Good morning! I am using a statewide dataset of adult inpatients with COVID-19 (n=18k+) to assess racial mortality disparities that may exist, particularly among American Indians, despite comparable comorbidity experiences. A weighted Elixhauser comorbidity index (ECI) score has been calculated for each patient based on the van Walraven algorithm developed in 2009. I then created the categorical variable ECIgroup to stratify patients into 1 of 5 groups (0-4) based on their weighted ECI score. Those in ECIgroup==0 have ECI scores ≤0 (and therefore lowest mortality risk), while those in ECIgroup==4 have the highest ECI scores (≥16) and therefore greatest mortality risk. I want to present both the adjusted predicted probabilities and adjusted odds ratios of inpatient mortality for American Indians, Whites, and Blacks by ECI group. American Indians have a race code of '3' in the my dataset and serve as my reference group; the reference group for ECIgroup==0.

I ran the below logistic regression command to include the binary 'dead' outcome variable with categorical race and ECIgroup interaction, along with age (continuous) and male (binary 0/1) predictors. I am unsure of how to interpret the attached results, however. Since ECIgroup==0 is my reference group, would the adjusted OR for a white patient in ECIgroup==0 be 0.37? Then for a white patient in ECIgroup==1, OR=0.67, ECIgroup==2, OR=0.57, etc.? The interpretation with the interaction has me a bit unsure.

I believe the -margins- command following this logistic regression will give me the adjusted predicted probabilities I need.

Thank you in advance for any guidance! I am unfortunately green.

Code:
logistic dead b3.race##ECIgroup age male if race<=3
Code:
margins race, at(ECIgroup=(0 1 2 3 4))
Array