Objective: To describe the medication component of the SAGE (Systematic Assessment of Geriatric drug use via Epidemiology) database and to present a summary of the population of nursing facility residents and their medication use. Design: Population-based observational study. Setting: All Medicare- and Medicaid-certified nursing facilities in Kansas, Maine, Mississippi, New York, and South Dakota. Study Population: 363,354 residents admitted to nursing facilities between 1992 and 1995. Main Outcome Measures: Percentage of the nursing facility population receiving selected drugs. Results: Over 70% of the residents in the nursing facilities included in the SAGE database are women and over three-fourths are 75 years of age or older. Cardiovascular agents, anti-ulcer medications, and ophthalmic preparations are the most commonly prescribed medications. Overall, while 31.5% of residents used at least one inappropriate drug as defined by the Beers criteria, only 8.6% of residents used a medication with a high potential for unnecessary adverse outcomes such as amitriptyline. However, significant variation existed between facilities with respect to the percentage of residents receiving potentially inappropriate medications. Conclusion: The SAGE database is unique in that it permits longitudinal assessments of medication use and links to many health outcome variables associated with elderly nursing facility residents. In the future, the database will offer opportunities to measure the effects of the prospective payment system and the use of drugs as quality indicators. Key Words: SAGE database, Nursing facility, Minimum Data Set, Medication use. Abbreviations: ACE = angiotensin converting enzyme; HCFA = Health Care Financing Administration; MDS = Minimum Data Set; NDC = National Drug Coding; OSCAR = On-line Certification Automated Survey; p.r.n. = as needed; SAGE = Systematic Assessment of Geriatric drug use via Epidemiology; Consult Pharm 1998;13:1356-64..

The elderly are a vulnerable population; many of them are frail and at high risk for drug-related problems. In older persons, the outcomes of medication use, both intended and inadvertent, may differ from those seen in younger populations. For example, an antihypertensive agent's adverse effects on cognition, continence, and mobility may be as important as its therapeutic efficacy. Despite limited information on the safe and effective use of medications in the elderly, they are likely to receive multiple drugs owing to their concomitant illnesses.

Since passage of the Omnibus Budget Reconciliation Act¹ of 1987, consultant pharmacists have been required to review the drug therapy of all nursing facility residents for appropriateness in an attempt to lessen medication-related problems.² An evaluation of the appropriateness, or potential inappropriateness, of drug therapy should be based on information published in the medical literature.

When information on drug use in a specific population is inadequate, large computerized databases from health maintenance organizations, pharmaceutical marketing companies,^3,4 or Medicaid populations frequently provide valuable information on the outcomes of medication use. However, the first two types of databases usually do not include the elderly, and none provide comprehensive information on patients' clinical and functional characteristics. In addition, other payment-based databases do not include non-prescription drugs, which are commonly used and known to have observable therapeutic effects.

The SAGE (Systematic Assessment of Geriatric drug use via Epidemiology) data system is a unique database that can be used for population-based research into prescribing patterns, adherence to established criteria for geriatric practice, as well as the association between medication use and health outcomes in elderly nursing facility residents. In addition, it may be used for evaluating the impact of regulations, such as the prospective payment system, and the use of drugs as quality indicators. The purpose of this report is to describe the SAGE database and its development, as well as to provide an overview of the use of medication, including potentially inappropriate uses, in this older population.

Methods
In collaboration with the University of Michigan Assessment Archive Project, the SAGE data system was assembled linking data from the nursing facility resident assessments, drug use, Medicare claims data, organizational data on nursing facility providers, and the Area Resource File. Over 360,000 individual residents were represented in the database, with over one million assessments and over 15 million drug records. Figure 1 illustrates the SAGE data sources, which have been summarized in the following sections.

