Methods for comparing healthcare costs between two treatment regimens in a non-CT setting?

Hi there, I’m new to the community so apologies if I’ve made any errors in asking my question here.

I’ve recently been asked to do an analysis comparing the healthcare costs between two groups of patients who differ based on the pattern they take a particular medication of interest. We’ve selected the cohort of patients from claims data which include their healthcare costs and we’ve broken down the costs by specific categories such as inpatient admissions-related costs, outpatient service costs, etc.

In measuring the healthcare costs, each claim record has a net and total payment amount associated with the claim and I summed these costs up - separately - within the different strata above (inpatient admissions, outpatient services, etc.). I examined the distribution of the costs and they are generally heavily skewed to the right with some substantially higher values that I’ve been told should not be treated as outliers whatsoever. I also assumed that no additional claims during the follow-up constituted a net or total cost of 0$ - thus the distributions are also zero-inflated.

My question is about what modeling methods exist to analyze healthcare cost outcomes and if there are any “best” practices when analyzing cost outcomes. In this case, we are interested in assessing if there are any differences in the healthcare costs between the treatment groups.

I don’t know if it’s relevant or not, but here is some additional background info:

Since the two groups have some significant differences in their baseline characteristics (identified by using standardized differences on the characteristics of interest, one at a time), we chose to use IPTW methods to control for these differences for our other outcomes analyses. The propensity for the treatment group of interest was estimated using high-dimensional propensity score estimation (Rassen et al., 2011) and after some examination, we chose to use trimming to refine the IPTW. As our primary outcome was all-cause mortality, we used this as a hard end point to follow-up. In addition, we also had some censoring end points such as loss of insurance coverage, the end of our study follow-up period, and discontinuation of the treatment of interest. The costs were considered only during the period of follow-up between the initiation of treatment and the end of follow-up defined here.

Thank you in advance for your help!