Availability and use of claims in general

1. Prior to COVID-19, claims data were used to explore trends in health care use and pricing, such as trends in medication dispensing and/or refills, screening, and vaccinations, and variation in prices for services, for example, by payer type.
2. During the COVID-19 PHE, claims data were most frequently used to explore changes to health care utilization, such as changes in utilization of telehealth, normal preventive services (e.g., cervical cancer screening, routine vaccination), and prescription medication fills. Most KII participants reported using claims data to identify changes in service utilization from before and during the COVID-19 PHE. Some articles identified through the e-scan also explored trends in COVID-19 treatment (e.g., proportion of patients who undergo dialysis or a tracheotomy) or trends in COVID-19 testing.
3. During the COVID-19 PHE, access to claims data largely did not change. KII participants reported no significant changes to the types of data available within their databases or to the time lag from data collection to availability.
4. The most regularly cited benefit of claims data for research was large sample sizes. Researchers noted that commercial databases typically include data from multiple payers and health systems. Other benefits are that claims data are quick and relatively cost effective to analyze, enable researchers to track individuals longitudinally, and are not reliant on self-report.

Limitations of using claims data for research

KII participants and e-scan publications noted several common limitations of claims data for public health research, including COVID-19 PHE-related research. Several individuals commented that the root cause of these limitations is that claims data are simply not designed to be used for public health surveillance or research but rather for payment.

Although generally applicable for all claims databases, these limitations can be more complex and difficult to overcome for commercial claims databases, mainly due to how they are currently compiled and made available for use during a PHE and for research.

1. Lack of data on relevant health and social factors (e.g., lack of clinical information, lack of sociodemographic data). The KIIs and e-scan found that lack of clinical and diagnostic data, particularly related to COVID-19 test results, vaccination data, and long-term COVID-19 outcome data, as well as frequent lack of other important sociodemographic data (notably, race/ethnicity and language) currently limits the utility of claims data to address PHE-related research priorities. Participants also noted that the time needed to issue and adopt new ICD diagnostic codes limited their ability to use claims data to identify COVID-19 cases during the first months of the COVID-19 PHE.
   This lack of data stems from the fact that health care claims are generated for administrative purposes. Therefore, claims typically only include data necessary for payment and were initially used only for fraud analysis.
   COVID-19 testing and vaccination data, for example, are often incomplete because these services are not consistently submitted as claims to payers. Some relevant clinical factors (e.g., stage of diagnosis, surgical complexity) are also not collected for claims databases because they are not needed for payment purposes. In addition, claims databases are not typically linked to clinical data sources that do have this data, such as EHRs. Moreover, only diagnoses associated with services are included in claims, such that undocumented conditions and conditions for which services were unreported to insurance (e.g., those managed with over-the-counter medications) are not collected.
2. Time lags. All KIIs discussed the issue of time lags between health service provision, claims data collection, and availability of claims data to researchers.
   Generally, time lags occur because processors must fully adjudicate claims before they are available within a database.
3. Coding discrepancies or inaccuracies. As with any data source, claims data include errors. The KIIs and e-scan found that claims data are subject to human error because they rely on people to report and document information. For example, inconsistent methods of classifying provider type or inaccuracies in classifying facility type can make claims data difficult to interpret.
4. Limited time, capacity, and resources to collect, access, and analyze data. State agency and HHS interviewees noted that limited staff time and expertise to analyze claims data, particularly during the COVID-19 PHE, limited their research using claims data. HHS staff also noted that costs to access commercial claims can be prohibitively high.
   Limited staffing capacity, competing priorities, staff shortages during the COVID-19 PHE, and insufficient funding were among the reasons noted by participants.
5. Inadequate representativeness of claims data. The KIIs and e-scan found that a significant, general limitation of claims databases is their representativeness. In particular, no databases include uninsured individuals. In addition, government and nongovernment claims must be linked to capture all insured individuals in a population. APCDs were created to address this issue and have a consolidated approach to accessing and using both government and nongovernment claims; however, APCDs have their own challenges and limitations. More information on APCDs can be found in other ASPE reports.²
   Participants acknowledged that commercial claims databases are a product of a multi-payer system where payers largely are siloed—each only collects data on its own enrollees or beneficiaries. In addition, claims are not in a common data model format to allow standardization, which translates into a lack of interoperability (i.e., multiple data formats, lack of a universal person-level identifier) that makes linking databases to obtain a more complete picture of insured individuals in the United States difficult.
   However, claims data are generally available at the person level and are linkable to other data sources. Insurers, commercial claims database providers, and APCDs all have person-level data with person-level identifiers. Linking claims data to other sources is sometimes possible using variables like date of service or national provider identifiers (NPIs). However, there are persistent challenges to linking data due to restrictions on access to person-level data from other databases and interoperability (i.e., common data formats) among many databases.

Availability of and access to data in claims databases

Following are key differences between claims databases explored in the KIIs and e-scan.

