Completeness of Provider ID Variables in CMS TAF Files

The Provider ID and TAF APR

Provider IDs are state-reported variables, meaning that states can include their own provider numbers when submitting records to T-MSIS. In the TAF data, these numbers are delineated by variables with a specific Provider ID suffix. These identifiers can serve as an alternative form of linkage to the TAF APR, which contains Medicaid provider characteristics.

There are many types of Provider IDs in the TAF data, and each TAF file can contain different Provider ID types. These differences are determined by CMS billing and reporting requirements. For example, admitting providers are available in TAF Inpatient (TAF IP) and TAF Long Term Care (TAF LTC) files. Admitting Provider ID variables are not available in TAF Outpatient (TAF OT) and TAF Pharmacy (TAF RX) files because beneficiaries are not admitted into these facilities. Table 1 provides a breakdown of TAF Provider IDs by file. This table is accurate at the time of writing, so please refer to CCW's Data Dictionary for current information.

Table 1. TAF Provider IDs by File

Table 1 caption: A list of Provider IDs by TAF file

Provider IDs and NPIs, a quick caveat on equivalency

The National Provider Identifier (NPI) is a commonly available provider identifier in CMS’s research identifiable file (RIF) data as it is used for covered health care providers(1). For many NPI variables in TAF data, there is a Provider ID equivalent variable. There are a few exceptions to the rule in the TAF IP and TAF OT, which are outlined in Table 2. If analysts are automating linkages using both NPI and Provider ID, processes that assume that NPIs have a Provider ID equivalent may result in error for these variables.

Table 2. TAF Provider NPIs with No Provider ID Equivalent

Table 2 caption: A list of NPIs that do not have a Provider ID equivalent and which TAF file they are in.

Methodology

This analysis is sourced from the 100% 2022 TAF IP, RX, and LTC files and 5% TAF OT. The 5% TAF OT represents a fully random sample and was used to reduce computational burden. The year-by-year summary is sourced from 100% 2018-2022 TAF IP file. No data cleaning was applied, so service tracking and managed care records have not been removed. These records were not removed for an overall sense of data completeness, but they tend to have a reduced amount of information available compared to fee-for-service records(2).

All Provider IDs, except for Servicing Provider, are sourced from the header record file and its denominator is the total number of header records. The Servicing Provider ID was sourced from the line record file, and its denominator is the total number of line records.

Non-missing values are defined as values that are not blank and not 0- or 9-filled.

Results

Table 3 describes the percent of non-missing Provider ID counts in 2022 TAF records. The Billing Provider ID variable is non-missing over 98% of the time except for TAF OT (77.4%). The Referring Provider ID variable has the lowest percentage of non-missing values across all files (10.7%-31.9%). All other values fall somewhere in between. TAF RX presents higher completeness when compared to the other files with a range of 90.2% to 99.8%.

Table 3. Non-missing Provider ID (%) in TAF Records, 2022

Table 3 caption: This table shows the percent completeness of Provider ID by file in 2022

Table 4 displays the percent of non-missing values for Provider IDs in TAF IP from 2018 to 2022. Admitting Provider is filled less than 50% in this time frame. The Billing Provider variable improves in reporting from 2018 (90.5%) to 2022 (98.8%). Total claim count decreases from 2018 (14.6 million claims) to 2022 (12.6 million claims). Total line count varies between 127 million to 140 million lines.

Table 4. TAF IP Provider Fill over Time, 2018-2022

Table 4 caption: Provider ID fill in TAF IP from 2018 to 2022. The last two rows describe total claims and total lines.

Discussion

Provider ID completeness is dependent on TAF file type, which aligns with variable requirements. TAF RX Provider ID variables have mandatory coding requirements while other TAF files have situational requirements(3–5). The distinction between requirement versus non-requirement values could lead to reporting differences.

TAF OT has a lower completeness for Billing Provider ID compared to other TAF files. For Medicare Fee-for-Service claims, we often assume that billing provider will be completed as it is required for reimbursement. One possibility is state-specific reporting differences as this occurs with Billing Provider NPI(6). Another possibility is potential reporting differences between fee-for-service and managed care Medicaid providers, where capitated payment may reduce the likelihood of billing provider reporting.

Admitting provider is consistently missing over half the time in TAF IP. Information from the T-MSIS Data Guide seems to say that all non-denied records should contain an admitting provider number value(7). TAF RIF contains only approved records, so this is unlikely the cause of missingness. Since TAF IP is claim-level and not the stay level, merged claims could contain an admitting provider.

Servicing Provider IDs are unique in that they are a situational variable, but they are reported consistently. High reporting could be associated with their relationship to a mandatory variable. This variable is allowed to be equal to two mandatory variables, which is not a coding requirement available for other Provider IDs(8). The reduced administrative burden may result in higher reporting.

Provider IDs provide another method of linkage between TAF RIF and TAF APR. The availability of this variable for linkage is dependent on file and provider type. This may result in varied utility during linkage. Researchers should assess completeness prior to use.

Limitations

This article has several limitations. First, it is not recommended to generate a fully random sample from TAF data due to state-specific policy differences. This can result in biased reporting. Second, non-missing values do not include erroneous values (incorrect Provider ID or partial 0- or 9-fill), which could increase the perceived non-missingness in the data. Since there is not a single repository for Provider IDs and Provider IDs are not always publicly posted, it is not feasible to determine validity. Finally, this article is lacking state-level disaggregation. When working with TAF data, it is important to consider state-level differences as states have flexibility in their Medicaid programs. Reporting on a national level does not account for state-to-state nuances. National-level reporting still serves as a useful first step in understanding Provider ID completeness.