Figure 1. SAGE Data Sources

Minimum Data Set
Between 1992 and 1995, all 1,492 Medicaid- and Medicare-certified nursing facilities in Kansas, Maine, Mississippi, New York, and South Dakota participated in the Health Care Financing Administration's (HCFA) Multi-State Case-Mix and Quality Demonstration Project. All residents had an assessment within 14 days following admission to the facility, at 30 days, and quarterly thereafter. The Minimum Data Set (MDS) of the Resident Assessment Instrument includes over 300 data elements.^7,8 Nursing and social work staff who were primarily responsible for the resident completed the MDS.⁹

Seven direct measures of cognition covering short- and-long term memory, recall or orientation items (season, location or room, staff names or faces, orientation to the nursing facility), and decision-making ability were recorded on the MDS.¹⁰ Resident physical functioning was based on the activities of daily living classification of six-levels of self-performance including dressing, eating, toilet use, bathing, locomotion, transfer, and continence.¹¹ Staff evaluated residents in each of these areas using a five-point scale as independent, needing supervision, needing limited assistance, needing extensive assistance, or totally dependent. In addition, communication and hearing problems, psychosocial well-being, mood state, activity and recreation, disease diagnoses, health conditions, nutritional status, oral and dental status, skin condition, and special treatments were also recorded.¹²

Medicare Data
The Medicare data, including information on eligibility and claims, were merged with the MDS file with use of the Health Insurance Claim number of the beneficiaries. The eligibility file contained data on gender, date of birth, survival status (verified date of death), and insurance type (health maintenance organization or fee-for-service). The claims data included all health services claims such as hospitalizations, skilled nursing facility admissions, hospice, and home health agency bills. The match rate of SAGE data to Medicare eligibility is approximately 85%.¹³

On-line Certification Automated Survey Data
On-line Certification Automated Survey (OSCAR) data, which provide nursing-facility information, have also been merged to the MDS file by facility code.¹⁴ Facility administrators provided information on nursing-facility ownership, structure (e.g., size, number of beds), staffing levels, and a profile of residents in the facility on the day of the inspection, including functional deficits, nursing care needs, and receipt of "high-tech" nursing care. In addition, the OSCAR data have been merged with the most recent version of the Area Resource File5 (ARF) to provide contextual geographical information. The ARF is a compilation of multiple data sets used to characterize the population residing in all U.S. counties, including the level of economic and health care delivery resources available in that area, for example, the number of hospital beds, average occupancy rate, and wage rates. Currently, the match rates exceed 90%.¹⁵

Drug Data
As part of the MDS, staff coded up to 18 drugs taken within the seven days preceding the assessment. Nursing facility staff coded each drug according to the National Drug Coding (NDC) system using the 10,000 NDCs included in the MDS training manual¹⁶ or, when needed, the Physicians' Desk Reference.¹⁷ These unique 10-digit codes were then converted to 11 digits, according to the National Council on Prescription Drugs Program standards.

NDC codes are commercially oriented and do not contain any mechanism by which to group drugs according to ingredients or categories of ingredients. Therefore, we linked NDCs into a therapeutic classification scheme to enable research.¹⁸ NDC codes were translated into therapeutic classes and subclasses through the use of the Master Drug Data Base (MediSpan).¹⁹ MediSpan included over 88,000 generic drug products, products from regional manufacturers, and information on over 90,000 inactive drugs. The hierarchical identifier, the Generic Product Identifier (GPI) contained in MediSpan, was a 14-character field consisting of seven subsets, each providing increasingly more specific information about the drug. While MediSpan incorporated the American Hospital Formulary Service²⁰ (AHFS) system based on the pharmacologic group, drugs were also classified with comparable compounds in the same therapeutic class.

