Presentation (CTI)

OBJECTIVES: Administrative claims are foundational in health economics and outcomes research studies, with payer-sourced, fully adjudicated “closed claims” datasets preferred given completeness of patient capture. Sometimes, however, only clearinghouse-sourced “open claims” data with selective capture of medical events and treatments is available. Activity criteria are often used in lieu of enrollment in such cases but may incur selection bias towards a more clinically active cohort. There is a gap in knowledge to quantify potential inadequacies in open claims and identification of corrective strategies for improving reliability of retrospective studies using open claims. This study evaluates several activity criteria for a similarly defined open claims population across common healthcare resource utilization (HCRU) metrics compared to a closed claims population.
METHODS: We conducted a retrospective, real-world analysis of administrative claims data across January-December 2023. Patients with two or more Type 2 diabetes claims, at least 30 days apart, were identified in closed claims Merative MarketScan data and open claims Symphony Health data. A random sample of closed claim patients with 12 months of continuous enrollment in 2023 were matched against open claim patients on age and gender (1:1 ratio) using propensity score matching. Both cohorts were evaluated by the number of healthcare episodes by setting. This analysis was repeated thrice with variable minimum activity thresholds. Performance is being evaluated based on the difference in endpoints.
RESULTS: A framework to catalog and evaluate each activity criterion’s performance against closed claims is being developed. Healthcare cost analysis was excluded because reimbursement amounts in open claims are mostly unadjudicated with low capture rates.
CONCLUSIONS: Initial results indicate that open and closed claims studies yield different results, even with the application of proxy criteria for enrollment. Among type 2 diabetes patients, we observe that although some options have better performance, none directly replicate the results found using closed claims.