OBJECTIVES: To assess the representativeness of provider encounter data from a large claims database on adults with major depression.

METHODS: We used self-reported data from National Survey on Drug Use and Health (NSDUH) from 2017-2019 to benchmark the representativeness of data on provider encounters for treatment of depression among commercially insured adults experiencing major depressive episodes in the prior year from the Healthcare Integrated Research Database® (HIRD). After applying suppression criteria, we used standardized mean differences (SDMs) to compare the similarity of the two data sources based on the annual proportion of prescription drug use and encounters with six categories of providers: primary care physicians, psychiatrists, other types of physicians, psychologists, social workers, and other non-physician mental health providers. We used thresholds familiarized by Cohen in 1988 for interpreting SMDs between non-randomized groups.

RESULTS: The HIRD and NSDUH benchmarks were highly similar based on the percentage using prescription medication (54.3% vs 48.7%, respectively; SMD=0.11) and receiving care from a physician or other non-physician provider (66.1% vs. 61.9%; SMD=.09). The similarity of the two data sources differed by provider type with SMDs ranging between highly similar for the percentage treated by psychiatrists (20.2% vs. 21.6%; SMD=.04) and similar for the percentage treated by psychologists (5.0% vs 18.7%; SMD = 0.42).

CONCLUSIONS: Despite differences between claims and survey data sources, results suggest claims data can represent patterns of depression treatment for commercially insured U.S. adults who experienced major depressive episodes. Differences between treatment from psychologists the HIRD and national benchmarks may reflect unmeasured geographic differences in scope of practice laws. While it may be challenging to identify suitable benchmarks for insured populations for many types of conditions, our analysis demonstrates that national health data can be valuable resource for benchmarking the representativeness of real-world data.