Health Conditions, Utilization, and
Expenditures of Children in Foster Care

CHAPTER II:
DATA AND METHODS

Contents

1. Data Source
2. Selection of Study States
   1. Use of Medicaid Managed Care
   2. Variation in State Program Characteristics
3. Selection of Study Sample
4. Construction of the Analytic File
   1. Demographic Characteristics
   2. Diagnostic Conditions
   3. Medicaid Expenditures
   4. Health Care Utilization

A. DATA SOURCE

The data source for this study is the State Medicaid Research Files (SMRF), a series of analytic data files created from the Medicaid Statistical Information System (MSIS) data that are submitted by states to the Health Care Financing Administration (HCFA) on a quarterly basis. The SMRF files have several advantages over data from MSIS and individual state Medicaid management information systems (Dodds 1997):

• SMRF files are organized by calendar year date of service rather than quarterly date of payment.
• SMRF claims are adjusted to reflect final bills and include hospital stay records created from multiple interim bills.
• A person summary file contains monthly eligibility information in a single record, together with annual utilization and expenditure information.
• Analytic variables, such as condition flags, type of service indicators, and eligibility categories, are created for research purposes.
• Enhanced editing and verification procedures are performed to identify and document problems and to ascertain the validity of data for particular applications

The SMRF data are organized into five files, including the person summary file and four claims files: inpatient, long-term care, drug, and other (which comprises claims for professional services, outpatient clinics, and premium payments). We used the person summary file for descriptive analyses of the dynamics of Medicaid eligibility and distribution of Medicaid expenditures. We used the claims files primarily to compare patterns of diagnoses and utilization.

Mathematica Policy Research (MPR) obtained SMRF files from three states for two years. These data were obtained from HCFA solely for the purpose of this study through a data use agreement between HCFA, ASPE, and MPR. The states and study periods are as follows:

These study periods represent the most recent SMRF data available for each state that met our selection criteria. The next section explains the criteria used to select the three states for the study.

[ Go to Contents ]

B. SELECTION OF STUDY STATES

We used the following hierarchical criteria to aid in the selection process:

1. Availability of Medicaid claims and enrollment data in the SMRF files
2. Ability to identify foster care children in the SMRF files
3. Identifiable foster care population of at least 10,000 children, to detect significant health conditions that are relatively rare events
4. Degree to which children are enrolled in Medicaid managed care
5. Variation in features of state foster care systems

As of August 1998, 34 states were participating in SMRF in 1995; 27 of these states were able to identify children who qualify for Medicaid because they are in some form of foster care or receive adoption assistance (Table II.1).⁽¹⁾

Seven of the 27 states — California, Pennsylvania,⁽²⁾ Florida, New Jersey, Washington, Wisconsin, and Georgia⁽³⁾ — had foster care populations of more than 10,000 children, and Indiana had a population of nearly 10,000 children. Four additional states — Missouri, Colorado, Kansas, and Minnesota — each had populations between 6,000 and 8,000 children. The remaining states all had foster care populations of less than 5,000 children.

After we narrowed the list of states to the seven with at least 10,000 children in foster care, we considered two additional factors, namely the extent of Medicaid managed care enrollment and variations in state program characteristics. We now turn to a discussion of each factor.

1. Use of Medicaid Managed Care

There has been a trend in recent years toward the use of managed care for Medicaid-eligible children in general and foster care children in particular (Battistelli 1997). By 1996, 22 states had enrolled at least some foster care children into capitated (prepaid) Medicaid managed care, and 17 of these states required at least some of these children to enroll in managed care (NASHP 1997). The use of Medicaid managed care poses significant challenges for this study because the claims data for children in capitated managed care plans are missing from the SMRF files. And without claims data, we cannot answer the research questions posed in this study.

Of the seven states with at least 10,000 Medicaid foster care children, all but one had overall Medicaid managed care penetration rates of 20 percent or less in 1994:

One caveat is that the managed care penetration rate was likely to vary across age groups and eligibility categories and children may have had above average rates of managed care enrollment. Our strategy, therefore, was to choose the three states with the largest foster care populations — California, Pennsylvania, and Florida — to ensure adequate sample sizes for the foster care analyses, while recognizing that the sample sizes in the other categories of eligibility would be more than adequate for our purposes.