Summary of Drug Usage within the SAGE Database
Two analyses were performed to characterize the SAGE database. The first analysis determined the overall prevalence and method of prescribing (i.e., scheduled versus as-needed) of residents' medications in the database. Pulmonary medications were defined by the Medispan system as inhaled anticholinergics and anti-inflammatory drugs, sympathomimetics, xanthines, and various asthma combination products. Antihypertensive agents included angiotensin converting enzyme (ACE) inhibitors, adrenolytic agents, reserpine, alpha blockers, non-specific vasodilators, and all antihypertensive combination products. Beta blockers encompassed selective, non-selective and alpha-beta blockers. Antihyperlipidemics included bile acid sequestrants, fibric acid derivatives, statins, and miscellaneous agents. The hypoglycemic category covered both insulin and oral hypoglycemic agents. Antidepressant drugs included tricyclics, tetracyclics, selective serotonin re-uptake inhibitors, and monoamine oxidase inhibitors. Finally, histamine H₂-receptor antagonists, prostaglandins, proton pump inhibitors, gastrointestinal antispasmodic-anticholinergic agents, and antibiotic with bismuth combinations (specifically for peptic ulcer disease) comprised the anti-ulcer category. To determine prevalence, the numerator consisted of the number of people taking drugs in each drug category, while the denominator was the total study sample.

The second general analysis determined the prevalence of medications potentially inappropriate for use in the elderly nursing facility resident, as defined by Beers et al.^21,22 These criteria were used for the analysis because they are widely known in general medical practice, as well as in the specialty of geriatrics. The medications primarily included older drugs for which either safer (e.g., fewer anticholinergic properties, shorter half-life) or more effective alternatives exist. Differences among facilities with respect to the percentage of residents receiving at least one potentially inappropriate medication were also evaluated.

Statistical Analysis
The data for all residents in this report (n=363,354) were obtained at the time of their first assessment in the nursing facility between 1992 and 1995. Data were analyzed using SAS statistical software.²³

Results
The characteristics of the study population are shown in Table 1. Over 70% of the residents were women. Over 75% of the study population were 75 years of age and older. While most residents were white, African Americans and Hispanics comprised 10.2% and 2.5% of the population, respectively. Admission to the nursing facility was from acute-care hospitals in almost two-thirds of the cases, with only 20% of admissions from home. Although most residents had at least a moderate degree of impairment in physical functioning, over one-third were relatively cognitively intact. Residents were taking a mean (± SD) of 6.5 drugs (± 4.5) with a range of 0-18 different medications. Only 6.8% of residents were taking no medications.

The overall prevalence of medication use by age is shown in Table 2. As expected, cardiovascular drugs were the most commonly prescribed medications. In the broad Medispan category of antihypertensive agents, ACE inhibitors comprised approximately 70%-75% of the drugs in this category. The use of digoxin, diuretics, and ophthalmic agents were associated with increased age, being most commonly used by residents 85 years of age and older. Men and women did not differ in their use of the medications listed in the table (data not shown).

For agents in which 2% or more of those with the prescription had it ordered as needed (i.e., p.r.n.), the prevalence of scheduled versus p.r.n. medication use by age and gender is shown in Table 3. Eleven percent had used a narcotic analgesic either on a scheduled or p.r.n. basis; however, use on a scheduled basis was rare. Among all residents, 15% had a scheduled order for an antipsychotic agent, with only 2% using these drugs on a p.r.n. basis. Approximately 30% of residents had a standing order for a laxative. Men and women did not differ in their use of the medications either on a scheduled or p.r.n. basis.

The prevalence of medications, defined as potentially inappropriate, is shown in Table 4. Overall, 31.5% of residents used at least one inappropriate drug as defined by the Beers criteria. While this rate appears serious, only 8.6% of residents used a medication with a high potential for unnecessary adverse outcomes such as amitriptyline. The most commonly prescribed drugs included propoxyphene (12.8%), diphenhydramine (5.5%), and amitriptyline (3.2%). The prevalence of other potentially inappropriate medications ranged from 1%-2%.

The distribution of residents receiving at least one potentially inappropriate medication and at least one medication specifically with a serious risk are shown by facility in Figures 2 and 3, respectively. Most importantly, over 16% of facilities had 50% of their residents receiving at least one potentially inappropriate medication and 20% of their residents receiving a medication with a serious potential for an adverse outcome.

