The CMHS contains enrollment and claims data for a 5 percent random sample of the Medicare population (based on the last two digits of the Health Insurance Claim [HIC] number), covering the most recent 30-year period for which data are available. Medicare payment and utilization data are aggregated by type of claim by calendar year; claim-specific data such as diagnoses and procedures are not contained on the file. An indicator for managed care enrollment is included for each year and intensive care unit (ICU) and coronary care unit (CCU) charges from inpatient hospital stays have been aggregated for each year after 1982. Payments for prescription drugs, which Medicare began covering in 2006, are not included in this file. The CMHS is longitudinal, that is, beneficiaries remain on the file until death unless their HIC numbers change and cause them to drop out of the sample.
The file used for this study covered the years 1978–2007; results are presented through 2006, as explained below. Medicare payments for the years 1998–2000 were not accurately recorded in the CMHS because of a programming error. Consequently, the years 1998–2000 were excluded from all analyses. For each study year, the analysis was restricted to beneficiaries who were age 65 and older and were not enrolled in a managed care plan at any time during the year. Managed care enrollees were eliminated from the analysis because Medicare does not receive claims payment data for them. Within the study years, managed care enrollment peaked at 19.5 percent of the CMHS sample in 2006. Annual Medicare payments were inflation adjusted to 2006 using the Consumer Price Index (Bureau of Labor Statistics 2008
Our primary goal was to assign Medicare payments either to decedents (persons in their last year) or to survivors (all others) for each calendar year, following methods used in an earlier study (Lubitz and Riley 1993
). In the case of calendar year 2006, for example, persons who survived through December 31, 2007 were identified as survivors for 2006 (see nos. 3 and 6 in ). Their person-years of enrollment in 2006 and all their Medicare payments for services provided in 2006 were assigned to survivors. For persons who died in 2006, person-years of enrollment and payments in 2006 were assigned to decedents (nos. 1 and 4). For those dying in 2007 (nos. 2 and 5), a portion of the payments and person-years of enrollment for calendar 2006 was assigned to survivors and a portion to decedents. The portion assigned to decedents (with the exception of hospice payments, as described below) depended on the proportion of calendar 2006 that was spent within 365 days of death. Thus, if person no. 2 in died on the 100th day of 2007, then 265/365 of their 2006 enrollment and nonhospice payments were assigned to decedents and 100/365 to survivors. This method differs from that used in the earlier study, where dates of discharge were used to assign inpatient hospital and skilled nursing facility services to decedent and survivor categories. In the current study, payments for all services were prorated because dates of discharge are no longer available on the CMHS database. Payments assigned to decedents may therefore be understated because service use tends to increase as death approaches; prorating payments by days may result in too few payments being assigned to the last year of life. A comparison of overlapping years (1980, 1985, and 1988) between this analysis and the earlier study suggests that the change in methodology reduced the estimated payments going to the last year of life by less than two percentage points. For hospice services, all 2006 payments for 2007 decedents were assigned to decedents rather than being prorated between decedents and survivors, because it was assumed that all hospice services were provided close to death. The same methodology was used to assign person-months and payments to decedents and survivors for all years of the study.
Examples of Allocation of 2006 Medicare Payments between Decedents and Survivors
The percents of payments going to decedents and survivors in each year were adjusted to the age, sex, and survival status of the 1978 sample to account for changes in the Medicare population over time. Adjustment for a given year was made by applying average payment amounts for each cell to the age, sex, and survival distribution of the 1978 sample. Both unadjusted and adjusted estimates are presented. Per capita payments for a given year were computed by dividing total dollars assigned to decedents (or survivors) by the total number of person-years for decedents (or survivors) in that year.
Hospitalization and use of ICU and CCU services near death represent two measures of the intensity of terminal care (Earle et al. 2004
; Wennberg et al. 2004
;). For beneficiaries dying in March of selected years, we determined the percent who were hospitalized and the percent using ICU/CCU services (measured as ICU/CCU charges greater than U.S.$0) in the calendar year of death. March decedents were chosen for this analysis to capture hospitalization and ICU/CCU use in the last 2–3 months of life, given the limitations of the CMHS file structure. Rates of hospitalization may be understated because hospital stays are counted from inpatient hospital claims files and Medicare does not receive inpatient hospital claims for some hospitalizations undergone by hospice enrollees. Those hospitalizations are billed as hospice services, provided that the hospitalization is for care related to the terminal illness. Undercounting of hospitalizations would be greater in the later years of the study when hospice use was highest.