Construction and Validation of Claims‐Based Medication Regimen Complexity Index. (20th August 2020)

• Record Type: Journal Article
• Authors: Chang, H.‐Y.; Shermock, K.; Kitchen, C.; Kharrazi, H.; Weiner, J.; Bishop, M.
• Is Part Of: Health services research. Volume 55:Number 1(2020)
• Page Start: 89
• Page End: 89
• Publication Date: 2020-08-20
• Subjects: Medical care -- Periodicals; Medical care -- Evaluation -- Periodicals; Hospital care -- Periodicals; Health services administration -- Periodicals; 362
• DOI: 10.1111/1475-6773.13456
• Languages: English
• ISSNs: 0017-9124

Abstract

Abstract : Research Objective: Population health management programs often use administrative claims to measure medication refills (eg, medication possession ratio); however, claims‐based indexes lack measuring medication complexity and adherence together. Medication Regimen Complexity Index (MRCI) is a validated measure of medication complexity and adherence levels and has been originally developed using nonclaims data sources. A claims‐based version of MRCI is lacking. This study aimed to create and validate a claims‐based version of MRCI (ie, cMRCI). Study Design: Retrospective cross‐sectional analysis of administrative claims data (QuintileIMS). MRCI consists of components A, B, and C, but the input for component C was unavailable in claims. For each pharmacy record, we used the route and dose information to calculate component A while the quantity and days of supply to calculate component B. We then computed an individual's 2014 cMRCI by aggregating components A and B associated with an individual's entire pharmacy records in 2014. We also assigned an individual to one of three medication complexity levels based on the tertiles of cMRCI among enrollees with the same number of active ingredients. We evaluated the impact of adding two types of cMRCI (continuous or categorical) to three baseline models in explaining health care costs and utilization: (1) demographics only, (2) demographics plus the count of active ingredients, and (3) demographics, the count of active ingredients, and a JHU‐ACG‐based morbidity measure. We adopted generalized linear models with a log link and gamma distribution for costs (total, pharmacy, and medical costs) and logistic regression for binary utilization markers (any hospitalization, any 30‐day readmission, and any ER visit). Both costs and utilization variables were derived from the concurrent (2014) and prospective (2015) year. We reported adjusted cost ratios for costs and adjusted odds ratio for utilization. Population Studied: We included 1, 541, 873 eligible enrollees aged 1 to 63 with continuous medical and pharmacy enrollment in 2014 and 2015. Older adults (65+) were excluded due to missing Medicare claims. Principal Findings: Study subjects were 47.1% male with mean age of 42.6 years. Across all models and costs, a higher cMRCI was statistically significantly associated with higher costs both concurrently and prospectively, with the exception in concurrent total costs; such increases were the highest for pharmacy costs. Alternatively, compared to those with low cMRCI level, enrollees with high or medium cMRCI level consistently had higher costs across all models and costs. As for utilization, a higher cMRCI was statistically significantly associated with higher odds of prospective utilization but not concurrently; compared to those with low cMRCI level, enrollees with high or medium level generally had higher odds of utilization across all models, with the exception in high vs. low comparison of the concurrent utilization from the most complex model. Conclusions: cMRCI is a valid measure of medication complexity given its associations with health care costs and utilization. Categorical cMRCI may be a better format to be used than continuous cMRCI. Implications for Policy or Practice: Population health analytics using insurance claims data lack a mixed medication complexity and adherence measure. cMRCI can be used as a claims‐based medication regiment complexity index, to measure adherence and improve utilization prediction.