Figure 2. Distribution of Inappropriate Medication Use by Facility

Percent of Facilities Percent of Residents with an Inappropriate Medication

Figure 3. Distribution of Inappropriate Medication Use with Severe Potential by Facility

Percent of Facilities Percent of Residents with an Inappropriate Medication

Discussion
The SAGE database is a unique research tool for improving the care of elderly nursing facility residents. As expected, the population of the SAGE database primarily includes women over the age of 75 who use multiple medications and who exhibit varied functional and cognitive status. Cardiovascular agents, anti-ulcer medications and ophthalmic preparations are the most commonly prescribed medications. Overall, while 31.5% of residents use at least one inappropriate drug as defined by the Beers criteria, only 8.6% of residents use a medication with a high potential for an adverse outcome such as amitriptyline. Over 16% of facilities had 50% of their residents receiving at least one potentially inappropriate medication and 20% of their residents receiving a medication with a serious potential for an adverse outcome. Thus this report demonstrates that the SAGE database links many components of geriatric care including medication use, resident assessment, and facility data into one comprehensive system.

For the individual consultant pharmacist, research from the SAGE database has already provided specific information on unrecognized, or underappreciated, common prescribing problems potentially within their own nursing facility practices. The first major paper from the SAGE database²⁴ reported that the management of cancer pain in the elderly is inadequate, with 26% of individuals with daily pain receiving no analgesic therapy. More importantly, specific predictors of inadequate pain management were identified to include minority race, low cognitive performance, and the number of other medications the patient is receiving. These results have provided guidance to consultant pharmacists in developing programs within their practice to improve the prescribing of appropriate drugs (i.e., analgesics) and to reduce the use of psychotropic agents in elderly individuals who are cognitive impaired and perhaps combative because of underlying, untreated pain. Because the findings are population-based, including all nursing facility residents in five states, consultant pharmacists may be better able to generalize the study results to their practice, instead of relying on research that is limited to just a few nursing facilities that may have problems unique to those institutions. Finally, while functional and cognitive outcomes were not linked to medication use for the purpose of this report, the ability to do so is important information for the consultant pharmacist.

Large, automated, cross-linked databases such as SAGE permit research into the effects of drugs in populations traditionally underrepresented in randomized clinical trials such as the elderly nursing facility resident. However, any database to be used for research purposes must be evaluated with respect to completeness, quality of key patient identifiers, and follow-up, among other factors.^25,26 Only 5.4% of the original NDC codes were incomplete or incorrect and thus could not be converted into a Medispan code.13 Gender-specific medications had a match rate of over 90% with gender. For example, all residents taking tamoxifen were women, while only men had been prescribed goserelin. Women comprised 92% of estrogen users. Cross-linkages between drugs and MDS conditions indicated that 89% of residents using levodopa were recorded as having a Parkinson's disease diagnosis based on the MDS data, while 93.2% of individuals taking hypoglycemic agents had diabetes mellitus. These data support the accuracy and reliability of the MDS drug data.¹³

The SAGE data system does have limitations. First, clinical data such as blood pressure and laboratory values are not collected as part of the MDS assessment. However, specific information on diagnoses and functional status were compiled, and the data can be cross-linked to Medicare data to determine previous conditions. Second, the duration of drug use is not obtained on the drug inventories. Quarterly assessments of the drug inventory, however, do permit estimation of drug use throughout the follow-up period, and only medications such as antibiotics are clearly underestimated. Other potential limitations could include an incomplete listing of non-prescription products, especially if the resident were receiving more than 18 drugs, and the possibility of inaccurate reporting of drug use.

The SAGE data system is a unique data set in that it permits longitudinal assessments of medication use and changes in cognitive function, mood state, and physical function. Quarterly measurements of clinical care variables allow intensive study of an elderly population including the oldest of the old. In addition, the database may be used for evaluating the impact of regulations such as the prospective payment system, and the use of drugs as quality indicators.

