The Sage Database: Introducing Functional Outcomes in Geriatric Pharmaco‐epidemiology

To the Editor: Administrative databases have rarely been used in long-term care, largely because of the lack of adequate databases.1 Following the Omnibus Reconciliation Act of 1987, the Health Care and Financing Administration (HCFA) has begun to develop a national, long-term care database combining detailed clinical information and administrative claims data. Since 1991, a uniform data collection instrument, the Resident Assessment Instrument and its Minimum Data Set (MDS) has been used in all Medicare/Medicaid-certified nursing homes.2 MDS contains more than 300 data elements, demographic variables, numerous clinical items ranging from physical and cognitive function to the list of medical conditions, an array of symptoms, syndromes, and treatments provided, as well as indicators of residents' behavior, mood, and involvement in social activities.2 As part of HCFA's Multi-State Case-Mix and Quality Demonstration Project, all nursing homes in 5 states – Kansas, Maine, Mississippi, New York and South Dakota – have computerized each MDS assessment. In these states, an assessment is completed within 14 days of admission to the nursing home, again at 30 days, and quarterly thereafter.3 An MDS is also completed annually and whenever a resident's clinical condition requires a revision of care plan. In addition to MDS data, nursing staff record up to 18 different drugs each resident received in the 7 days preceding the assessments.3 Drugs are coded using the National Drug Code (NDC) system, and information includes brand/generic name, dosage, route, and frequency of administration, and whether standing or prn (pro re nata). The Medi-Span™ coding system, similar to the American Hospital Formulary Service (AHFS), has been adopted for providing clinically meaningful classification. We have assembled a database (see Figure 1) that cross-links these MDS data (1992–1996) with Medicare enrollment files (HISKEW (Health Insurance Skeletonized Write-off)) and the Medicare Provider Analysis and Review (MEDPAR) database, which contains information on all persons covered by Medicare Part A (97% of the US population aged ≥ 65 years). In addition to the data from the eligibility files (e.g., residence, HMO status, vital status) we have extracted selected information (e.g., dates of services, medical diagnoses, procedures performed, and reimbursements) about all hospitalization, skilled nursing facility, hospice, home health, and outpatient Part A claims between 1991 and 1996. All 1492 nursing homes included in the SAGE database have been linked to the OSCAR (On-line Survey Certification of Automated Records) archive.4 OSCAR describes facility characteristics, including ownership, staffing patterns, and a profile of the residents, as a part of the annual certification process (see Figure 1). These data are available for the years 1992 through 1996, and each MDS assessment in the SAGE database can be matched to the pertaining facility data. Finally, OSCAR data have been linked to the Area Resource File, a publicly available data set summarizing a large array of census, health and social information for all the counties in the contiguous United States.5 . Schematic diagram of the SAGE database. Although data collection will continue through the year 1998, as of December 1996, the SAGE database comprises more than one-half million unique residents, with more than 2.5 million assessments and nearly 20 million drug records. Here we report some preliminary data from the initial assessment of 296,379 residents between 1992 and 1994. As shown in Table 1, only 8.1% of the residents are less than 65 years of age, whereas more than 43% are 85 years of age or older; 4127 residents (1.4%) are centenarians. Women are over-represented (71.5%), as are residents of the white race (86%). Based on the ADL scale, 17% of residents are independent in feeding, dressing, and transferring (ADL 0,1), and 44% are highly or totally dependent (ADL 4,5). About 40% of residents score 0 or 1 on the Cognitive Performance Scale,6 which corresponds to a score greater than 20 on the Mini-Mental State Exam. The average number of medical diagnoses that affect function or influence the treatment plan is 3.3 ± 1.9, and residents receive an average of 6.1 ± 3.4 drugs. We believe that the SAGE database has the potential to capture the clinical complexity of long-term care residents,7 making it possible to measure functional outcomes related to specific pharmacologic treatments in a population traditionally excluded from randomized controlled trials.7,8 Because of the size and breadth of content of the SAGE database, it is possible to analyze clinically meaningful subpopulations and still enable extensive stratification not possible in previous studies. These advantages are able to ameliorate some of the methodological problems inherent in observational studies. Although, the MDS demonstrated reliability and validity when completed by research staff,9 use of the ongoing clinical and administrative data for research has been questioned.10 Extensive investigation is required to ascertain the reliability and validity of the information contained in the SAGE database and to make sure that its potential can be exploited.