Dichotomization: 2 x 2 (x2 x 2 x 2...) categories: infinite possibilities

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Standard epidemiological practice often relies on a single measure of association, like an odds ratio (OR), from dichotomized continuous variables.
• This approach can obscure valuable information by ignoring the impact of different cutoff points for dichotomization.
• Reporting results from a single, arbitrary cutoff can misrepresent the true exposure-disease relationship.

Purpose of the Study:

• To demonstrate how varying cutoffs for continuous exposure variables affect measures of association.
• To advocate for a graphical presentation of odds ratios across a range of cutoffs.
• To illustrate the impact of different cutoffs on the body mass index-cholesterol relationship.

Main Methods:

• Utilized National Health and Nutrition Examination Survey data for analysis.
• Examined the relationship between body mass index (BMI) and high cholesterol.
• Applied multiple, theory- and data-driven cutoffs to the continuous BMI variable.

Main Results:

• Varying cutoffs for BMI yielded odds ratios ranging from 1.1 to 1.9.
• A single odds ratio can be misleading, allowing selective reporting of findings.
• An odds ratio curve provides a more nuanced view of the exposure-disease association than a single OR and confidence interval.

Conclusions:

• An odds ratio curve offers a richer understanding of the exposure-disease relationship across different cutoffs.
• This graphical approach helps assess whether study findings represent typical trends or outliers.
• It highlights the implications of changing cutoffs and random variability on observed associations.