The Residential History File: Studying Nursing Home Residents' Long-Term Care Histories*

RESEARCH ARTICLE

The Residential History File: StudyingNursing Home Residents’ Long-TermCare Historiesn

Orna Intrator, Jeffrey Hiris, Katherine Berg, Susan C. Miller, andVince Mor

Objective. To construct a data tool, the Residential History File (RHF), that summa-rizes information from Medicare claims and nursing home (NH) Minimum Data Set(MDS) assessments to track people through health care locations, including non-Med-icare-paid NH stays.Data Sources. Online Survey of Certification and Reporting (OSCAR) data for 202free-standing NHs, Medicare Denominator, claims (parts A and B), and MDS assess-ments for 60,984 people who were present in one of these NHs in 2006.Methods. The algorithm creating the RHF is outlined and the RHF for the study dataare used to describe place of death. The identification of residents in NHs is comparedwith the reports in OSCAR and part B claims.Principal Findings. The RHF correctly identified 84.8 percent of part B claims withplace-of-service in NH, and it identified 18.3 less residents on average than reported inthe OSCAR on the day of the survey. The RHF indicated that 17.5 percent non-Medicare NH decedents were transferred to the hospital to die versus 45.6 percentskilled nursing facility decedents.Conclusions. The population-based design of the RHF makes it possible to conductpolicy-relevant research to examine the variation in the rate and type of health caretransitions across the United States.

Key Words. Medicare, Minimum Data Set (MDS), transitions in care settings,linking administrative files, tracking health care utilization

The adoption of prospective payment, first for hospitals in the early 1980s andthen postacute settings a decade ago, created conflicting payment silos, withhospitals reducing length of stay and postacute providers accepting complexpatients likely to be rehospitalized. These conflicting Medicare reimbursement

nThe views expressed in this article are those of the authors and do not necessarily reflect theposition or policy of the Department of Veterans Affairs or the United States government.

r Health Research and Educational TrustDOI: 10.1111/j.1475-6773.2010.01194.x

120

Health Services Research

incentives have been associated with high rates of transitions between pro-viders because there are no consequences for maximizing reimbursements inthis way. In particular, patients who are disabled or chronically or terminallyill, who are often served in nursing homes (NH) as their main long-term careprovider have been subject to the consequences of the conflicting reimburse-ment incentives and have thus suffered from multiple transitions. In spite ofthe increasing recognition of the importance of care transitions among long-term care residents, most of the literature has concentrated on reporting totalutilization per service type (mostly either inpatient or Medicare-paid skillednursing facility [SNF] care) (Coleman et al. 2004; Mor et al. 2010). Even therecent focus on geographic variation in Medicare costs has emphasized re-gional and hospital differences in the average intensity of inpatient use ratherthan the extent of variation in patients’ experiences across the continuum ofcare within and between geographic areas (http://www.dartmouthatlas.org).

Historically it has been difficult to assemble patient histories using ex-isting claims data for those discharged from hospital to postacute settingsbecause conflicting reimbursement incentives also translate to disparate re-imbursement systems and therefore administrative data systems. Thus, com-posing transition histories using only Medicare claims does not provideinformation on long-term NH care (Burton et al. 1995; Brown et al. 1999;Cooper et al. 2000). Alternatively, using only NH federally mandated regularassessment of all NH residents using the Minimum Data Set (MDS) residentassessment instrument data provides limited information about transitionsoutside of the NH (Coburn, Keith, and Bolda 2002).

