Analisi di dati clinici attraverso diverse metodiche di clustering

The amount of genetic and transcriptional data that are produced and stored today in bioinformatics and computational biology is hugely increased, as well as the knowledge that might be extracted from them with the proper computational techniques. Clinical trials are also a source of data that might be analyzed in order to provide useful informations for clinicians. Data acquired in clinical practice may be suited to design indicators that help physicians to predict prognosis and classify patients as accurately as possible. In hematology and oncology, the Mantle Cell Lymphoma Prog- nostic Index (MIPI) and its variations, as the engineered MIPI (eMIPI), are widely utilized in order to divide patients with different prognosis. They are based on both clinical and biological features, like the proliferation marker ki-67. Several data mining methodologies, already applied in other fields but not usually adopted with clinical data, may also be used to divide patients on the ba- sis of clinical features relevant to cluster patients. Here, state-of-the-art clustering techniques are adopted to cluster advanced stage Mantle Cell Lymphoma (MCL) adult patients recruited by the MCL-0208 Trial, funded by the FIL (Fondazione Italiana Linfomi). Data have been used in two variants: their continuous form and a dichotomized version, in which variables have been asso- ciated to 0/1 values following different criteria (e.g. 0: values inside a clinically relevant range, 1: values outside a clinically relevant range). Three different clustering methodologies have been utilized: non negative matrix factorization, Bayesian clustering and the standard hierarchical clus- tering commonly used. The clusters of the patients have been compared among them as well as with prognostic indexes used in MCL, even utilizing cutting-edge visualization tools. Differences in clinical features among obtained clusters have been tested with the proper statistical tests for either categorical or continuous variables, both parametric and non parametric. The features se- lected should be considered to design new prognostic indexes for MCL that can be used in clinical practice.