Matching using the rangejoin command vs. other methods

[jennifer Thornton]

I am using STATA Version 17.0 to assess survival time in a cohort of over 100K patients with a specific type of cancer. I am comparing two groups of patients with different insurance coverage.

I successfully used the rangejoin command to "match" the patients with two different types of insurance coverage on potentially confounding factors such as age of diagnosis (within 5 years) and sex (male or female). By successful, I mean that I was able to match every member of the group with one type of insurance coverage to up to four members of the group with another type of insurance coverage and look at their outcomes such as survival time.

I have since had some questions from my team about alternative methods of matching the two groups of patients such as propensity score matching using the psmatch2 routine.

Can anyone guide me to some literature on the relative merits of matching by propensity score instead of the way that I did it?

Thank you for your time, and if there is any additional information that I should provide please let me know.

[Andrew Musau]

See https://scholarworks.umass.edu/cgi/v...0&context=pare and the references therein.

[Rich Goldstein]

a lot depends on why you want to match - if, e.g., you want to match to balance a set of covariates that are currently not balanced, then you need to check the balance after the matching - if balance is achieved, then fine, regardless of what type of matching you did; note that there are lots of methods of matching and there is no simple guide to their relative merits (that I know of anyway); so, why are you matching?

[jennifer Thornton]

Thank you for the help and guidance. I am going to check the balance after the matching and proceed from there. Thank you again!