The absence of data on NH use may lead to misleading conclusions.Some studies of Medicare expenditures at the end of life found those to belower for older than younger persons (Gornick, McMillan, and Lubitz 1993;Levinsky et al. 2001). However, Roos, Montgomery, and Roos (1987), using aCanadian longitudinal data set that included data on both acute and long-termcare utilization, found end-of-life public health care utilization expendituresdid not decrease with age. The discrepancy between the United States and

Address correspondence to Orna Intrator, Ph.D., Associate Professor (Research), Center forGerontology and Health Care Research, Brown University, PO Box G-S121-6, Providence, RI02912; e-mail: [email protected]. Orna Intrator, Ph.D., Health Research Scientist, VAHSR&D REAP, Providence, RI. Jeffrey Hiris, M.A., Senior Systems Analyst, Susan C. Miller,Ph.D., Associate Professor (Research), Vincent Mor, Ph.D., Professor, are with the Center forGerontology and Health Care Research, Brown University, Providence, RI. Katherine Berg,Ph.D., P.T., Associate Professor and Chair, is with the Department of Physical Therapy, Universityof Toronto, Toronto, ON, Canada.

The Residential History File 121

http://www.dartmouthatlas.org

mailto:[email protected]






Canadian findings may be explained by the absence of NH stays in the U.S.research and highlight the need for accurate information on all utilizationwhen studying expenditures and budgeting for care.

The purpose of this paper is to describe the creation of a ‘‘residentialhistory file’’ (RHF), using an algorithm that links Medicare claims and NHMDS assessments that results in a dataset (the RHF) which tracks the timingand location of health service use. Initially, we developed this method to trackpostacute care for patients hospitalized for hip fracture or stroke (Intrator et al.2003). Subsequently we expanded the method to other applications rangingfrom studies of hospice use among Medicaid NH residents to tracking post-hospitalization disposition of NH residents (Miller et al. 2004; Intrator et al.2007). In this paper we present the structure of the RHF algorithm and apply itto a cohort of all Medicare beneficiaries who were in one of 202 free-standingnonpediatric NHs at some time in 2006. We describe the resulting RHF,present an alternative RHF based only on MDS data, and conduct an illus-trative analysis of a study of place of death. We then present results fromcomparisons identifying patients in NHs based on the RHF versus on OnlineSurvey of Certification and Reporting (OSCAR) and on the place of servicecodes recorded on Medicare part B claims.

METHODS

Administrative data are a compelling source for the study of population-widehealth care utilization, patient outcomes, and organization and system eval-uation. In the United States several administrative data sources are available toresearchers from the Center for Medicare and Medicaid Services (CMS). Weused Medicare claims and NH resident assessments and a facility-levelresource, the OSCAR, which reports results of the annual certification ofNHs and contains information regarding NH deficiencies as well as infor-mation about the NH residents in aggregate. In particular, we used the re-ported total number of NH residents and the number of Medicareresidents on the day of the survey.

MDS Data from NH Residents’ Assessments

Filing of the MDS resident assessment instrument is mandated for every NHresident at admission, quarterly, annually, at discharge, and at any significantchange of care. The instrument contains nearly 400 data elements which havebeen combined to create composite outcome scales, quality indicators, and

122 HSR: Health Services Research 46:1, Part I (February 2011)

various case mix and risk-adjustment measures (Fries et al. 1994; Hawes et al.1995; Mor et al. 1995, 2003; Phillips and Morris 1997; Gambassi et al. 1998;Hirdes, Frijters, and Teare 2003; Grabowski, Angelelli, and Mor 2004; Moret al. 2004; Wu et al. 2005). Since 1999 the MDS data have been used todetermine level of payment of Medicare billable SNF services using 44–53Resource Utilization Groups (Fries et al. 1994). Because the MDS is used todetermine Medicare SNF payment levels, special Medicare MDS assessmentsare conducted more frequently during the first several weeks of a SNF stay toproperly adjust payment levels to residents’ care needs. Since 2002 the MDSdata have been used to publically report quality of NH care (http://www.medicare.gov/nhcompare.). The use of MDS for both payment andquality monitoring requires that the MDS is done in a timely manner and ableto withstand audit.

Medicare Eligibility and Claims Data

Medicare’s routine administrative databases include detailed demographic,financial, and clinical data. All records in the claims and enrollment filesinclude unique identifiers for Medicare enrollees that allow longitudinal link-age, enabling detailed descriptions of the specific clinical services provided topatients. Enrollment files include demographic data (birth date, age, gender,race, and place of residence), eligibility information (in particular, Medicareparts A and B eligibility and periods of HMO enrollment), and vital status(date of death [DOD]). Medicare part A, institutional claims files, have beenused extensively in research. Generally, they contain information on dates ofclaims, diagnoses, services provided, charges, and reimbursements. Outpatientclaims may be used by an NH to bill for skilled services for residents who hadexhausted their part A SNF benefits or were otherwise ineligible for SNF care.

