(283) Data Lag in a Large Open and Closed Claims Dataset: Navigating the Completeness-Timeliness Tradeoff

Publication Number: 957

Background: Unlike traditional adjudicated (closed) insurance claims, open claims data may be available for research purposes within weeks or even days. Thus, open claims can provide rapid insights to inform timely decision-making for public health, as demonstrated during the COVID-19 pandemic. However the tradeoff between the timeliness and completeness of these data remains to be further evaluated.

Objectives: Describe the latency and completeness of medical claims sources in a large real-world dataset, overall and stratified by claim type and care setting.

Methods: ​​Using HealthVerity claims, we described trends in data latency from Oct 2019 to Oct 2020 by comparing claims data received in Oct 2020 to an updated cut of data for the same population received 8 months later (in Jun 2021). Percent completeness was calculated at weekly intervals as the ratio of the number of patients with claims in the original (Oct 2020) data cut relative to the number of patients with claims in the updated (Jun 2021) data cut, for the same calendar week. We report completeness separately for both an open and closed claim source, overall and by care setting (inpatient, IP; emergency room, ER; non-ER outpatient, OP).

Results: The cohort included 36 million patients observable in both Oct 2020 and Jun 2021 data cuts. Percent completeness for closed vs open claims was 0% vs 46%, 75%, and 85% at 1, 2, and 4 weeks prior to the end of the study period (Oct 2020), and increased to 35% vs 94% and 69% vs 96% at 8 and 12 weeks prior. After 1 year, 98% of open claims and 92% of closed claims were available. Latency trends varied by medical setting. At 1, 2, and 4 weeks, 21%, 64%, and 77% of IP open claims and 68%, 81%, and 89% of OP open claims were available. After 8 weeks, open claims completeness was >90% for all care settings. In closed claims, 54% of IP and 73% of OP claims were available within 12 weeks with 85% of IP claims and 92% of OP claims at 24 weeks.

Conclusions: Open claims in this real-world dataset were available more quickly than closed claims, with >75% of open claims available within 2 weeks. Time lag for OP claims was shorter than IP claims for both open and closed sources, highlighting the need to consider care setting when determining fitness of claims data for specific study questions.

This study evaluated completeness by comparing claims within the same data sources at different time points, which does not account for gaps in observability (open claims lack enrollment files), nor for agreement in open vs closed claims content. Further research is needed to describe the validity of open claims on these dimensions, and to demonstrate when and how such sources can be used for public health surveillance and evidence generation.