Five myths about variable selection.

Area of Science:

• Transplantation research
• Biostatistics
• Medical data analysis

Background:

• Multivariable regression models are frequently employed in transplantation research to identify variables associated with outcomes.
• Variable selection is a common statistical method used to simplify these complex models.
• However, its application in transplantation research, particularly with limited sample sizes, presents significant challenges.

Purpose of the Study:

• To critically evaluate the application and implications of variable selection in multivariable regression models within transplantation research.
• To highlight common misconceptions leading to inappropriate use of variable selection.
• To propose expert knowledge as a viable alternative to mitigate analytical complications.

Main Methods:

• Discussion of common misconceptions surrounding variable selection.
• Analysis of the impact of variable selection on statistical inference (P-values, confidence intervals).
• Exploration of alternative approaches, emphasizing the role of expert knowledge.

Main Results:

• Variable selection, despite its aim to simplify, often complicates analysis and invalidates standard statistical inference tools.
• These issues are exacerbated in transplantation research due to small to moderate sample sizes.
• Computer-intensive stability investigations and cautious interpretation are required when variable selection is used.

Conclusions:

• Variable selection in transplantation research can lead to erroneous conclusions due to invalidated statistical inference.
• The use of expert knowledge is recommended as a more robust approach to model building and variable identification.
• Avoiding variable selection can prevent analytical complications and improve the reliability of transplantation outcome research.