Rheumatoid arthritis presents considerable diagnostic challenges, particularly in its early stages. This study analyzes clinical variables and results from the DAS-28 and HAQ-DI instruments to identify which factors most effectively distinguish between levels of functional impairment in RA patients. Using a dataset of 124 patients from a Mexican hospital, we applied a comprehensive set of statistical methods including ANOVA, Kruskal-Wallis, mutual information analysis and mean decrease in impurity assessment, to evaluate the discriminatory power of individual features. Two dataset variants were compared: one with HAQ-DI category scores and another with responses to each individual question. Results show that certain HAQ-DI components and specific questions offer greater predictive value than general clinical variables. Notably, strength in the right hand, and HAQ-DI categories such as “Reaching” and “Other,” stand out as informative. These findings may support simplified and more accurate machine learning models for future RA diagnosis by guiding dimensionality reduction and feature selection strategies.