California had by far the largest foster care population (nearly 100,000 in 1995), and, for that reason alone, was of great interest as a potential study state. Pennsylvania and Florida were next in size of foster care population, with 24,000 and 21,000 Medicaid children in foster care, respectively. We concluded that the relatively large size of the foster care population compensated for the level of managed care enrollment in these two states (20 percent in Pennsylvania, 16 percent in Florida). These large sample sizes have afforded us the opportunity to compare patterns of utilization and expenditures within the foster care population.

2. Variation in State Program Characteristics

We researched two key program characteristics to ensure that the three states varied on important factors. The first is whether the foster care programs are administered at the state or county level. The foster care programs in two of the states, California and Pennsylvania, are state supervised and county administered, while the program in Florida is state-administered (Child Welfare League of America 1999). Thus, we might expect to see more intrastate variation in utilization patterns in the two county-administered programs.

The second characteristic is the presence of a health passport program.⁽⁴⁾ All three states have implemented health passport programs statewide, with Florida's and Pennsylvania's passports dating back to 1989 and 1990, respectively. California's health passport program was implemented statewide in February 1995. None of the states use an electronic (computerized) passport record (Lutz and Horvath 1997).

[ Go to Contents ]

C. SELECTION OF THE STUDY SAMPLE

An important step in the file construction phase involved the selection of the study sample. The selection of the study sample took place in several steps, as follows:

Step 1: Create a Subset by Age

We created a subset of the administrative data containing records for children under the age of 19.

Step 2: Identify Foster Care Children

The SMRF person summary file contains SMRF eligibility codes that identify both the monthly Medicaid eligibility categories and the primary Medicaid eligibility category during the year. The SMRF eligibility codes were created by classifying state-specific eligibility codes according to the standardized set of eligibility codes that make up the SMRF coding system. During the file construction phase, we discovered a problem with the original mapping of Florida's state-specific eligibility codes into the SMRF eligibility code. We remapped the eligibility codes in Florida to correctly identify categories of Medicaid eligibility. We also discovered that the SMRF foster care category includes children receiving adoption assistance as well as those receiving emergency assistance in conjunction with child welfare services.

Table II.2 shows, for each of our three study states, the state-specific codes that identify which children are receiving foster care assistance, adoption assistance, or emergency assistance, and the number of children in each category of Medicaid eligibility (based on the main category of eligibility for the year).

Table II.3 presents additional information on the number of children eligible for Medicaid due to foster care, adoption assistance, or emergency assistance. This table shows that the primary SMRF eligibility code — which reflects the main category of Medicaid eligibility for a child — understates the number of children who were eligible for Medicaid due to an out-of-home placement at any time during a given year. The number of children with any period of foster care ranges from 15 percent to 23 percent higher than the number of children whose main category of Medicaid eligibility was some form of foster care.

For the purpose of this study, we reconstructed variables related to Medicaid eligibility, to show whether a child had any period of foster care, adoption assistance, or emergency assistance. All results are shown separately for children in foster care and for those receiving adoption assistance. Children receiving emergency assistance are excluded from the main analysis, but basic descriptive information about this group is presented in Appendix A.

Step 3: Create Comparison Groups

Next, we defined three comparison groups of children enrolled in Medicaid: those receiving adoption assistance through Title IV-E or other sources, those receiving cash assistance through Aid to Families with Dependent Children (AFDC), and those receiving Supplemental Security Income (SSI) benefits.⁽⁵⁾

These groups were chosen to compare to the foster care population because they are of interest to policymakers. The SSI population includes children with disabilities, and there is considerable interest in understanding how this population is similar to or differs from foster care children. The AFDC population includes children receiving public assistance and, as we shall see in Chapter III, there is substantial overlap between the foster care and AFDC groups. Many of the foster care children were eligible for Medicaid through AFDC either before or after their foster care eligibility. Children receiving adoption assistance are also of interest to policymakers who would like to know to what extent these children are similar to or different from those who remain in foster care in terms of diagnoses, utilization, and costs.

