Appendix A.

Data and Variables Used for Hospice in Nursing Facility Analyses

March 2000

This appendix was used for both Report 4 (Use of Medicare's Hospice Benefit by Nursing Facility Residents) and Report 5 ( Outcomes and Utilization for Hospice and Non-Hospice Nursing Facility Decedents).


Resident Assessment Instrument (the Minimum Data Set or MDS)

The Omnibus Budget Reconciliation Act of 1987 (OBRA '87) contained the most far-reaching revisions to the standards, inspection process and enforcement system in nursing facilities since the passage of Medicare and Medicaid in 1965 (Hawes, 1998). A major feature of this legislation was the introduction of a uniform, comprehensive resident assessment instrument (the MDS) to guide the clinical care planning process in order to systematically document residents' needs. The MDS is not only used to systematically assess the resident and to generate a comprehensive care plan to document clinical progress as that plan is implemented, but it is used by regulators to focus on resident outcomes and by facilities to improve their performance. In the time period studied here documentation of the resident assessments were required: at admission (by 15th day), quarterly (by 90th day), and annually (by 365th day). Reassessments were required when a resident was readmitted after hospital admissions and when significant change occurred. Resident assessments were to have been completed on all nursing facility residents cared for in facilities receiving any Medicare or Medicaid payment. These nursing facilities represent 96 percent of the facilities in the United States.

Topics covered in the MDS include cognitive function, communication/hearing problems, physical functioning, continence, psychosocial well-being, mood state, activity and recreation, disease diagnoses, health conditions/symptoms, nutritional status, oral/dental status, skin condition, special treatments, and medication use. A number of studies (Morris et al., 1994; Frederiksen et al., 1996; Hartmaier et al., 1994 & 1995; Phillips et al., 1993; Mor et al., 1994) demonstrate that researchers and clinicians using the MDS can achieve high levels of inter-rater reliability. Using an earlier version of the data set used in this study, Gambassi and colleagues found reasonably high levels of validity and good internal consistency comparing diagnoses on the MDS with HCFA claims and medical conditions with patterns of use of specific drugs (Gambassi et al, 1988). While the accuracy of the MDS data have been questioned (Berlowitz et al., 1997; Kramer et al., 199 ), and considerable anecdotal evidence reveals that some facilities have not taken the time to train their staff properly in its use, it is unlikely that misclassification errors in recording of information will be differential with respect to the outcomes of interest.

For the comparative study hospice and non-hospice patients had to have had at least 2 MDS assessments performed. The need for the presence of 2 MDS assessments for our comparative analyses was originally recognized after preliminary analysis documented the presence of ascertainment bias on selected symptoms. Specifically, hospice residents were significantly more likely to have pain and dyspnea recorded than were non-hospice decedents. For example, controlling for other patient factors, residents with a dementia diagnoses were 3 times as likely to have pain recorded than were non-hospice residents. Therefore, to more correctly represent the hospice influence on the presence and management of symptoms (rather than merely the increased likelihood of hospice to assess symptoms) we felt that it was necessary to control for the status of symptoms at the time of the penultimate MDS.

Health Care Financing Administration (HCFA) Claims Data

The HCFA claims data were merged to the MDS file using the Health Insurance Claim number of Medicare beneficiaries. To ensure confidentiality, these identifiers were replaced with unique identifiers using the claim number as a seed. Two files (beneficiary information and claims data) comprise the HCFA data. The beneficiary file (Denominator file) contains gender, date of birth, and survival status (verified date of death). The claims data used include all Medicare Part A claims including hospital, skilled nursing facility, hospice, and home health agency claims. We achieved a match rate of MDS data to HCFA beneficiary data of approximately 85 percent using HCFA data from 1991through 1997.