Part B claims have been used less frequently in research. Like part Aclaims, these claims contain information on the date of service, charges,services provided, and reimbursements. They also provide information on theplace of service. Among the 50 or so place of service codes two indicate NHlocation (SNF and nursing facility). One study examined the utility of part Bclaims place of service code to identify NH utilization (Iwashyna 2003). Usingself-reported utilization data from the 1993 Medicare Current BeneficiarySurvey (MCBS) for validation, it reported that the place of service code was89.7 percent sensitive and 97.7 percent specific in detecting any NH utilizationwithin a year. Although an impressive rate over a 1-year period of time, thatpaper did not attempt to define the duration of an episode of NH utilization


http://www.medicare.gov/nhcompare




nor to validate the accuracy of the actual place of service. We examine thesensitivity and specificity of the RHF (as the test) in identifying part B claims inthe NH (assumed as the gold standard).

RHF Methodology

The goal of the RHF is to create a per-person chronological history of healthservice utilization and location of care within a prespecified calendar (e.g.,throughout a calendar year). The first step of the algorithm assigns utilizationsand associated locations to days in a calendar. Depending upon the type ofclaim, the basic information from a claim is the location of care (e.g., hospital,NH, emergency room, and home) type of provider (e.g., free-standing, hos-pital based, or swing bed SNF), and service type (e.g., hospice, SNF). Theorder of information entered into the daily calendar structure which controlsthe RHF (the data hierarchy) gives precedence to the records with dates thatare most likely to be complete and accurate because Medicare paymentsdepend upon them. Thus, inpatient claims are first filled into dates of thecalendar followed by outpatient emergency department (ER) and observationdays. Next SNF claims are entered onto days, followed by outpatient claimsfor skilled nursing service in a nursing home, and lastly home health claims arefilled into days. The above claims are location specific. Hospice claims are notlocation specific, because hospice can be provided in community or institu-tional settings. Thus, the location of hospice care is defined using other dataand includes hospice at home and hospice in NH. Consecutive days with thesame location and provider form episodelets of care.

Once the calendar is populated by all information obtained from claims,remaining nonfilled days may be populated by projected NH days based onMDS assessment dates and type. For example, admission assessments arerequired within 2 weeks of admission; therefore, the RHF fills up to 14 daysback during consecutive ‘‘gap’’ days to form an NH stay. Quarterly assess-ments are required every 3 months; thus, any gap days within 3 monthspreceding the quarterly assessment will be filled in the calendar as NH. An-nual assessments are required each calendar year, around the time of theclosest designated quarterly assessment. Discharge tracking assessments arerequired by CMS and are used to determine date of NH discharge, whenpresent. A full list of MDS rules is available from the authors.

The DOD is determined using information from the Medicare Denom-inator file and claims. Death is recorded as occurring on the last RHFepisodelet, thus enabling an easy identification of place of death.


We note that the RHF only summarizes the existing data and that severalepisodelets may need to be linked to define an ‘‘episode of care’’ which isrelevant to the research question. For example, if the total number of NH daysis of interest, several types of NH episodelets (e.g., SNF and MDS type) mayneed to be joined to span the complete time in NH.

Example 1: An RHF-Based Patient History

We present an example of a history constructed for a fictional Medicare ben-eficiary. A home health care claim was made on behalf of this individualspanning from January 1 through January 10. However, an ER outpatientclaim was made on January 3, and an inpatient claim was made from January 8through January 11. This was followed by a Medicare SNF claim January 11–15. The NH conducted an admission MDS assessment on January 12 and thena Medicare specified assessment on January 27. An MDS discharge trackingform was filed on January 31. Hospice filed a claim January 26–31, and an-other inpatient claim was made for January 30–31. The Denominator filereported a verified DOD on January 31.

