CCW administrative claims, enrollment, and chronic condition indicators for 2005 were used in these analyses. Since the CCW data files are already linked by a unique beneficiary key across time and claim type, no beneficiary linkage efforts are required by researchers (e.g., traditionally it has been challenging to link all data for a patient over time because of changes in the Medicare health insurance claim number due to changes in eligibility status). This linkage strategy simplifies examination of the full continuum of care as well as longitudinal studies. Minimal merging of files is required prior to development of the analytic code to address the study objectives.
The CCW contains all Medicare FFS institutional and non-institutional claims, assessment data, and enrollment/eligibility information from January 1, 2000 forward. A random 5% sample of Medicare beneficiaries is the standard data file available to researchers, although the database contains information for 100% of beneficiaries and can be used to select a wide range of cohorts. There are predefined chronic condition indicator variables which are made available to researchers for cohort selection and data extraction, as well as for chronic disease research.
The twenty-one predefined condition indicator variables are coded within the CCW and disseminated to researchers as variables in the Chronic Condition Summary File. Algorithms involving Medicare claims-based utilization information are used to make the chronic condition determinations (i.e., an indicator that the beneficiary received services or treatment for the condition of interest within the specified time period). The identification of each of these conditions is limited to the information available from Medicare administrative claims (e.g., based on ICD-9-CM  and HCPCS codes ). Treatment information is not available for those enrolled in Medicare managed care plans.
Institutional (i.e., inpatient, outpatient, skilled nursing facility, home health, and hospice) and non-institutional (i.e., physician/supplier and durable medical equipment) FFS claims for services provided in 2005 were used in the analyses. The 5% random sample of the Medicare population, based on the standard sampling methodology used by CMS , formed the sampling frame for this study, from which a narrower cohort was identified.
The Medicare beneficiary enrollment and eligibility information was obtained from the CCW Beneficiary Summary File, which also contains beneficiary demographic and Medicare coverage information. The predefined chronic condition indicator variables were obtained from the CCW Chronic Condition Summary File. Since these condition indicators are defined using only FFS claims-based criteria (e.g., ICD-9-CM codes, specific combinations of claim types, etc.) and no managed care utilization information, only FFS beneficiaries with Part A and B coverage were included in the cohort. Beneficiaries who were alive on January 1, 2005 and enrolled in Medicare Parts A and B for at least 11 of the 12 months in the year, or until the time of death (i.e., covered for every alive and eligible month, or covered for all except one of the alive and eligible months), and who had one month or less of managed care coverage, were considered eligible for the study cohort. Since this cohort was selected from the random 5% sample, some of whom had the chronic conditions of interest, the findings may be generalized to the larger Medicare FFS population.
Nine of the 21 predefined chronic condition indicator variables were used in this study. Four types of cancer were combined into one variable, including female breast, colorectal, prostate, and lung cancer, due to similarities in the patterns of care (e.g., settings used), the desire not to unduly inflate the numbers of distinct disease types being treated simultaneously for a beneficiary, and for simplicity in the analyses. This resulted in six chronic condition variables which were used for these analyses. The diseases represented included cancer, CKD, COPD, depression, diabetes, and HF. A summary of the types of services used to define these conditions is provided in Additional file 1.
The comparison group used throughout this study consisted of the remainder of the random 5% sample who were not receiving treatment for any of these six conditions during 2005. Please note that it is possible that some of the beneficiaries within this comparison group may have been receiving treatment for other types of medical conditions (or for any of the other 12 CCW conditions), which were not a part of the current study (i.e., it is not necessarily a disease-free group). The administrative claims data for the study cohort were extracted from the CCW and aggregated by beneficiary using the unique beneficiary identifiers created in the CCW. The resulting beneficiary-level, aggregate claims utilization and cost file was used for all further analyses.
Cancer, COPD, and depression are CCW algorithms which consider services occurring during a one-year look-back period. The CCW uses a two-year look-back period for CKD, diabetes, and heart failure. The algorithms use these look-back periods as the length of time during which a certain service(s) can be provided to a beneficiary for inclusion in the chronic condition category.
Medicare utilization was assessed using each of the claim types. These included inpatient, skilled nursing facility, home health, outpatient, hospice, physician/supplier and durable medical equipment claims. Unique inpatient and skilled nursing facility (SNF) stays were defined as those with a paid Medicare amount and discharge date in 2005, regardless of the reason for the stay. The number of days was calculated by taking the sum of all covered Medicare FFS days of care chargeable to Medicare in 2005. The number of visits (i.e., home health, institutional outpatient, and physician office) was defined as the average number of FFS visits per beneficiary in 2005. Home health (HH) visits were counted using a total visit count variable on the claims. Institutional outpatient (OP) visits were averaged from the sum of the number of outpatient claims. Physician office visits represent the number of evaluation and management visits where the HCPCS ranged from 99201-99205 or 99211-99215, as indicated on the Carrier (physician office) claims.
Costs were defined as total Medicare payment (per claim type), or the sum of all FFS claim payment amounts, per beneficiary for 2005. For each beneficiary, total Medicare payments were summed across all claim types for all services provided during the year, regardless of the diagnosis on the claim. The average Medicare payments per beneficiary were calculated. These population totals and averages were examined for each claim type, then for each of the selected conditions and for beneficiaries with varying numbers of conditions.
There are various methods by which the chronic condition indicator variables may be used in the calculation of population prevalence rates for chronic conditions. A technical paper describing some of the basic methods for performing analyses with these indicator variables is available on the CCW web site http://www.ccwdata.org. The methods used for this study to ascertain prevalence for the chronic conditions, including the rationale for allowing a one month break in FFS Medicare coverage for the study cohort, are more fully described and justified in the technical paper . To summarize, allowing for a one month break in Medicare A or B coverage (or allowing one month of managed care coverage), rather than requiring full Medicare coverage for a 12 month surveillance period, allows for retention of a fair number of beneficiaries in the cohort for whom there is evidence that treatment for the condition(s) of interest occurred. Eleven months (rather than 12 months) FFS coverage may be sufficient for denominator criteria (note that numerator criteria may use different look-back periods) for the purposes of examining population period prevalence of chronic conditions.
The utilization data presented in this paper focus on beneficiary averages rather than simply raw utilization statistics for this cohort. This per capita comparison controls for the number of persons in each category.
For further comparison of utilization across conditions, odds ratios (ORs) were calculated for each care setting. ORs allow for the comparison of the likelihood of the type of care for beneficiaries with a condition, compared to beneficiaries with no condition (i.e., none of the six conditions of interest in this study). For example, the OR for beneficiaries with diabetes receiving inpatient care was computed by dividing the odds of those beneficiaries having an inpatient stay, by the odds of beneficiaries with none of the six conditions having an inpatient stay during the year. The identification of this reference group allows for comparisons regarding the relative importance of the six conditions, and accounts for the fact that the six conditions are not mutually exclusive categories (e.g., beneficiaries may have CKD and diabetes). ORs were also calculated for the comparison of utilization likelihood for beneficiaries with multiple conditions to beneficiaries with none of the six conditions. Comparisons of utilization across conditions are presented for the most frequently used settings of care.
Cost comparisons of total Medicare payments and average-per-beneficiary Medicare payments, by condition and number of conditions present, were also explored in order to more adequately understand the costs of care for beneficiaries with each condition(s). Ratios of means (ROM) were calculated to further compare the differences in average payment amounts per beneficiary by chronic condition and care setting. Each ratio of means was calculated by dividing the average payment amounts per beneficiary for those with the condition, by the average payment amounts per beneficiary for those with none of the six conditions.