Date on Master's Thesis/Doctoral Dissertation


Document Type

Master's Thesis

Degree Name



Bioinformatics and Biostatistics

Committee Chair

Rai, Shesh Nath


Patients--Classification--Mathematical models


The identification of subgroups in clinical studies is an important aspect of personalized medicine. In order to develop tailored therapeutics, the factors that characterize subgroups with differential prognosis, response to treatment, and incidence of adverse events or toxicities must be elucidated. We present a generalization of a statistical learning algorithm, Patient Rule Induction Method (PRIM), that is well suited for this task given a right-censored time-to-event outcome measure. This algorithm works to recursively partition a covariate space into mutually exclusive boxes that can be utilized to define subgroups. Conceptually the algorithm is similar to classification and regression trees but rather than satisfying the goal of minimizing overall prediction error, PRIM works to find the extrema of the response surface. The algorithm's performance in prognostic subgroup identification is demonstrated with simulation studies and a case study using data from the Framingham Heart Study. We find that the algorithm has much utility as it provides a set of easy to interpret rules that define subgroups with maximal (minimal) survival or differential response to an intervention as measured by a survival outcome.

Included in

Biostatistics Commons