Metodi statistici per la stima di profili di rischio personalizzati basati sulla medicina di precisione del cancro nei pazienti oncologici

Precision medicine is beginning to emerge as a well-defined discipline with specific goals, areas of focus, and tailored methodology. Specifically, the primary goal is to discover treatment rules that leverage heterogeneity to improve clinical decision making in a manner that is reproducible, generalizable, and adaptable as needed. This endeavor spans a broad range of scientific areas including drug discovery, genetics/genomics, health communication, and causal inference, all in support of evidence-based, i.e., data-driven, decision making. Precision Medicine allows patients to be discriminated according to their level of risk (e.g. low or high) and identifies subgroups of patients according to their characteristics in order to assign the treatment to those who are likely to benefit. Statistics research in precision medicine is broadly focused on methodological development for estimation of and inference for treatment regimens that maximize some cumulative clinical outcome. The process for using statistical inference to establish personalized treatment strategies requires specific techniques for data-analysis that optimize the combination of competing therapies with candidate genetic features and characteristics of the patient and disease. The present dissertation focuses on the implementation and application of statistical methods for establishing optimal treatment rules for personalized medicine and discuss specific examples in various medical contexts with oncology as an emphasis. I have focused my research activity mainly in the study of the following topics. 1) Statistical methods to analyze continuous biomarkers. Several approaches were considered according to the design of study: from classical approach - median or mean value, percentiles, optimal cut-point identified by means standard receiver operating characteristic (ROC) analysis-to more complex analysis - time-dependent ROC, conditional inferential tree and subpopulation Treatment Effect Pattern (STEPP) method. 2) Statistical methods for time-to-event endpoints. Competing risks occur commonly in medical research. In the analysis of competing risks data, methods of standard survival analysis lead to incorrect and biased results. In the presence of competing risks, data analysis has to be performed including methods to calculate the cumulative incidence of an event of interest, to compare cumulative incidence curves in the presence of competing risks, and to perform competing risks regression analysis. 3) Meta-analysis for synthesizing evidence. 4) An important topic reviews to use of several statistical methods that handle the issue of treatment switching. The contribution aims at assessing tamoxifen treatment effect taking into account treatment switches, in order to provide a robust assessment of treatment effect applying causal inference methods. 5) The last topic deals with the use of population-based registry and administrative databases. The objective of this project is to develop an acceptable claims-based algorithm to identify second breast cancer events during a 10-year follow-up through a record-linkage of two data sources:the Friuli Venezia Giulia population based-cancer registry and the administrative individual-record FVG database.