Drug Data

As part of the resident assessment, nursing facility staffs code up to eighteen drugs taken within the seven days preceding the assessment. Nursing home staffs code each drug according to the National Drug Coding (NDC) system. Field tests of the MDS showed that 100 percent of the medication use items were reliable with the average reliability being 0.73 (Hawes et al., 1995b). While prescription drug products must use an NDC, most non-prescription drugs are also primarily referenced by the NDC. NDCs are unique 11-digit codes that identify discrete drug products. The first five digits refer to the manufacturer. The next four digits correspond to the drug product. The last two digits indicate the packaging. As pharmaceutical companies merge, new products are introduced, and drugs are no longer active, changes in NDCs occur continuously. Consequently, the NDCs are commercially-oriented and do not contain any mechanism to group drugs according to ingredients or categories of ingredients. Therefore, linking NDCs to specific descriptive information is critical to enable research. This NDC matching entails several steps. To match the NDC codes, we used a historical reference archive for drug products that listed all NDCs ever attributed and eventually discontinued between 1991 and 1996. For scientific drug research, we translated NDC codes into a hierarchical therapeutic classification scheme as recommended by WHO (Pahor et al., 1994). NDC codes were merged to useable therapeutic class and sub-class information using the Master Drug Data Base (MediSpan™) (1995). MediSpan™ contains complete records for prescriptions common in retail pharmacy as well as unit-dose and injectables used by hospitals and external facilities. MediSpan™ now includes over 100,000 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™, is a 14-character field consisting of seven subsets, each providing increasingly more specific information about the drug. (See below example.) While MediSpan™ incorporates the American Hospital Formulary Service (AHFS) (1994), a classification system based on the pharmacological uses of drugs, MediSpan™ also groups drugs with comparable compounds in the same therapeutic class and allows the same drug to be classified into multiple therapeutic classes.

Medi-span™ Classification System - Example of an Antidepressant
GPI Coding Example
58- Drug group Antidepressants
58-20- Drug class Tricyclic agents
58-20-00- Drug sub-class --
58-20-00-60 Drug name Nortiptyline
58-20-00-60-10 Drug name extension Hydochloride
58-20-00-60-10-01 Dosage form 10mg

A recent study analyzed the MDS drug codes with respect to: (1) completeness; (2) internal consistency; and (3) external validity (Gambassi et al., 1998). Investigators found the overall match rate between the NDC and the MediSpan™ greater than 90 percent with only 5.4 percent of the original NDC codes contained in the MDS data from the states being studied in this project to be incomplete or incorrect. Gender-specific medications had a high concordance with gender (>90 percent). For example, all residents taking tamoxifen were women; all residents taking goserelin were men; and 92 percent of estrogen users were women. Cross-linkages between drugs and MDS condition variables revealed adequate to high rates of concordance (range: 51 percent (gout) - 100 percent (rheumatoid arthritis)). High rates of concordance were reported when cross-checking levo-dopa with Parkinson's disease (88.9 percent); hypoglycemic agents with diabetes mellitus (93.2 percent); and sore care products with pressure sores (83.7 percent). These data show that the MDS drug data are consistent and reliable (Gambassi et al., 1998).

Hospice Provider of Service File

The Provider of Service (POS) File is compiled and managed by HCFA to determine the capacity of Medicare/Medicaid institutional providers to render acceptable care. This file contains information on program characteristics, collected by State surveyors under Federal guidelines. Since the periodic inspection of hospices is not mandated, surveys are conducted according to state priorities and resources, and, as a consequence, hospice and nursing facility data does not match across time. We used the hospice provider number on the HCFA hospice claim to link the hospice provider information to MDS and claims data. For our descriptive analyses, we used 1995 hospice provider information. For our analyses we were most interested in hospice provider type (freestanding or home care, hospital or nursing facility based), and ownership.


Table of Variables

Table A1 lists all variables studied in both the descriptive and comparative hospice in nursing facility analyses. Variable measurement and data source are shown. For selected variables, the text below provides more information on their documentation and/or their reliability and validity.


Pain was not a major focus of the MDS version used in these states during the time period in question, and, as such, there are several limitations to the measure used in these analyses. A major limitation is that the level of pain intensity is not recorded; the newer MDS 2.0 requires documentation of pain intensity.

