Use of Fractional Polynomials in Dynamic Prediction and Assessment of Prediction Accuracy: A Simulation Study
Abstract
Non-linear relationships can be modeled by various approaches such as polynomials, splines, and fractional polynomials. Fractional polynomials (FPs) have attracted much attention in recent years, as they can provide a parsimonious function and a global interpretation. However, little is known about the influence of the choice of FP basis functions on the numeric properties of the resulting estimators. In this thesis, we implemented many combinations of FP basis functions, and tested their performance in estimating time-varying coefficients in dynamic prediction of quantile residual life and estimations of time-dependent AUCs through simulations and a real data application. Our findings suggest that an intermediate number of basis functions leads to relatively smaller bias, standard deviations, and MSEs, and therefore might be a good choice to approximate time-varying functions in models of both quantile residual life and estimation of AUCs.
Subject Area
Biostatistics
Recommended Citation
Peng, Fen, "Use of Fractional Polynomials in Dynamic Prediction and Assessment of Prediction Accuracy: A Simulation Study" (2017). Texas Medical Center Dissertations (via ProQuest). AAI10680425.
https://digitalcommons.library.tmc.edu/dissertations/AAI10680425