IMPLEMENTAZIONE DI MODELLI PER VARIABILI RISPOSTA QUALITATIVE E MISURE PER LA VALUTAZIONE DELLA CAPACITÀ PREDITTIVA: APPLICAZIONE ALLA STADIAZIONE DELLA FIBROSI EPATICA IN PAZIENTI CON EPATITE CRONICA C.

In the context of medical, epidemiological and health organization research there are many situations in which it is necessary to develop prediction models that identify multiclass categorical variable outcomes, where the latter may be nominal or ordered. When the outcome variable is binary, consolidated modeling techniques based on logistic function are available. On the contrary, up to now, models for multiple classes outcome, taking into account ordinality or not, were developed from a theoretical point of view but not widespreadly used. Limited software availability and the difficulty in making the results suitable for pratictioners in a simple way are the main reasons for such a limited use. This has leaded to many published studies in wich a dichotomization of the outcome variable was introduced in order to allow the application of the logistic model, thus losing part of the data information and reducing the power of results. This dissertation provides an overview of the available theoretical models and shows their implementation for a problem of three-classes diagnostic classification of the severity of liver fibrosis in HCV patients. Moreover, a representation of the selected model by means of immediately readable nomograms is also provided.