Semester 1

Date 
Presenter 
Title 
Abstract 
17 Feb 
Michael Fitzgerald
The University of Newcastle

Hot topics in clinical trials statistics 
[Click to show or hide] 


The clinical trials industry is constantly reviewing and revising the incumbent standard methods however new ideas often take a while to gain traction, probably due to both the risk associated and a feel of "we've always done it this way" (a phrase which makes me shudder every time I hear it), an example is that adaptive Bayesian study designs have only become common place in the last couple of years even though they were shown to be superior over ten years ago.
A couple of the latest hot topics in clinic trials are:
 the proper use of estimands, and
risk based management of data
Estimands are nothing new to statisticians but the relationship between study objectives (estimands) and analysis methods (estimates) is under increased scrutiny by regulators and the guidance body ICH have started a working group on this topic. The challenge is getting some common language and understanding of the concepts across multidisciplinary teams.
Risk based management of data is a departure from traditional data validation methods which primarily look for transcription errors. With central data management we can used statistical methods to identify a range of possible data issues, from naive errors to fraud. This provides statisticians in clinical trials to apply a wide variety of data analysis methods and also an opportunity to learn methods from other industries where similar methods are already applied regularly.

10 Mar 
Dr Robert King
The University of Newcastle

Exploring the Quantile Statistical Universe 
[Click to show or hide] 


[Technical series]
The standard way of specifying statistical distributions is via their probability mass or probability density function, or via their distribution function, F(x) = P(X ≤ x).
This seminar is a guide to the distributions that can be defined by the inverse of the distribution function, the quantile function, Q(p), where Q(p)=x where x is the smallest x such that P(X ≤ x)=p.
There are theoretical and computational advantages to the quantile approach, which I will illustrate by introducing a number of different distributions, including the generalised lambda, the quantile defined skew logistic, and the logisticexponential.

12 May 
Prof Jimmy Efird
The University of Newcastle

Risk Stratification using a Weibull proportionalhazards model 
[Click to show or hide] 


Although commonly used to assess disease risk, adjusted survival estimates generated from a Cox proportionalhazards model often deviate significantly from their true values with increasing time from exposure. In this presentation, we present a family of survival models based on the generalized Weibull distribution for performing risk stratification of time to event data. Given a set of covariates, we derive the corresponding density, cumulative distribution, hazard, and survival function for the Weibull proportionalhazards model. Additionally, we show how to compute the Hessian and Fisher information function corresponding to model beta coefficients and how to test the null hypothesis that beta=0. With new methods appearing in the literature for efficiently imputing censored time to event data, the parametric Weibull proportionhazard model presents a practical alternative approach for analyzing such data.
Key Words: Risk exposure, eventtime models, generalized Weibull distribution

22 Jun 
Ines Wilms
KU Leuven

Sparse cointegration 
[Click to show or hide] 


Cointegration analysis is used to estimate the longrun equilibrium relations between several time series. The coefficients of these longrun equilibrium relations are the cointegrating vectors. We provide a sparse estimator of the cointegrating vectors. Sparsity means that some elements of the cointegrating vectors are estimated as exactly zero. The sparse estimator is applicable in highdimensional settings, where the time series length is short relative to the number of time series. Our method achieves better estimation accuracy than the traditional Johansen method in sparse and/or highdimensional settings. We use the sparse method for interest rate growth forecasting and consumption growth forecasting. We show that forecast performance can be improved by sparsely estimating the cointegrating vectors.
