Evaluation and Editing of Health Insurance Data

A feature that distinguishes the NHIS estimates of health insurance coverage from other survey-based estimates is the use of responses to follow-up questions to evaluate the reliability of the reported health insurance coverage and to adjudicate conflicting information. For many survey respondents, health insurance is a complex topic, and some inconsistencies in survey response are expected. If the follow-up questions clearly suggest that the original responses were incorrect, the original responses are edited. As a result, a portion of the sample is reassigned to a different type of coverage or reclassified from insured to uninsured (or vice versa). The evaluation and editing of health insurance coverage data on the NHIS is not new, and surprisingly many of the same issues in obtaining information on health insurance that the NHIS has today are similar to those encountered more than 50 years ago.

Since 1963, the evaluation and editing of health insurance coverage data has been facilitated by the collection of insurance plan names. Today, respondents who report any private insurance coverage and respondents who report coverage under Medicare managed care, Medicaid managed care, CHIP or any other state-sponsored or government plan are asked to provide the plan name. These data are used in conjunction with external sources to edit the reported source of coverage, as the plan name may identify the plan as something other than the reported source of coverage.

The evaluation and editing procedures involve a number of distinct operations, but there are several main components. The first is a series of automated edits based on string searches of the private plan names. If a reported name includes wording that indicates that it is a single service plan, the plan’s classification will be changed from a comprehensive health insurance plan to a single service plan. Examples include stand-alone dental, vision, and prescription drug plans. Later, detailed manual coding of the plan names provided by respondents who indicated coverage from a private source or selected public sources (including Medicare, Medicaid, and military coverage) can sometimes identify additional cases for which the reported coverage type is not accurate. For example, if a respondent stated that he or she has a private comprehensive plan and provided the name of the plan as “Medicaid,” the person would be coded as having Medicaid coverage rather than private coverage.

The key to the coding of plan type is the construction of a coding manual, which is based on information abstracted from a number of commercial and government sources. This manual consists of lists of names of private health insurance plans and public programs and plans. NHIS has been constructing coding manuals for more than 50 years. As early as 1962, a comprehensive list of “Blue” plans (Blue Cross and Blue Shield plans) was maintained for editing the NHIS data. A revision of the list was made in 1963, “not because any of the Blue Cross or Blue Shield programs had gone out of existence since the original 1962 indexes were prepared, but because numerous mergers, separations, and name changes had occurred. It was therefore necessary to retain all the names of plans in the original index since the outmoded name might be used by a respondent.” (Hoffmann, 1965). The coding scheme in these early days was more simplified; health insurance plans were coded as either being a Blue plan (contained on the list) or “Other". The “Other” plans for the most part were private insurance companies, but this category also included independent prepayment plans such as the Health Insurance Plan of Greater New York and the Kaiser Foundation Health Plan. The category “Other” was not subdivided because of difficulties in making accurate classifications (Hoffmann, 1965). Beginning with 2004, the person number variable changed to FPX which is unique within each family.

Today, the manual contains between 4,000 and 5,000 plan names, organized by state, and updated annually. The coding manual identifies each plan by its registered name and includes a code that indicates the nature of the plan. For example, the code may indicate that the plan provides only single-service coverage—for example, dental or vision care—rather than comprehensive coverage. Medicaid, CHIP, and other state and government plans are listed in a separate section of the coding manual, with codes identifying the respective source. Public plans with private-sounding names may be included in the private plan list—and often in both lists—if they were first identified through one of the commercial sources.

