Neutral comparison studies: illustration through a benchmark > study on omics based prognostic modelling and discussion

Abstract

In the first part of my talk, I will first briefly present several  methods for deriving
prognostic models of patient outcomes using  clinical and multi-omics data. IPF-LASSO consists of
using a different penalty parameter lambda for variables of different (omics) types in > lasso
regression. If the user has prior knowledge about the importance of the different omics types or
a preference for practical reasons (e.g., because one of the omics types is much cheaper to
measure), the > recently developed "priority-lasso" may be a valuable option. I will then present
an extensive benchmark study comparing these and other  methods for multi-omics based prognostic
modelling.

This study is intended as "neutral" and will serve as case example in  the second part of my talk,
which addresses different aspects of what > may be called the "replication crisis in methodological
(statistical)  research". With a focus on methodological studies (i.e. studies > considering
statistical methods as research objects), I will discuss  problems such as fishing expeditions and
over-optimism , publication  bias, the design and necessity of neutral comparison studies, and the
importance of reporting and research synthesis.

Organizzatore: Christian Hennig