Figure 1 is a diagram of the process of filling a calendar for the month ofJanuary for this patient. First circles are entered to mark inpatient care (Jan-uary 8–10, 30–31), then triangles to mark SNF care (January 11–14). Nexthome health periods are entered on claimed days (January 1–10), although wenote that they overlap ER on January 3 and inpatient claims on January 8–10.Following home health claims, MDS assessments are entered with their type(January 12 admission 1A, January 27 Medicare required 7O, and January 31,discharge tracking 8D). Rules based on regulations infer NH days during gapdays (January 15–29). Hospice recorded on January 26–29 are hospice in NH,but on January 30–31 they are added as a secondary type of episodelet to theexisting inpatient stay. Finally, death is noted on January 31st from Denom-inator eligibility records. The table under the monthly calendar presents theresulting RHF records.

Cohort

The RHF algorithm was applied to a cohort of all Medicare beneficiariesidentified in any one of 202 free-standing NHs during 2006 collected foranother study (Katz et al. 2009). All Medicare eligibility records, parts A and Bclaims and MDS data, were obtained for these residents under a data useagreement (DUA) with CMS. The RHF algorithm was applied to these data tocreate an RHF that is described and used in the following sections.


Inpatient SAF Home Health SAF Hospice SAF M=MDS Assmnt

NH SNF SAF NH (Based on MDS) Outpatient ER Death

January 2007

S M T W Th F S

1 2 3 4 5 6

7 8 9 10 11 12

M(1A)

13

14 15 16 17 18 19 20

21 22 23 24 25 26 27

M(7O)

28 29 30 31

M(8D)

*

#

* #

Figure 1: Sequence of Filling Calendar Days with Claims and MDSInformation for Fictional Patient


Analyses

Site of death has been studied before using information from death certificates,or only to determine whether death occurred in the hospital. The RHF makesit possible to locate a person when s/he dies, allowing an examination of NHand home as places of death, and transfers in care sites at the end of life. Thoseare tabulated and the distributions described.

RHFs created using only MDS and Medicare Denominator data versususing all part A claims, Denominator and MDS data were compared in orderto gain an understanding of the difference in identifying days in NH for res-idents who are not Medicare beneficiaries or who are receiving MedicareManaged Care. Total days in each episodelet type per person were compared,and the distribution of the difference described.

We examined the completeness of the RHF in determining NH ep-isodelets by comparing it with two other sources. The first is the number ofresidents in an NH on the day of certification reported on the OSCAR. Thesecond is the place of service codes on Medicare part B claims.

Because the RHF can be based on all NH residents regardless of theirpayer source (all residents must be assessed), the number of residents identifiedin the RHF on the day of the survey should be comparable to that reported inthe OSCAR. Moreover, the OSCAR also includes a count of Medicare res-idents which should correspond to the number of residents receiving eitherSNF or hospice care in an NH on the day of the survey. The distribution of thedifference between the OSCAR and RHF reports is described.

NH place of service codes in part B claim, and the episodelete typereported in the RHF on the day of the part B claim, are crosstabulated. Resultsare presented with the claim/visit as the unit of analysis. Assuming that part Bplace of service indicating NH is the ‘‘truth,’’ we calculated the sensitivity andspecificity of the RHF to identify days in NH. Because part B visits are notconducted every day a resident is in an NH, we do not expect the specificity gobe good.

Notes. MDS assmnt, Minimum Data Set assessment (types are 1A, admission; 7O, other Medicare required; 8D,discharge); NH, nursing home; SAF, Medicare claims from standard analytic files; SNF, skilled nursing facility(Medicare paid nursing home care). Below calendar, upper panel shows the locations of care for the person inExample 1. Location variables are prefaced by HEE_, with variables presented for dates from and through, twooverlapping locations, and indicator of death in the episodelet. Lower panel shows the data segments (claims,MDS assessments, Denominator indicator of death), which pertain to each of the episodelets. Segment variablesare prefaced by HED_ with variables presented for dates of episodelet and segments and segment type (HED_TYPE).