In earlier work, we included a broader set of comparison groups, namely children eligible for Medicaid through poverty-related expansions, children who are medically needy, and children in other categories of Medicaid eligibility. Based on a preliminary assessment of the data, we decided (in consultation with ASPE and HCFA staff) to streamline the tabular displays to include only comparisons with children receiving adoption assistance, AFDC, or SSI. The totals, however, include all Medicaid children, regardless of category of Medicaid eligibility.

Step 4: Exclude Children Enrolled in Managed Care

Because providers do not submit individual claims for services provided to children enrolled in prepaid managed care, it was necessary to omit managed care enrollees from the analyses of diagnoses, utilization, and costs. Therefore, we developed specifications to identify children enrolled in prepaid, or capitated, managed care.

Identifying this group proved more complex than we anticipated because there is no direct, accurate approach to measuring managed care enrollment based on the SMRF eligibility or claims data. We developed state-specific algorithms to utilize the data available in each state. In Florida and Pennsylvania, we excluded children who had any premium payment during the year, that is, one or more claims reflecting a capitation payment to a managed care organization. In California, we were unable to use this approach because a large number of children had premium payments, but only for dental care. Thus, we relied on plan identifiers to exclude children enrolled for one or more months in managed care organizations. Children enrolled only in prepaid dental plans remained in the sample. Fortunately, the California Medicaid program, known as Medi-Cal, receives shadow claims for dental services from dental plans, which allowed us to analyze dental utilization in California even for those enrolled in prepaid dental plans.

Children enrolled in managed care plans are included in the demographic analyses and the analyses of Medicaid eligibility dynamics. They are excluded, however, from all analyses related to diagnoses, utilization, and costs. Managed care participation rates are discussed further in Chapter III.

[ Go to Contents ]

D. CONSTRUCTION OF THE ANALYTIC FILE

Once we selected the study sample, we created analytic files reflecting demographics, health status, utilization, and expenditures.

1. Demographic Characteristics

Demographic variables available on the SMRF files include age, gender, and race/ethnicity. We created a measure of urbanicity by merging ZIP codes from the SMRF person summary file to the Area Resource File to identify large urban, small urban, and rural areas. Large urban areas are metropolitan statistical areas (MSAs) with one million or more residents, small urban areas are MSAs with less than one million residents, and rural areas are those located outside of an MSA.

2. Diagnostic Conditions

We explored various approaches to classifying diagnoses within the Medicaid population. Our goal was to identify an algorithm that would classify the types of diagnoses found within the foster care population and then allow us to compare the distributions of diagnoses to those found in the general Medicaid population. Specifically, we wanted to be able to differentiate between physical and mental conditions. If possible, we also wanted to distinguish the level of severity within diagnostic groups. We explored approaches developed by Kronick et al. (2000), Burwell et al. (1997), and Perrin et al. (1999), as well as the crosswalk used by the Social Security Administration (SSA 1998).

We selected the Chronic Illness and Disability Payment System (CDPS), because it not only identifies severe and chronic conditions among children but also differentiates according to the severity or complexity of the case within a given diagnostic category (Kronick et al. 2000). Specifically, the CDPS identifies 20 diagnostic categories and identifies gradients of high-, medium- and low-cost subcategories within each.⁽⁶⁾

For our purposes, the approach by Burwell et al. (1997) was more limited because it did not define specific diagnostic groups (beyond physical and mental conditions in the aggregate) and did not allow for differentiation of severity within the broad categories. Nor did the crosswalk by Perrin et al. (1999) allow for as fine a breakdown of diagnostic groups. Although the SSA crosswalk contains a wide range of diagnostic codes to classify those receiving SSI benefits into broad diagnostic categories, it is not appropriate for identifying chronic or disabling conditions within the general population.

