Researchers in rheumatology should avoid categorization of continuous predictor variables

Area of Science:

• Rheumatology
• Biostatistics
• Epidemiology

Background:

• Continuous predictor variables are frequently categorized in rheumatology research.
• This practice may lead to altered results in observational studies within the field.

Purpose of the Study:

• To demonstrate how categorizing continuous predictor variables impacts results in rheumatology observational studies.
• To compare findings from categorical versus continuous analyses of body mass index (BMI) changes on osteoarthritis outcomes.

Main Methods:

• Two analyses were conducted comparing percentage change in body mass index (BMI) to 26 structure and pain outcomes in knee and hip osteoarthritis.
• Analysis 1: BMI change was categorized (≥5% decrease, <5% change, ≥5% increase).
• Analysis 2: BMI change was treated as a continuous variable. Generalized estimating equations with a logistic link function were used for both.

Main Results:

• Results differed between categorical and continuous analyses for 8 of 26 outcomes (31%).
• Categorical analyses sometimes showed associations in only one direction of BMI change, unlike continuous analyses revealing bidirectional effects.
• Discrepancies included potential false positive (categorical showed association, continuous did not) and false negative (continuous showed association, categorical did not) findings.

Conclusions:

• Categorizing continuous predictor variables significantly alters analytical results and can lead to different scientific conclusions.
• Rheumatology researchers are advised to avoid categorizing continuous predictor variables to ensure accurate and reliable study findings.