Aims and objectives
In all areas of medicine,
the measurement of continuous variables is commonly used as an aid in making diagnostic or prognostic determinations relating to patient treatments and outcomes.
based on the value of a continuous variable,
clinical decision making often involves classifying individuals based on the value of a continuous variable as having or not having a medical condition,
or being graded into one of several categories of severity . In medical research,
Methods and materials
standard “For presentation” screening mammograms were obtained from 2,374 women attending a population-based breast screening program between 2009 and 2011.
These women were part of a case-control study that has been previously described . An automated PMD measurement software (DM-Density 2.0,
Densitas Inc.) was used to assess baseline PMD for each mammography study in 1% increments.
Less precise PMD measures were subsequently derived by rounding upward to the nearest 2%,...
Compared to the baseline PMD measure,
when rounded into broader,
less precise values,
R-squared decreased from 1.00 (1% PMD) to 0.58 (50% PMD),
while RMSE increased from 0.00 (1% PMD) to 29.10 (50% PMD). When used to evaluate cancer risk,
the AUROC curve for baseline PMD (0.554) was not significantly different from the AUROC curve for PMD measured in 5% increments (0.551,
p = 0.228) but was significantly higher than that for PMD measured in 25% increments (0.535,
p = 0.006) and that measured...
Using PMD assessed in 5% vs.
1% increments resulted in no significant loss of information in modelling breast cancer risk.
less precise density scales,
such as PMD measured in 25% or 50% increments,
lead to loss of information and lower performing cancer risk models. Breast cancer risk model performance benefits from the use of continuous or near-continuous measurements of breast density.
More course or broadly defined breast density scales,
such as the BI-RADS density scale and any...
Contact Author: Mohamed Abdolell,
Department of Diagnostic Radiology,
The cost of dichotomising continuous variables.
Br Med J.
2006;332(7549):1080. Bennette C,
Against quantiles: categorization of continuous variables in epidemiologic research,
and its discontents.
BMC Med Res Methodol.
2012;12(1):21. Cohen J.
The cost of dichotomization.
Appl Psychol Meas.
1983;7(3):249-53. Austin PC,
Inflation of the type I error rate when a continuous confounding variable is categorized in logistic regression analyses.