Outpatient diagnostic data were only available for two of the three states; Florida did not report diagnostic information on its outpatient SMRF files. Therefore, the diagnostic comparisons were performed only for California and Pennsylvania.

In addition to examining chronic illness and disability among low-income children, we also compared the prevalence of deliveries across each of the Medicaid eligibility categories. We developed a measure of the number of girls, ages 15 to 17, who delivered a baby in 1994. We used the SMRF delivery indicator, which was based on ICD-9-CM codes signifying a live birth.⁽⁷⁾ Age was measured as of the end of the year (December 31, 1994). We restricted the measure to include only girls ages 15-17, to compare these rates with national benchmarks from vital statistics (Ventura et al. 1996).

3. Medicaid Expenditures

The analysis of Medicaid expenditures is based on data reported in the SMRF person summary file, which aggregates annual amounts paid by Medicaid within 24 types of service (TOS) categories.⁽⁸⁾ These categories were further aggregated into four service groups to create subtotals of analytic interest. The TOS classification is shown in Table II.4. For most analyses, we compare average (mean) monthly expenditures across groups, which were derived by dividing total expenditures for the year by the number of months of enrollment. As we will discuss in the next chapter, there are some inconsistencies in how states classify claims by type of service. Most notable for the purpose of this study is state variation in the classification of mental health services.

4. Health Care Utilization

To analyze health care utilization patterns, we created indicators of the probability and level of service use. In general, we used definitions developed as part of the Health Plan Employer Data & Information Set (NCQA 1998). Our goal was to create a parsimonious set of measures, while still capturing the range of variation within and among groups. Table II.5 lists the measures, the SMRF source file, and the method used to construct each measure.

Where possible, we included state-specific procedure codes to classify emergency room, preventive, and mental health/substance abuse services. These codes were obtained from internal files as well as from follow-up discussions with states.

Separate utilization measures were constructed for mental health and substance abuse services. As discussed earlier, Florida's outpatient SMRF file did not contain any diagnostic data. As a result, we were unable to distinguish between mental health and substance abuse services in Florida; we therefore decided to combine these services into a broader measure reflecting behavioral healthcare.

[ Go to Contents ]

Footnotes

1.  This can include children receiving foster care assistance (both Title IV-E and non-Title IV-E), adoption subsidies (both Title IV-E and non-Title IV-E), and children in group homes or those who are otherwise wards of the state. The remaining seven states either do not identify foster care children or have problems with their eligibility data that make the states unsuitable for this study.

2.  Data for Pennsylvania are from 1994. A SMRF file for 1995 is available, but it has significant data quality problems which prohibit it from being used for this study.

3.  One of the limitations of the Georgia data was the omission of eligibility data for the first three months of 1994. This would censor the data to less than two full years of eligibility and claims history.

4.  A health passport is a traveling medical record for children in foster care, which tracks their medical history and documents their health care utilization.

5.  The SMRF file shows only one Medicaid eligibility category per month. Children were classified according to the eligibility category accounting for the majority of the eligibility period, with the exception of children in foster care who were classified in the foster care category if they were "even enrolled" in foster care during the year. Foster care children who received SSI benefits were counted in the foster care category only if the SMRF file identified one or more months of foster care eligibility. It is possible that some foster care children are counted in the SSI category if states code only their SSI eligibility on the SMRF file.

6.  For this study, we excluded the pregnancy and newborn complications categories included in the CDPS. While these categories are relevant for purposes of risk adjustment (predicting higher costs in a subsequent year), they are not reflective of chronic illness or disability per se.

7.  The diagnostic codes are: 650, 646.0-656.3 (fifth digit 1, 2), 656.5-676.9 (fifth digit 1, 2),V27.0, V27.2, V27.3, V27.5, V27.6, V27.9, V30-V39.21.

8.  Two categories — mental hospital for aged and skilled nursing facility/intermediate care facility (SNF/ICF) mental health services for aged — were not applicable to this study. Another category — ICF-all other — was no longer in use. The premium payment category was not applicable since children enrolled in managed care were excluded from this study, and dental premiums in California were reported under the dental type of service.