I want to explore the relationship between several image-derived vessel metrics (continuous) and disease severity (5-class ordinal). The rationale is that the existing severity grading is somewhat arbitrary, and I'm interested in exploring new severity markers. I have two observations per individual (i.e. the data is nested). This is my current approach:
- Calculate Spearman's correlations coefficients between each metric and the ordinal severity grade
- Perform simple linear regression using
lmwith severity as the predictor and each vessel metric as the outcome (separate models)
- Perform linear mixed effects analysis using
lmer, again with severity as the predictor and each vessel metric as the outcome, and a random intercept effect for the individual identifier
- Perform multivariable linear mixed effects analysis with the addition of age, sex, and an image quality score
My question is whether it's acceptable to compare different standardized beta coefficients across models to comment on which vessel metrics best reflect disease severity. If not, I would be grateful for suggestions about how to approach this. Thank you!