RESULTS

Description of Study Data and Cohort

We received 263,040 MDS assessments from 59,810 unique residents in any ofthe 202 free-standing nonpediatric facilities during 2006 and 71,473 SNFclaims from 30,627 residents. A total of 61,479 residents had either SNF orMDS assessments, of whom 28,756 (46.8 percent) had both SNF claims andMDS assessments, another 1,774 (2.9 percent) had only SNF claims, and30,949 (50.3 percent) residents had only MDS assessments. Of the 61,479residents with either an MDS or an SNF claim during 2006, 132 were notMedicare eligible reducing the sample to 61,347 residents. The RHF alsoidentified 363 residents who had claims after death and who were removedfrom the RHF. Thus, the final RHF dataset included 60,984 individuals.

The final RHF included 506,477 episodelets lasting an average of 38.7days (median 11 days). Among the cohort residents, 456 (0.7 percent) had nopart A eligibility throughout the year, 683 (1.1 percent) had no part B eligibilitythroughout the year, and 10,446 (17.1 percent) received Medicare benefitsthrough a managed care organization some time throughout 2006, with 6,016(9.9 percent) having MCO coverage throughout 2006.

The average length of stay in the hospital was 7.6 days, with a median of5 days. The durations of SNF, MDS-based, and outpatient-based NH epi-sodelets were 24.3 (median 5 18), 55.6 (median 5 7), and 13.9 (median 5 10)days, respectively.

Example 2: Site of Death

Overall, 15,341 (25.2 percent) residents died during the year. Death in thehospital occurred for 2,974 patients (19.4 percent). NH was the place of deathfor 9,134 patients (59.5 percent) with 2,423 (15.8 percent) dying while receiv-ing SNF care and 3,450 (22.5 percent) while receiving hospice care. Theremaining 3,233 decedents (21.1 percent) died at home. Residents who died inthe ER or during an observation stay (n 5 480, 3.1 percent) were assigned aplace of death based on the location they originated from. The majority 390(2.5 percent) died after being transferred to the ER from NH and the rest (90)died after being transferred to the ER from home.

The RHF allows us to identify the location before the site of death(Table 1). Interestingly, 1,485 of 3,233 residents who died at home (45.1percent) were in an NH before that. Among residents who died in the hospital19.1 percent were transferred from home, 68.2 percent from an NH (36.1


Tab

le1:

Tra

nsi

tion

sfr

omth

eL

ocat

ion

bef

ore

the

Loc

atio

nof

Dea

thto

Loc

atio

nof

Dea

thfo

r15

,341

Coh

ortD

eced

ents

Loc

atio

nat

Tim

eof

Dea

th,N

/Row

%

Loc

atio

nbe

fore

loca

tion

atti

me

ofde

ath

Hom

eH

ome

Hos

pice

Hos

pita

lN

HN

HSN

FN

HH

ospi

ceT

otal

Hom

e24

111

.650

024

.255

026

.656

527

.317

58.

539

1.9

2,07

0H

ome

hos

pic

e38

4.3

279

31.3

171.

92

0.2

252.

853

059

.589

1H

osp

ital

942.

962

319

.137

711

.525

17.

71,

645

50.3

280

8.6

3,27

0N

H87

116

.258

1.1

941

17.5

1.87

134

.835

0.7

1,60

329

.85,

379

NH

SNF

371.

624

1.0

1,07

645

.652

122

.153

722

.816

36.

92,

358

NH

hos

pic

e38

2.8

430

31.3

130.

951

3.7

60.

483

560

.81,

373

Tot

al1,

319

8.6

1,91

412

.52,

974

19.4

3,26

121

.32,

423

15.8

3,45

022

.515

,341

Not

e.L

ocat

ion

ofd

eath

(col

umn

s),a

nd

loca

tion

bef

ore

loca

tion

ofd

eath

(row

s).

