Non-linear relationships in clinical research

Nicholas C Chesnaye; Merel van Diepen; Friedo Dekker; Carmine Zoccali; Kitty J Jager; Vianda S Stel

doi:10.1093/ndt/gfae187

Non-linear relationships in clinical research

Nephrol Dial Transplant. 2024 Aug 21:gfae187. doi: 10.1093/ndt/gfae187. Online ahead of print.

Authors

Nicholas C Chesnaye^{1

2}, Merel van Diepen³, Friedo Dekker³, Carmine Zoccali⁴, Kitty J Jager^{1

2}, Vianda S Stel^{1

2}

Affiliations

¹ ERA Registry, Amsterdam UMC location University of Amsterdam, Medical Informatics, Meibergdreef 9, Amsterdam, The Netherlands.
² Amsterdam Public Health Research Institute, Quality of Care, Amsterdam, The Netherlands.
³ Department of Clinical Epidemiology, Leiden University Medical Center, Leiden, The Netherlands.
⁴ Associazione Ipertensione Nefrologia Trapianto Renale (IPNET), c/o Nefrologia, Grande Ospedale Metropolitano, Reggio Calabria, Italy.

PMID: 39169463
DOI: 10.1093/ndt/gfae187

Abstract

True linear relationships are rare in clinical data. Despite this, linearity is often assumed during analyses, leading to potentially biased estimates and inaccurate conclusions. In this introductory paper, we aim to first describe - in a non-mathematical manner - how to identify non-linear relationships. Various methods are then discussed that can be applied to deal with non-linearity, including transformations, polynomials, splines, and Generalized Additive Models (GAMs), along with their strengths and weaknesses. Finally, we illustrate the use of these methods with a practical example from nephrology, providing guidance on how to report the results from non-linear relationships.

Keywords: dichotomania; generalized additive models; non-linear relationship; polynomials; splines; transformations.