The MDS pain data used in this study is based on assessment by nursing personnel which is supposed to be performed according to instructions provided in the MDS manual (HCFA, 1991). Nursing home personnel are supposed to evaluate signs and symptoms of pain, but since pain is a subjective experience, they are instructed to record whatever the residents said it was. Residents were to be asked whether they had experienced any pain in the last seven days. Furthermore, residents were to be asked to describe the pain and how often it was manifest. To elicit complete and satisfactory answers, the assessors were instructed to ask neutral and non-directive questions. Questions such as: "What do you mean?" "Tell me what you have in mind." "Tell me more about that." "Please be more specific." "Give me an example". Moreover, the assessors were instructed to validate their understanding of what the resident was really saying. Statements like "I think I hear you saying that ….." or "Let's see if I understood you correctly. You said ….. Is it right?" were suggested in the MDS Instruction Manual.

For MDS assessment purposes, pain refers to any type of physical pain or discomfort in any part of the body experienced on a daily basis. If the assessor had difficulty discriminating the frequency, the instructions were to code as daily. Pain could have been localized or more generalized. It could have been acute or chronic, continuous or intermittent, occurring at rest or with movement. Pain recording could have depended exclusively on the observation of signs of pain. According to the MDS Manual, these include moaning, crying, and other vocalizations; wincing or frowning and other facial expressions; or body posture such as guarding/protecting an area of the body, or lying very still. In these cases, the assessors were instructed to ask the nurse assistants and therapists who might have been working with the resident, whether he/she had complaints or signs of pain during their shifts. In some residents, those who have dementia and cannot verbalize the pain experience, the assessor was instructed to look for particular behaviors such as calling out for help, pained facial expressions, refusing to eat, or striking out at a nurse assistant who was trying to move them or touch a body part.


Shortness of breath (dyspnea) is recorded on the MDS if the problem is present in the "last 7 days" prior to completion of the MDS assessment. The degree or frequency of dyspnea is not indicated, only its presence. We compare the presence or absence of dyspnea in hospice versus non-hospice decedents stratifying for its presence on the penultimate MDS assessment.

Persistent Mood Disturbance

Persistent mood disturbance is defined as "persistent sad or anxious mood that has existed over the last 7 days and was not easily altered by attempt to "cheer up" the resident." (MDS+ Reference Manual, 1993) For MDS assessment purposes, a sad or anxious mood is a distressed mood that is characterized by explicit verbal or gestural expressions of feeling depressed or anxious (or a synonym such as feeling sad, miserable, blue, hopeless, empty, or tearful). Assessors are instructed to draw upon their own interactions with the residents as well as to statements of direct-care staff, social workers, and licensed personnel who may have evaluated the resident in this area. Suggested cues are: Does the resident cry or look dejected (unhappy) when no one is talking with him or her? When you talk with the resident, does he or she sound hopeless, fearful, sad, anxious? Does the resident appear withdrawn, apathetic, without emotion? (MDS+ Reference Manual, 1993)

Cognitive Performance

The MDS includes seven direct measures of cognition: short and long term memory, recall or orientation items (season, location or room, staff names/faces, orientation to nursing home), and decision-making ability. Good reliability (0.7) of these items has been reported (Hawes et al, 1995b). The cognitive performance scale (CPS) used in this study is a categorical measure of cognition using these MDS items and several items which indirectly evaluate cognitive function (i.e. comatose state, total dependent eating) (Morris et al., 1994). Based on two standard cognitive assessment tools, the Mini-Mental State Examination (Folstein et al., 1975) and the Test for Severe Impairment (Albert & Cohen, 1992), the CPS has excellent reliability with estimates published in the range of 0.66 - 0.88 (159). Using the MMSE as the gold standard, the CPS has high sensitivity (>90 percent) and specificity (>85 percent), yielding high diagnostic accuracy, regardless of patient education level (Hartmaier et al., 1995). Furthermore, the CPS has excellent reproducibility (Kappa >0.76) (155).

Activities of Daily Living

The reliability of the ADL scores range from 0.87-0.92 (Hawes et al., 1995b) and is highly correlated (0.89) with the Physical Signs and Symptoms Scale (Lawton & Brody, 1969). Furthermore, a recent study found these measures useful in pharmacoepidemiologic studies (Bernabei et al., 1998).