References
1. OBRA (Omnibus Budget Reconciliation Act) 1987. U.S. Congress. Public Law 101-508.
2. Tessier EG. In: McCure JD, Tessier EG, and Gaziano P (eds). Geriatric Drug Handbook for Long-Term Care. Baltimore: Williams & Wilkins Co; 1993:165-74.
3. National Prescription Audit. Plymouth Meeting, PA: IMS America Ltd; 1978-1988.
4. National Disease and Therapeutic Index. Plymouth Meeting, PA: IMS America Ltd; 1978-1988.
5. Stambler H. The Area Resource File: a brief look. Pub Health Rep 1988;103:184-8.
7. Morris JN, Hawes C, Fries BE et al. Designing the national resident assessment instrument for nursing facilities. Gerontologist 1990;30:293-307.
8. Hawes C, Morris JN, Phillips CD et al. Reliability estimates for the Minimum Data Set for nursing home resident assessment and care screening. Gerontologist 1995;2:172-8.
9. Morris JN, Hawes C, Murphy K et al. Multistate Nursing Home Case Mix and Quality Demonstration Training Manual. Natick, MA: Eliott Press, 1992.
10. Morris JN, Fries BE, Mehr DR et al. MDS Cognitive Performance Scale. J Gerontol 1994;49:M174-82.
11. Lawton M, Brody E. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist 1969;9:179-86.
12. Mor V, Branco K, Fleishman J et al. The structure of social engagement among nursing home residents. J Gerontol 1994;50B:P1-8.
13. Gambassi G, Landi F, Peng L et al. Validity of diagnostic and drug data in standardized nursing home assessments: potential for geriatric pharmacoepidemiology. Med Care 1998;36:167-79.
14. Zinn JS, Mor V. Nursing home special care units distribution by type, share and facility characteristics. Gerontologist 1994;34:371-7.
15. Bernabei R, Gambassi G, Lapane K et al. Introducing functional outcomes in geriatric pharmaco-epidemiology: The SAGE data base (submitted).
16. Minimum Data Set Plus Training Manual, Natick, MA: Eliot Press, 1991.
17. Physicians' Desk Reference, Montevale, NJ: Medical Economics, 1996.
18. Pahor M, Chrischilles EA, Guralnik JM et al. Drug data coding and analysis in epidemiologic studies. Eur J Epidemiol 1994;10:405-11.
19. Master Drug Data Base (MDDB) Documentation Manual. Indianapolis: MediSpan Inc., 1995.
20. American Hospital Formulary Service. Drug Information. Bethesda, MD: American Society of Hospital Pharmacists, 1994.
21. Beers MH, Ouslander JG, Fingold SF et al. Inappropriate medication use in skilled nursing facilities. Ann Intern Med 1992;117:684-9.
22. Beers MH. Explicit criteria for determining potentially inappropriate medication use by the elderly. An update. Arch Intern Med 1997;157:1531-6.
23. SAS/STAT User's Guide, Version 6, 4th ed, Vol. 2, Cary, NC: SAS Institute Inc., 1989.
24. Bernabei R, Gambassi G, Lapane K et al. Management of pain in elderly patients with cancer. SAGE study group. Systematic Assessment of Geriatric Drug Use via Epidemiology. JAMA 1998;279:1877-82.
25. Stergachis AS. Evaluating the quality of linked automated databases for use in pharmacoepidemiology. In: Hartzema AG, Porta MS, Tilson HH (eds), Pharmacoepidemiology: An Introduction, 2nd ed. Cincinatti, OH: Harvey Whitney Books, 1991.
26. Roos LL, Nicol JP, Cageorge SM. Using administrative data for longitudinal research: comparisons with primary data collection. J Chronic Dis 1987;40:41-9.

Copyright © 1998, American Society of Consultant Pharmacists, Inc. All rights reserved.