NH

,nur

sin

gh

ome;

SNF

,ski

lled

nur

sin

gfa

cilit

y.


percent from SNF and 32.1 percent from non-SNF), while 14.6 percent ofresidents who died in the NH were transferred from home.

Example 3: Comparison of Full RHF and Resident History File Based Only onMDS Data

We compared the ‘‘full RHF’’ created from all Medicare parts A claims, De-nominator, and MDS data with an ‘‘MDS only RHF’’ created only from MDSand Denominator data in terms of the total duration in MDS-identified NHstay, any NH stay, and gaps, per person (Table 2). On average, the MDS-onlyRHF listed more MDS days than the full-RHF (127 versus 101 MDS days,respectively). Correspondingly, there were many more gap days in the MDS-only RHF than in the full RHF (238 versus 181 days, respectively). However,when comparing the total number of NH days of any type in the full RHFversus those accounted for by MDS only, the difference was very small, onaverage 1 more NH day in the full RHF versus the MDS-only RHF (inter-quartile range [IQR] 0–2 days).

Comparison 1: RHF versus OSCAR Number of Residents on Day of Certification

Among the 202 NHs in the cohort 138 were surveyed in 2006. The OSCARreported an average of 130.8 (IQR 85–159) residents in the NH on the day ofthe survey, and 18.6 (IQR 8–26) residents with Medicare as their primarypayer. Using the RHF we identified an average of 112.5 (IQR 65–134) res-idents in the NH on the day of the survey, and an average of 21.0 (IQR 10–29)residents receiving Medicare paid care. On average, the RHF identified 18.3(IQR 6–24) less residents than did OSCAR and the RHF identified 2.4 (IQR0–6) more residents on Medicare than OSCAR.

Comparison 2: RHF NH Episodelets and Part B Claims Place of Service Code

We received a total of 5,365,457 part B claims for 53,527 beneficiaries fromour sample (91.1 percent of residents in the cohort). Of these claims 917,059(17.1 percent) reported place of service as an NH for beneficiaries eitherreceiving SNF care or other non-SNF NH care. On the other hand, 1,920,728(35.8 percent) part B claims had a date of service corresponding to when theRHF identified the beneficiary as residing in an NH. Table 3 shows the cross-tabulation of place of service and location of resident on the day of the part Bclaim based on the RHF episodelets.

Among the 917,059 claims with place-of-service NH, 777,628 were ob-served in an NH by the RHF. Thus, the sensitivity of the RHF in identifying


Tab

le2:

Com

par

ison

bet

wee

nM

DS-

On

lyan

dF

ullR

HF

sin

Ter

ms

ofA

vera

geN

umb

erof

Day

sP

erR

esid

ent

Day

sin

Ful

lRH

FM

DS-

Onl

yR

HF

Diff

eren

ce:

Ful

l-M

DS

Onl

y

Mea

n(S

D)

Med

ian

(Q1,

Q3)

Mea

n(S

D)

Med

ian

(Q1,

Q3)

Mea

n(S

D)

Med

ian

(Q1,

Q3)

MD

Sep

isod

elet

s10

1(1

18)

26(1

,239

)12

7(1

15)

78(2

4,25

5)�

26.2

(35.

4)�

12(�

41,0

)A

ny

NH

epis

odel

ets

128

(115

)78

(24,

257)

127

(115

)78

(24,

255)

0.9

(14.

3)0

(�2,

0)G

ap18

1(1

23)

162

(73,

310)

238

(115

)28

7(1

10,3

41)

�57

.0(9

3.5)

�12

(�58

,0)

MD

S,M

inim

umD

ata

Set;

NH

,nur

sin

gh

ome;

RH

F,R

esid

enti

alH

isto

ryF

ile.


NH is 84.8 percent (777,628/917,059 part B claims). The specificity is 74.3percent (1� [917,059–777,628]/[5,565,457–1,920,186] part B claims).

