Biostatistics: Methodological Innovation 1


Biostatistics: Methodological Innovation 1 
Chair: Caitlin Daly (McMaster University) 
[PDF]

YING WU, University of Waterloo
A Two-Phase Model For Chronic Disease Processes Under Intermittent Inspection  [PDF]
 
A two-phase model is developed for chronic diseases with an indolent phase which is followed by a phase with more active disease resulting in progression and damage. Weakly parametric models with piecewise constant baseline hazard and rate functions are specified and an expectation-maximization algorithm is described for model fitting. Simulation studies examining the performance of the proposed model show good performance under maximum likelihood and two-stage estimation. An application to data from the motivating study of disease progression in psoriatic arthritis illustrates the procedure and identifies new human leukocyte antigens associated with the duration of the indolent phase.
 
WENYAN ZHONG, University of Calgary
Group Selection in Proportional Odds Model for Right-Censored Data  [PDF]
 
We consider the problem of variable selection for proportional odds model with right-censored data. Proportional odds model is an important alternative to Cox model when proportional hazards assumption is not satisfied. To fit the proportional odds model, we proposed an estimation by minimizing a negatively weighted partial log-likelihood subject to a group bridge penalty. This penalty encourages sparse solutions and selects significant groups and, simultaneously, individual variables within significant groups. To implement this newly proposed method, we developed a computational algorithm that has been shown efficient through simulation studies. Some theoretical properties of the proposed method were also presented. 
 
STEVE FERREIRA GUERRA, Université de Montréal
A Self-Selecting Procedure for the Optimal Discretization of a Continuous Timeline for Longitudinal Causal Inference Methods  [PDF]
 
In health care research, administrative databases have become abundantly used to conduct causal inference for drug effects. In longitudinal settings, we generally rely on a discretization of the patient timeline and bias may result when coarsening is arbitrarily chosen by the researcher. This is partially due to discarded information about time-dependent confounders. Using a Longitudinal Targeted Maximum Likelihood Estimation loss based function, we developed a cross-validation self-selecting procedure of the optimal discretization of a continuous timeline. We use a simulation study to evaluate the bias-variance tradeoff of such a method. 
 
ZELALEM NEGERI, McMaster University
Bivariate Random Effects Meta-Analysis Models for Diagnostic Test Accuracy Studies Using Arcsine-Based Transformations  [PDF]
 
The bivariate random-effects model using the logit transformation is commonly employed to synthesize the sensitivity and specificity of diagnostic test accuracy studies. We propose the arcsine square root and the Freeman-Tukey double arcsine transformation to overcome some shortcomings of the logit transformation. We evaluated the performance of the three transformations using extensive simulations in terms of bias, root mean square error and coverage probability. We varied several parameters including number of studies and sample size. The proposed transformations outperformed the logit transformation in terms of all of the performance criteria. The methods have also been illustrated using real data sets. 
 
MAXIME TURGEON, McGill University
Principal Component of Explained Variance: An Efficient and Optimal Data Dimension Reduction Framework for Association Studies  [PDF]
 
The genomics era has led to an increase in the dimensionality of the data collected. In this context, dimension-reduction techniques can be used to summarize high-dimensional signals, to further test for association with the covariates of interest. We revisit one such approach, renamed here as Principal Component of Explained Variance (PCEV). This method seeks a linear combination of outcomes by maximising the proportion of variance explained by the covariates of interest. We propose a general analytical framework that is conceptually simple and free of tuning parameters, and we provide a simple computational strategy for high-dimensional outcomes, along with testing procedures. 
 
MARIE-HÉLÈNE TOUPIN, Université Laval
On the Family of Chi-Square Copulas  [PDF]
 
The family of Chi-square copulas that has recently appeared in the literature is very attractive because it generalizes the Gaussian copula and allows for flexible modeling for high dimensional random vectors. This presentation will explore the theoretical properties and the practical usefulness of this class of dependence structures. An application of the Chi-square copulas will also be developed, namely parameter estimation.