The following are the current sources of information for health insurance plan coding:

• The Competitive Edge (HealthLeaders-Interstudy)
• The National Directory of Managed Care Organizations (Health Resources Publishing)
• SMG HMO/PPO & Chains Directory (FirstMark, Inc)
• HMO/PPO Directory (Grey House Publishing)
• AIS Directory of Health Plans (Atlantic Information Services)
• Listing of Medigap Plans from the Centers for Medicaid and Medicare Services
• Mark Farrah and Associates
• Medicaid Managed Care Plan list (compiled by NCHS from Internet searches)
• Medicare Advantage List
• Guide to Federal Benefits
• Exchange Issuer and Exchange Plan name list (Department of Health and Human Services)
• Blue Cross and Blue Shield Plans (compiled by NCHS from Internet searches)
• Names of other public programs (compiled by NCHS from Internet searches)

Another main component of the evaluation and editing process is the use of manual edits based on follow-up questions for respondents who report no coverage at all. Respondents who reported one or more family members to be without coverage were later asked how long each such member has been without coverage and the reasons why the family member either stopped being covered or does not have health insurance. In a few instances, the respondent indicated that the subject actually does have coverage and, with or without prompting, provided the source of such coverage. Interviewers enter this information into a text field, and this information is used to assign coverage and the appropriate coverage type. This results in a small fraction of those who were originally classified as uninsured being assigned a type of coverage.

The editing process generates the series of new variables. These include indicators of coverage for each of several sources as well as a detailed code describing the type of coverage provided by each of up to four private plans reported by respondents. On the data file, the recodes are: MEDICARE, MEDICAID, PRIVATE, SCHIP, IHS, MILCARE, OTHPUB, OTHGOV, SINGLE, and NOTCOV. NOTCOV reflects the definition of noncoverage as used in Health, United States (in which persons with only Indian Health Service coverage are considered uninsured). Analysts are strongly advised to use the recodes MEDICARE, MEDICAID, PRIVATE, SCHIP, IHS, MILCARE, OTHPUB, OTHGOV, and SINGLE for estimates of types of health care coverage and NOTCOV to derive estimates of uninsurance. If users want to count IHS as coverage or create any other definition of uninsurance, they may also use the health insurance recodes (MEDICARE, MEDICAID, PRIVATE, SCHIP, IHS, MILCARE, OTHPUB, and OTHGOV) for this purpose. However, for those users who wish to review the originally reported health insurance coverage types prior to editing and back-coding, they are available in the HIKINDA through HIKINDJ, MCAREPRB, MCAIDPRB, AND SINCOV variables.

 

Procedures for Early Release Program Reports

The Early Release Program of the NHIS provides quarterly analytic reports and preliminary microdata files on an expedited schedule. NHIS data users can have access to these very timely reports and microdata files without having to wait for the final annual NHIS microdata files to be released in late June following the end of each data collection year. A more complete description of the program can be found here.

To produce the preliminary quarterly estimates that are included in the Early Release reports, NCHS developed a streamlined version of the usual final editing process. This alternative methodology bypasses the assignment of detailed codes to individual health plans but still attempts to determine whether coverage that is attributed to a private source or to a public source other than Medicare, Medicaid, or the military is in fact provided by another source. The streamlined editing makes use of a mix of automated and partially automated procedures with limited, targeted manual review.
 

Three operational choices make it possible to conduct the editing on a streamlined basis. The first is the use of the prior year’s coding manual to determine whether the reported source of coverage is correctly identified as a comprehensive private plan, CHIP, other public, or other government. The second is the use of automated keyword searches to identify the most easily recognizable single service plans, Medicare, and Medicaid program names misreported as another form of coverage. The keywords are developed from the coding manual used in the previous year. The third is the choice to produce only three variables for health insurance coverage type: INSURED6, TYPEPRI, and TYPEPUB. INSURED6 provides estimates for the insured and uninsured. TYPEPRI provides estimates for those with private comprehensive plans. TYPEPUB provides estimates for those with public coverage. Public coverage includes Medicaid, Medicare, CHIP, TRICARE, VA, and CHAMP-VA, other state-sponsored health plans, and other government programs. The Early Release Program does not produce individual variables for individual types of public programs.