DISCUSSION

The Residential History methodology provides extremely useful informationfor many health care utilization research studies. By tracking peoples’ healthcare utilization over time, the RHF provides a longitudinal picture of utilization,making possible the examination of access barriers, and discontinuity of care.There are many important research questions that require this technology inorder to be conducted; for example, identifying a cross-section of NH residentsat a particular day to describe their conditions (http://www.ltcfocus.org). An-other example is identifying 30-day rehospitalizations from NH, currentlyhotly debated by policy makers, requires that NH admission be verified after anindex hospitalization, and that rehospitalization only from NH be identified.While some of these differentiations can indeed be done without the benefit ofthe RHF, having it available makes creating such complex analyses files muchmore efficient. Even more important, the RHF summarizes information andknowledge about Medicare claims and MDS records that have been obtainedand accumulated by experience, and thus provides a single resource for uti-lizing these components of data together. The RHF framework also provides a

Table 3: RHF Location at Time of Part B Claims versus Part B Place ofService (row percents)

RHF Location

Nursing Home Hospital HomeTotal

Place of Service N % N % N % N

31. SNF 370,659 94.8 7,553 1.9 12,916 3.3 391,12832. NH 496,969 94.5 6,566 1.2 22,396 4.3 525,93121. Inpatient 219,087 11.8 1,506,187 81.0 134,231 7.2 1,859,50522. Outpatient hospital 52,056 27.4 31,964 16.8 106,136 55.8 190,15623. ER hospital 44,475 19.8 113,182 50.4 66,704 29.7 224,36112. Home 8,652 16.2 670 1.3 44,210 82.6 53,53211. Office 173,067 16.3 20,075 1.9 870,104 81.8 1,063,24681. Independent lab 364,903 55.8 8,728 1.3 280,323 42.9 653,954Other 1,057,946 80.1 123,922 9.4 138,835 10.5 1,320,703Total 1,920,186 35.8 1,804,728 33.6 1,640,543 30.6 5,365,457

ER, emergency room; NH, nursing home; RHF, Residential History File; SNF, skilled nursingfacility.


http://www.ltcfocus.org










method to adjudicate overlapping claims. For example, determining DODrequires working with information from both Denominator and claims withknowledge-based algorithms designed to reconcile differences. Moreover, thecomparison of the RHF based on MDS and claims data to the RHF based onlyon MDS and Denominator data reveals that the RHF is able to track NHlocation quite accurately for those residents who are not Medicare beneficiaries(approximately 5 percent of the NH population) or who have elected to receiveMedicare Managed Care (approximately 17 percent).

The RHF algorithm resulted in an RHF file with much face validity.Compared with part B place of service, it was almost 85 percent sensitive inidentifying NH location. Moreover, while it is expected that episodelets inMDS-only RHF would be longer than episodeletes in the full RHF given thatpart A claims provide more transfer information, it was reassuring that thetotal number of NH days identified by the two methods was similar.

Indeed, the study of site of death presented in this paper points to com-plexities in studying site of death. Using the RHF, a more nuanced under-standing of end-of-life care is gained; for example, among decedents receivingSNF care before their site of death, 49.5 percent were transferred to the hos-pital to die. This compares to 18 percent of non-SNF NH decedents, raisingthe question, why that great a difference?

When comparing the RHF location with an NH place of service on thepart B claims, we found the RHF correctly identified 84.8 percent of the NHpart B claims. The majority of the part B dates not identified to be in the NHwere identified in ‘‘gap’’ (presumably, community not receiving institutionalcare). This suggests that there is a potential to extend the RHF and increase theidentification of NH stays using the part B claims. Indeed, if part B claimsindicating the NH as the place of service infers that the existing RHF epi-sodelet is in an NH, we find that 94.1 percent of NH episodelets had a part Bstay with place of service NH.

The comparison of the number of residents observed on the day of theOSCAR survey with the OSCAR report indicates that, on average, the RHFreported 14 percent less total residents and 12.9 percent more residents re-ceiving Medicare-paid (SNF or hospice) care than the OSCAR. However,based on the OSCAR, almost 10 percent of facilities a year have a discrepancyin the number of residents of at least 20 percent from the prior year, and almost60 percent of facilities each year have at least 20 percent difference in totalMedicare residents. Thus, it appears that the OSCAR report may be moreerroneous than the report based on the RHF. Several other studies havequestioned the validity of the OSCAR data (OIG 2003; Feng et al. 2005).










