Categorizing continuous variables resulted in different predictors in a prognostic model for nonspecific neck pain

Area of Science:

• Statistics
• Medical Informatics
• Epidemiology

Background:

• Multivariable logistic regression is a key tool for developing prognostic models.
• The handling of continuous variables in these models can significantly impact results.
• Categorization strategies are often employed but their effects require careful evaluation.

Purpose of the Study:

• To assess how different methods of introducing continuous variables affect logistic regression model content and performance.
• To compare prognostic models built using continuous, categorized, and dichotomized variables.

Main Methods:

• Backward multivariable logistic regression was used for patients with nonspecific neck pain.
• Continuous variables were analyzed as continuous, categorized, and dichotomized.
• Models were compared based on content, goodness of fit, explained variation, and discriminative ability.

Main Results:

• Models using continuous variables consistently showed better performance.
• Categorization strategies led to differences in model content, with dichotomization causing the most pronounced changes.
• No significant difference in performance was observed between categorizing before or after variable selection.

Conclusions:

• Categorizing continuous variables in logistic regression models results in altered model content and diminished predictive performance.
• Maintaining variables in their continuous form is recommended for optimal prognostic model development.