An important feature of the early release coding is the judicious use of manual review to check or in some cases supplement the automated procedures. The number of records that are reclassified by the automated procedures is about 3-4%. A manual check of all these records entails fairly little additional effort but makes an important contribution to quality control. In addition, the number of records for which an automated check is considered unreliable is fairly small as well. Including these records in a manual review provides an opportunity to make more informed decisions about whether or not they should be reclassified.

A comparison of preliminary estimates based on the streamlined coding methodology to final estimates based on the complete editing procedures indicates that the preliminary estimates fall within 0.1 percentage points of the final estimates of the proportion uninsured. A more limited evaluation of the preliminary estimates of private and public coverage indicates somewhat greater divergence (generally 0.2 or 0.3 percentage points) from the final estimates.

 

Procedures for Classifying Exchange-Based Coverage

The core questions in the Health Insurance section of the NHIS Family Core have remained largely unchanged since 1997, including the questions that collect insurance plan names. In 2014, in response to the Affordable Care Act of 2010 (P.L. 111-148, P.L. 111-152) (ACA), several new questions were added to the FHI section to capture health care coverage obtained through the Health Insurance Marketplace or state-based exchanges. These new questions were asked for persons covered by public plans (Medicaid, CHIP, state-sponsored health plans, and other government programs) as well as for those on private health insurance. This acknowledges that some respondents perceive exchange coverage as a public program and others perceive exchange coverage as private health insurance. For persons who indicate Medicaid, CHIP, state-sponsored health plans, other government programs, respondents were asked if they obtained this coverage through Healthcare.gov, the Health Insurance Marketplace, or the name of their state’s exchange. This was followed by a question that asked if there was a premium or enrollment fee associated with the plan. If this was the case, the respondents were asked if this premium was based on income. For persons covered by private insurance plans that were not employment-based, respondents were asked if the plan was obtained through Healthcare.gov, the Health Insurance Marketplace, or the name of their state’s exchange. For those who pay a premium for their private health insurance plan, a question was added in Quarter 4 of 2013 to ascertain whether that premium was based on income. The premium question was asked for both employment-based and non-employment-based plans.
 

In general, if a family member was reported to have coverage through the Health Insurance Marketplace or a state-based exchange, then that reported coverage is considered accurate unless there is some other information that clearly contradicts that report. Similarly, if a family member was not reported to have coverage through the Marketplace or state-based exchange, then that is considered accurate unless there is some other information that clearly contradicts that report. Specific applications of these general rules are detailed here [Detailed Exchange Editing Rules] and summarized below.
 

The NHIS considers a person reporting private coverage as having exchange-based coverage if they are reported to have a private, non-employment-based, directly purchased plan and the plan name provided is a) an exchange plan name or b) an exchange portal name (e.g. Healthcare.gov, kynect in Kentucky), or c) they have provided an exchange company name and the respondent indicated that the plan is through the Health Insurance Marketplace or state-based exchange or d) the plan name was unknown or refused and the respondent indicated that the plan was obtained through the Health Insurance Marketplace or state-based exchange. Providing an exchange plan name or an exchange portal name is weighed heavily in the decision to classify a person as having exchange-based coverage. Persons with employment-based coverage were not considered to have exchange coverage unless a very specific exchange plan name was provided.
 

When a state-sponsored health plan or another government program is reported, the person is classified as having exchange-based coverage if the plan name provided is a) an exchange plan name or b) an exchange portal name (e.g. Healthcare.gov, Covered California) or c) an exchange company name and the respondent has indicated that the plan was through the Health Insurance Marketplace or state-based exchange, or d) the plan name was unknown or refused and the respondent indicated that the plan was obtained through the Health Insurance Marketplace or state-based exchange and they had a premium associated with the plan. The source of coverage would be changed from public to private. As noted above, providing an exchange plan name or an exchange portal name is weighed heavily in the decision to classify a person as having exchange-based coverage.

All individuals who are classified as having exchange-based coverage are considered to have private health insurance, regardless of whether they were reported to have obtained the coverage from a private or public source.