Background: Inequities engendered from differences in socioeconomic status increase the gap between the groups and strengthen disparities. Decision trees are useful tools in predicting a binary or numeric outcome variable using several predictor variables. This study aimed to estimate the relative importance of several demographic, socioeconomic, and health status related variables such as gender, age, race, income, health insurance, over-night hospitalization, etc. in predicting unmet medical need among U.S. adults.

Method: Using pooled data from four years (2014-2017) of National Health Interview Survey (NHIS), variables with highest importance in association with unmet medical need are identified and ranked. Data were partitioned into training and testing partitions to perform CART analysis. Resulted decision trees and model fit statistics such as lift and gain charts are reported.

Results: Findings pronounce the importance of health insurance coverage. Among all the variables, lack of health insurance coverage was the main predictor, with over 90% model accuracy, of unmet medical need. Self-rated health status, family structure, and family income to poverty ratio were other potential predictors. Results from CART were constant with logistic regression analysis.

Discussion and collusion: Decision trees are powerful tools in studying and visualizing predictors of public health problems. Contrary to classic OLS, decision trees do not have major restrictive assumptions regarding the distribution of the data. Decision trees are easy to interpret and understand and provide great insight in identifying high-risk and vulnerable population groups. Lack of health insurance coverage was identified as the most important predictor of unmet medical need.