1. Data ownership varies by data steward. Health insurers own commercial claims data only from their own enrollees. Conversely, commercial claims databases and APCDs typically include data from multiple payers. APCDs have data from public insurers (e.g., Medicare and Medicaid) and select commercial insurers for individuals within their state. Commercial claims databases have claims data from select health insurers from states where the insurers operate. While HHS staff can readily access Medicare and Medicaid data from the Centers for Medicare & Medicaid Services (CMS), they have limited access to commercial claims, in part due to the cost of purchasing commercial claims databases.
2. The degree to which claims data are made available varies by database provider. Some insurers and all commercial claims database providers make their data available to external health organizations (e.g., federal agencies; pharmaceutical companies) at a cost. States (i.e., APCDs) sometimes charge a fee for use and tend to limit use of their data to those who can demonstrate a public benefit (e.g., academic researchers, other state agencies). HHS makes claims data (e.g., Medicare and Medicaid) available to external organizations at a cost. In addition, there are key differences in data availability among several large, multi-payer claims databases explored in the e-scan. Note that this information is limited only to what was described on database websites.
   1. Health Care Cost Institute (HCCI) data may be more transparent and accessible than other claims databases. For example, HCCI makes a data dictionary available on its website, lists which commercial insurers are included in its data, and describes its licensing process and pricing structure. However, unlike other claims databases, HCCI does not house its own, linkable nonclaims data. In addition, HCCI’s sample is smaller than other databases reviewed here.
   2. MarketScan data may be available for a longer time frame and have a shorter time lag than other claims databases. In addition, MarketScan’s claims data can be linked with its nonclaims datasets, including EHR, laboratory, and dental data. However, MarketScan does not make data variables or licensing and pricing information available on its website.
   3. IQVIA data may have a larger geographic scope and sample size than other claims databases. In addition, IQVIA claims data can be linked with its other nonclaims datasets, including EHR, pharmacy, and hospital data. However, like MarketScan, IQVIA does not make data variables or licensing and pricing information available on its website.

Limitations of using claims data for research

1. Access and quality of the various data sources differ across those who collect and use claims data for public health and public policy purposes. Medicare FFS data, the “gold standard” within HHS, only covers a specific population of adults aged 65 and older and people with disabilities, limiting the generalizability of using these data. Though federal agencies’ receipt of Medicare data can be timely and allow for near real-time analyses and action, for non-federal researchers, Medicare FFS data may become available later and only after Medicaid, Medicare Advantage, and other commercial data are made available. In addition, there may be time lags to receipt of Medicare data due to internal CMS processes which delay the distribution of these data to a large demand of external users.
2. Claims data lack information for some populations (e.g., individuals who are incarcerated, individuals without insurance) and nonclaims-based payments (e.g., episode-based payments). Moreover, state APCDs are limited in obtaining self-insured group health plans’ data, and commercial claims databases have challenges in gathering data from private health plans.

The TEP noted that there are broader issues with data availability for certain populations and health care processes across claims and other data sources. Truly addressing the limitations identified during the COVID-19 PHE would require addressing both the limitations to claims data and these limitations in health services data more broadly.

Potential Solutions

Following are several potential solutions proposed by the TEP to improve claims data access and quality for future PHEs.

1. Consider legislative changes to compel data sharing. TEP members suggested the federal government could introduce legislation that would require self-insured employer plans (currently exempt from reporting requirements) to share their claims data. Similarly, the federal government could establish a national APCD and/or compel health plans to provide data to one or more existing multi-payer data efforts. TEP members emphasized that private data providers have a financial interest in protecting their data and are only likely to share data if there is a more compelling financial incentive or if they are mandated to do so. Importantly, TEP members noted that federal agencies would need to have solutions in place to mitigate privacy concerns if data sharing were mandated.
2. Standardize data collection and reporting requirements. TEP participants stressed that a single data source is unlikely to be fully comprehensive. In addition to compelling data sharing, standardizing data collection and reporting methods—for example, by implementing a universal, patient-level identifier—would ultimately improve consistency and allow linkages across datasets.
3. Alternatively, identify approaches for linking data that do not require a universal, patient-level identifier. Identity resolution is critical for ensuring data are usable, but it must be done in a privacy-compliant manner to protect the data of individuals included in the dataset. One TEP member’s organization had implemented a workaround; an internal team of privacy analysts used probabilistic matching to link patient-level records across data sources based on patient-level characteristics (e.g., birthday), enabling them to track individuals’ service utilization and care outcomes over time with a high level of accuracy and without the need for a universal patient-level identifier.
4. Strengthen the analytic workforce. Investing in workforce development programs would familiarize more researchers with analyzing claims data and foster analytic skills before the next PHE. These investments could include training opportunities for public health departments and state agencies so they are prepared to quickly analyze data during the next PHE. Including historically black colleges and universities and other minority-serving institutions in these efforts would enable more communities to analyze and respond to a crisis with local data tailored to communities with the highest needs.