dependent on one more variables. Statistical methodology reviewed case dichotomous, continuous time (with possibility censored observations) when predictor variables are discrete (or combination both types). Methods known study reviewed. These include: grouping according comparing treatments separately within each group; covariance maximum likelihood. Knowledge useful defining control group be compared with treated non-randomized Three examples designed elucidate described, breast cancer, cancer prostate Hodgkin's disease. Methodes Statistiques D'Identification et D'Utilisation des Facteurs de Pronostic Le present expose analyse la facon determiner les facteurs pronostic methodes statistiques utilisees pour definir appliquer ces dans conception l'analyse etudes cliniques. Il est utile cliniques: cela peut donner un apercu du mecanisme maladie, permettre strates malades en vue planifier essais cliniques, faciliter comparaison entre l'issue maladies divers groupes sujets, aider le traitement malade donne une action curative. Une reponse mesure l'etat sante ou maladie futur d'un sujet sa valeur depend habituellement d'une plusieurs pronostic. Les sont analysees ici cas dichotomique, continue limitee temps (avec possibilite d'observations incompletes) previsionnelles discontinues continues (ou combinent deux d'application connues egalement etudiees article. Elles comprennent: classification selon traitements chaque groupe, probabilite maximale. La connaissance permet groupe temoin qui sera compare au traite non aleatoires. auteurs donnent trois exemples d'etudes destinees pronostic, l'un sein, l'autre troisieme Hodgkin.