TABLE A1. Comparative Descriptive and Analytic Analyses--Variables, Measures, and Data Sources Outcomes
Type Variable Empirical Measure Source Data
Patient Acute Care Hospitalization and Average Hospital Days
-- in 30 days prior to death
-- in 90 days prior to death
-- in 6 months prior to death

Hospice decedents in hospice for total time period studied (and their matched controls) who have acute care hospitalization/ hospice decedents in hospice for total time period studied (and their matched controls).

Days in hospital in time period for hospice decedents in hospice for total time period studied (and their matched controls)/ hospice decedents in hospice for total time period studied (and their matched controls).

Medicare Part A Hospital Claims
Pain Management (regular treatment) Decedents in pain and receiving WHO level analgesia at least twice a day, or, for level III drugs, having a drug patch. Resident Assessment and Drug Information
Persistent Mood Disturbance (treatment of) Decedents with persistent mood disturbance and receiving antiananxiety or antidepressant medication in the 5 to 7 days prior to MDS assessment data. Resident Assessment
Independent Variables
Type Variable Empirical Measure Source Dataset
Patient Hospice enrollment Resident elected hospice prior to nursing facility admission. Medicare hospice claims and Resident assessment
Covariates and Variables for Descriptive Comparisons
Type Variable Empirical Measure Source Dataset

Hospice Concentraction Unduplicated nursing facility residents on Medicare hospice in given year/ unduplicated nursing facility residents in a given year. Medicare hospice claims and Resident Assessment
For-profit ownership

Government ownership

Organizational type
A for-profit organization controls and operates the hospice.
A government entity controls and operates the hospice.
Hospice is freestanding or under administrative control of a hospital, home health agency or nursing home.
Hospice Provider of Service File

Level Variable Empirical Measure Source Dataset
Patient State of residence

Year of death
Race/ Ethnicity
State location of nursing facility in which resident resides
Year in which resident died
White/ Afr.Am./ Latino/ Native Am./ Asian
Resident Assessment
HCFA Denominator File
Gender Female/ Male     "
Age Years     "
Marital Status Married/ Widowed or Divorced/ Separated/ Never married - most recent status Resident Assessment
Activities of daily living ADL score - see text     "
Cognitive performance CPS score - see text     "
Body Mass Index Weight (kg)/ height2 (m2) Resident Assessment
Diagnosis Diagnostic categories of (1) Cancer without dementia (2) Cancer with dementia (3) Alzheimer's disease/ dementia, (4) Other Resident Assissments, Inpatient Claims 6 months before NF adm. & during stay
Individual selected diagnoses for multivariate analysis. Resident Assessment


Persistent Mood Disturbance

Complains or shows evidence of pain daily or almost daily (in 7 days prior to assessment).
Difficulty breathing occurring at rest, with activity, in response to illness or anxiety, or when lying flat.
Persistent sad or anxious mood that has existed over the last 7 days and was not easily attended by attempt to "cheer up" the resident.
Vomiting in 7 days prior to assessment.
Resident Assessment


Analgesic consumed

IM medications
IV medications

Tube feedings


Analgesic level received by WHO I, II and III levels, and daily frequency for multivariate analysis.
Any drug given intramuscularly.
Any drug or biological (e.g., contrast material give by IV push or drip) in the 7 days prior to assessment.
Presence of any tube that can deliver food/ nutritional substances, other directly into the gastrointestinal system.
Any use of truck or limb restraints or chair that prevents rising in 7 days prior to assessment.
Any speech, occupational or physical therapy in 7 days prior to assessment.
Resident Assessment and Drug Information

Resident Assessment



Advance directives--
Do not resuscitate
Do not hospitalize
Feeding restrictions
Medication restrictions
Other treatment restrictions
Documentation of preference must also be present in resident's healthcare record.     "
Average Expenditures--
  Routine home care
  Continuous home care
  Respite inpatient
  General inpatient
  Physician visits
Medicare Part A
  Acute care hospital
  Skilled nursing
  Home health care
Total expenditures per category divided by decedents (hospice, non-hospice and total).
-- for 30 days (stratified by hospice lengths of stay of: <30 days and 30+ days)
-- for last 6 months (stratified by hospice lengths of stay of: <30 days, 30-59 days, 60-119 days and 120+ days)
Non-hospice decedents are placed in the length of stay category of the hospice case to which they are matched.
Medicare claims.


