Alternative Approaches to Defining COVID-Related Inpatient Encounters in IBM® Marketscan® Data

Background: In late 2021, we observed that zero records in the IBM® MarketScan® Research Databases had an ICD-10-CM code for COVID-19 (U07.1 “COVID-19, virus identified”) as either a principal diagnosis or in the first diagnosis code position on inpatient claims. Given the significant morbidity and mortality related to the pandemic, it is highly unlikely that there were no COVID-related hospitalizations among > 40 million persons in one year. The missingness for this code has been identified as a third-party software anomaly, but a full correction is estimated to take months or years.

Objectives: Considering the potential of these data to answer questions related to COVID-19, we sought to identify alternative approaches for capturing COVID-related admissions in MarketScan.

Methods: We defined persons of any age as having a COVID-related inpatient stay in MarketScan Commercial Claims and Encounters and Medicare Supplemental and Coordination of Care Databases from April 2020 through June 2021 in three ways. First, an inpatient encounter with a second position diagnosis of U07.1. Second, an inpatient encounter with a principal diagnosis of ICD-10-CM B97.29 (“Other coronavirus as the cause of diseases classified elsewhere”). Third, an inpatient encounter with both a principal diagnosis of B97.29 and U07.1 in any other position.
To benchmark the validity, we compared the proportion of total inpatient encounters related to COVID-19 in MarketScan to three other administrative claims databases from commercial insurance sources: PINC AI™ Healthcare Data (PHD) Special Release COVID-19 edition, IQVIA PharMetrics® Plus and Optum Clinformatics® Data Mart.

Results: Among a total of 1,467,411 inpatient encounters, there were 63,367 with a second position diagnosis of U07.1, 61,884 had a principal diagnosis of B97.29, and 61,879 had both a principal diagnosis of B97.29 and U07.1 in any other position; these groups are not mutually exclusive. There were clear temporal patterns that mirrored national pandemic trends, with a rise in cases during Summer 2020, followed by a dramatic rise in Fall 2020/Winter 2021. The proportion of inpatient encounters classified as COVID-related was similar for each strategy considered, and closely resembled monthly trends in other datasets. For example, in July 2020, we identified 3.8% of admissions as COVID-related using each MarketScan search strategy, as compared to 4.4% in PHD, 4.1% in PharMetrics and 5.6% in Clinformatics Data Mart.

Conclusions: A stricter definition may consider a principal diagnosis of B97.29 with U07.1 in any diagnosis position; a broader definition would use U07.1 in any position. Users of these methods will need to balance the need for sensitivity to detect true cases with specificity to correctly rule out non-cases.