Article Text

PDF
Demystifying “exact” logistic regression for pathologists
  1. G Venkataraman,
  2. V Ananthanarayanan
  1. Department of Pathology, Loyola University Medical Center, Maywood, IL, USA
  1. Dr Girish Venkataraman, Department of Pathology, Loyola University Medical Center, Building 110, Room 2233, 2160 South First Avenue, Maywood, IL 60153, USA; gvenkat{at}lumc.edu

Statistics from Altmetric.com

Scientific research studies in a wide variety of biological disciplines aiming at predicting a binary outcome (for example, survived: yes/no) have traditionally used logistic regression (LR) extensively.1 Ever since the widespread availability of statistical software, it has become easy for many non-statisticians, including pathologists, to perform LR for predictive studies.

On the other hand, pathology research data is frequently plagued by issues as low case numbers and issues related to distribution of variables, and the default LR methods in most statistical software may yield erroneous and unreliable results. Yet, pathologists who wish to independently perform the analysis without seeking help of biostatisticians have difficulty identifying these issues and applying the right form of LR. We present a simple real-world prostate cancer dataset from our institution, highlighting the specific use of the underused “EXACT” variant of logistic regression and the situations for which it is appropriate, in a way that pathologists can relate to.

For the benefit of pathologists unfamiliar with the basics of LR, a brief introduction is given. LR is a class of statistical modelling which aims to predict a binary outcome variable (metastases, present/absent type), using a combination of predictors that may be continuous (eg, Ki67 index), ordinal (eg, Gleason score) …

View Full Text

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.