Albert, M. and Cohen, C. The Test for Severe Impairment: An instrument for the assessment of patients with severe cognitive dysfunction. J Am Geriatr Soc. 1992 May; 40(5):449-53.

American Hospital Formulary Service (AHFS)Drug Information. Bethesda, MD: American Society of Hospital Pharmacists; 1994.

Berlowitz, D. R.; Brandeis, G. H., and Moskowitz, M. A. Using administrative databases to evaluate long-term care. J Am Geriatr Soc. 1997 May; 45(5):618-23.

Bernabei, R.; Gambassi, G.; Lapane, K.; Landi, F.; Gatsonis, C.; Dunlop, R.; Lipsitz, L.; Steel, K., and Mor, V. Management of pain in elderly patients with cancer. SAGE Study Group. Systematic Assessment of Geriatric Drug Use via Epidemiology [see comments]. JAMA. 1998 Jun 17; 279(23):1877-82.

Folstein, M. F.; Folstein, S. E., and McHugh, P. R. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975 Nov; 12(3):189-98.

Frederiksen, K.; Tariot, P., and De Jonghe, E. Minimum Data Set Plus (MDS+) scores compared with scores from five rating scales. J Am Geriatr Soc. 1996 Mar; 44(3):305-9.

Gambassi, G.; Landi, F.; Peng, L.; Brostrup-Jensen, C.; Calore, K.; Hiris, J.; Lipsitz, L.; Mor, V., and Bernabei, R. on behalf of the SAGE study group. Validity of diagnostic and drug data in standardized nursing home resident assessment: Potential for geriatric pharmacoepidemiology. Medical Care. 1998; 36(2):167-179.

Hartmaier, S. L.; Sloane, P. D.; Guess, H. A., and Koch, G. G. The MDS Cognition Scale: A valid instrument for identifying and staging nursing home residents with dementia using the minimum data set [see comments]. J Am Geriatr Soc. 1994 Nov; 42(11):1173-9.

Hartmaier, S. L.; Sloane, P. D.; Guess, H. A.; Koch, G. G.; Mitchell, C. M., and Phillips, C. D. Validation of the Minimum Data Set Cognitive Performance Scale: Agreement with the Mini-Mental State Examination. J Gerontol A Biol Sci Med Sci. 1995 Mar; 50(2):M128-33.

Hawes, C. Regulation and the politics of long-term care. Generations Journal of the American Society on Aging. 1998; 21(4):5-9.

Hawes, C.; Morris, J. N.; Phillips, C. D.; Mor, V.; Fries, B. E., and Nonemaker, S. Reliability estimates for the Minimum Data Set for nursing home resident assessment and care screening (MDS). Gerontologist. 1995b Apr; 35(2):172-8.

Health Care Financing Administration (HCFA). Minimum data se plus multistate nursing home case mix and quality demonstration training manual. Natick, MA: Eliot Press; 1991.

Lawton, M. and Brody, E. Assessment of older people: Self-maintaining and instrumental activities of daily living. The Gerontologist. 1969; 9179-186.

Mor, V.; Allen, S., and Malin, M. The psychosocial impact of cancer on older versus younger patients and their families. Cancer. 1994 Oct 1; 74(7 Suppl):2118-27.

Morris, J. N.; Fries, B. E.; Mehr, D. R.; Hawes, C.; Phillips, C.; Mor, V., and Lipsitz, L. A. MDS Cognitive Performance Scale. J Gerontol. 1994 Jul; 49(4):M174-82.

Pahor, M.; Chrischilles, E. A.; Guralnik, J. M.; Brown, S. L.; Wallace, R. B., and Carbonin, P. Drug data coding and analysis in epidemiologic studies. Eur J Epidemiol. 1994; 10(4):405-411.

Phillips, C. D.; Chu, C. W.; Morris, J. N., and Hawes, C. Effects of cognitive impairment on the reliability of geriatric assessments in nursing homes. J Am Geriatr Soc. 1993 Feb; 41(2):136-42.

Return to:
  • Outcomes and Utilization for Hospice and Non-Hospice Nursing Facility Decedents
  • Use of Medicare's Hospice Benefit by Nursing Facility Residents