Dichotomizing continuous predictors in multiple regression: a bad idea

By Patrick Royston, Douglas G. Altman, Willi Sauerbrei. Statistics in Medicine, 2006

2,276 citations recorded by the source corpus.

Subfields: Epidemiology, Hepatology, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/d04da496ec5f80ed20919436801a2d623db12ece

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Epidemiology

In Hepatology

In Statistics and Probability

Citations in the corpus

Prognosis and prognostic research: Developing a prognostic model (2009, background): “Keeping variables continuous is preferable since much more predictive information is retained.”
Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. (2007, neutral): “wer [87] especially when dichotomization is used [82,85,88].”
Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies: The CHARMS Checklist (2014, methodology): “, using restricted cubic splines or fractional polynomials) are indicated [45,51,58,93,94,96].”
A researcher's guide to regression, discretization, and median splits of continuous variables (2015, background): “(2015), Irwin and McClelland (2003), Lemon (2009), Maxwell and Delaney ), Royston et al. (2006), Streiner (2002), Vargha et al. (1996) 09), Hunter and Schmidt (1990), Lagakos (1988), Morgan and Elashoff (1986) Callum et al. (2002) collinearity is absent should not be accepted as sufficient…”
External validation of a Cox prognostic model: principles and methods (2013, background): “There is no necessity to dichotomise continuous variables—indeed, it wastes information [21].”
Reporting Recommendations for Tumor Marker Prognostic Studies (REMARK): Explanation and Elaboration (2012, background): “Although categorization is ubiquitous in cancer studies, there are some major concerns about that approach, as discussed in Box 4 [153].”
Clinical drug response can be predicted using baseline gene expression levels and in vitro drug sensitivity in cell lines (2014, background): “While dichotomizing clinical response data is not ideal [33], the original clinical data is again not available, thus this is the only means of directly comparing the predictions.”
A prospective study of the factors affecting outcomes of nonsurgical root canal treatment: part 1: periapical health. (2011, positive): “(Royston et al. 2006).”
Dose‐response analyses using restricted cubic spline functions in public health research (2010, background): “Dose-response analyses are often carried out by categorizing the exposure, although several limitations have already been pointed out [4--6], including loss of information and reduction in power, and modeling a continuous dose–response association with a biologically implausible step function [7].”
Towards better clinical prediction models: seven steps for development and an ABCD for validation. (2014, background): “above a certain cut-off) in the model development phase, since valuable information is lost.(27) In a later phase, we may search for a user-friendly format of the prediction model and categorize some predictors (e.”
The cost of dichotomising continuous variables (2006, background): “Nevertheless, all these approaches are preferable to performing several analyses and choosing that which gives the most convincing result. Use of this so called “optimal” cutpoint (usually that giving the minimum P value) runs a high risk of a spuriously significant result; the difference in the out”