Assessment of comprehensiveness of Medicare Part B provider-level data sets within dermatology

Publicly available provider-level Medicare Part B data sets have been increasingly utilized for health services research in dermatology. Despite offering detailed insights, an important limitation of these data sets is suppression of services performed for < 11 Medicare beneficiaries at the level of each provider. This longitudinal review assesses the magnitude of this limitation by comparing service counts in provider-level Medicare data sets to those in aggregate data sets, ultimately identifying a concordance rate of 94.3% for dermatology services. However, facility-based visits (52.5%), inpatient evaluation and management visits (59.7%), phototherapy (62.9%), incision and drainage (61.1%), and nail procedures (38.0%) were less well-represented in the provider-level data sets. Provider-level data sets are most suitable for assessing dermatology services in aggregate and among specific high-volume procedure groups but alternative data sets should be considered when investigating inpatient services, facility-based services (more common in certain states), or rarely performed procedures.


Introduction
The Centers for Medicare and Medicaid Services has publicly released provider-level Part B claims data since 2014 to promote billing transparency and research innovation. These data sets have been increasingly utilized for health services research in many physician specialties. Within dermatology, there have been nearly 200 PubMed-indexed articles utilizing these data sets since 2014. Despite offering important insights, an often-highlighted shortcoming of these data sets is suppression of services performed for < 11 Medicare beneficiaries at the level of each provider. We aimed to estimate the magnitude of this limitation to better establish the comprehensiveness of these datasets when used for dermatology research.

Methods
We divided the total volume of services performed by dermatologists in the 2018-2020 Part B Provider-Level Public Use Files by the total services in the Part B Physician/Supplier Procedure Summary (PSPS) Files, which do not suppress data at the level of each provider [1,2]. This proportion was calculated across all services, by service location (e.g., facility, non-facility), and by specific service types.

Discussion
The findings suggest that despite suppression of small service volume, Medicare provider-level data sets account for most services performed by dermatologists. Use of providerlevel data sets enable unique characterization of practice patterns by geographic setting, practice type, physician characteristics, and rurality. They have recently been utilized to assess dermatologist density, biopsy and procedure rates, patch testing utilization, nail procedure utilization, dermatopathologist practice patterns, and more [3][4][5][6] These data sets appear most comprehensive in assessing dermatology services in aggregate and among specific high-volume procedure groups (e.g., biopsies, Mohs micrographic surgery).
The provider-level data sets may, however, be less suitable for assessing inpatient consults and certain procedures with lower utilization (e.g., phototherapy, incision and drainage, nail procedures). They are also less comprehensive in assessing dermatology services performed in outpatient hospital departments (facility services), which more often occur at academic centers and in certain states, including Vermont and South Dakota [7]. In assessing these procedures or practice settings, alternative data sets such as the aggregate PSPS file is likely more comprehensive despite being less granular.
There are important limitations to this analysis, including the inability to fully characterize other shortcomings of the Medicare provider-level data sets, which only include fee-for-service beneficiaries, are subject to data entry error, and do not offer outcome data [1]. Furthermore, the aggregate PSPS data set also undergoes suppression of service counts < 11, albeit at the much broader geographic level, and is, therefore, also not fully comprehensive [2]. Despite these limitations, this study should help to guide appropriate use of Medicare provider-level data sets in dermatology health services research.  Proportions in the table are calculated by dividing the total volume of services in the medicare Part B provider-level public use files by the total volume of services in the Part B physician/supplier procedure summary (PSPS) Files for each respective year. Current procedural terminology (CPT) codes were used to group services into similar categories consistent with American Academy of Dermatology definitions. Facility services primarily include those in the outpatient or inpatient hospital setting; non-facility services primarily represent those in the office setting Data availability The datasets generated during and/or analysed during the current study are available in the Centers for Medicare and Medicaid Services database at https:// data. cms. gov.

Declarations
Conflict of interest None relevant to this manuscript. Christian Gronbeck: none reported. Hao Feng: Consultant, Cytrellis Biosystems, Inc; Consultant, Soliton, Inc. The authors declare no competing interests.

IRB approval status
This study utilized publicly available online datasets and did not qualify as human subject research; therefore, institutional review board approval was not required at the University of Connecticut Health Center.