Others versions of the RHF reported appear to be more limited and notwell documented (Sood, Buntin, and Escarce 2008). One recent attempt tocreate a file that identifies periods of NH care has been conducted internallyby CMS (L. ‘‘Spike’’ Duzor, personal communication). The ‘‘Stay File’’ iden-tifies periods of NH stay and adjoining hospitalizations. It uses MDS data todetermine periods of NH stays by using the assessment types recorded on theMDS and their corresponding dates. Unlike the RHF, this file was created inan attempt to examine NH utilization in isolation of other types of utilizationand only uses the MedPAR data to determine hospitalizations surrounding orduring NH stays. Based on our experience with the RHF algorithm, it wouldappear that the use of more limited Medicare data may create a file that wouldidentify all NH days, but would incorrectly identify additional days as NHdays. Therefore, it is not likely that this file will be as useful in examiningtransitions in care and discontinuity of care.

Several limitations are noted. The RHF methodology is usable only forresearchers who have an approved DUA with CMS to use Medicare standardanalytic files and NH MDS data. Although the information provided by theRHF is comprehensive, it pertains mainly to the Medicare fee-for-service pop-ulation and does not include information of out-of-pocket expenses estimated tobe about 17 percent of total expenses for Medicare eligible population (MEPS2006). The program that implements the methodology is very computationallyintensive and quite complex spanning over 10,000 lines of SAS code.

The RHF methodology can be useful for many studies that have linkedsurvey data with Medicare claims and MDS data. For example, the MCBS, theSEER cancer registry, the Health and Retirement Survey, and the currentlyplanned National Health and Aging Trends Study all routinely link to Medicareclaims. MCBS adds MDS and OASIS data since 1999, building a crude timelinefile of MDS and OASIS stays (F. Epig, personal communication). All thesestudies can benefit from knowledge on transition sequences and NH stays.

There are several ways to share this algorithm; one is to make it availableas is on our website. However, the SAS program is very complex and wouldrequire a lot of support at our institution, which we cannot budget. Anothermethod is to provide this code to CMS and to allow researchers to ask in aDUA that an RHF be created for them. CMS could also incorporate the RHF,or a reduced version of it, into the Chronic Condition Warehouse, which isintended to serve a somewhat similar, although less dynamic purpose.

Increasingly, the U.S. health care system will need to provide care forfrail, older persons, many with a terminal disease trajectory of chronic, pro-gressive illnesses with prolonged periods of functional dependency. Our cur-


rent health care reimbursement system, for the most part, is based on fee-for-service payment to individual institutions, with recent policy providing in-centives for cost containment. These incentives, in part, resulted in shorterhospital stays and a higher rate of transitions to SNFs. The population-baseddesign of the RHF makes it possible to conduct policy-relevant research toexamine the variation in the rate and type of health care transitions in theUnited States, examine the role of state policies and market characteristics ontransition rates, and the impact on beneficiaries residing in geographic regionwith differing rates and types of transitions.

ACKNOWLEDGMENTS

Joint Acknowledgment/Disclosure Statement : Supported in part by NationalInstitute on Aging grants R01 AG-14427, R01 AG 020557, R21 AG 030191,and P01 AG027296, and Agency of Healthcare Research and Quality R01HS10549. Special thanks to Julie Lima, Ph.D., who helped create Figure 1 andwho has dealt with all the DUA issues over the past many years, to LindaLaliberte-Cote for her confidence in this concept and in continuing to supportthis effort throughout the years, and to Christian Brostrup-Jensen for his pro-gramming support and his comments and feedback in testing the residentialhistory file program. A previous version of this paper was presented at theannual meeting of the Gerontological Society of America in 2003.

Disclosures: None.Disclaimers: None.

REFERENCES

Brown, M. L., G. F. Riley, A. L. Potosky, and R. D. Etzioni. 1999. ‘‘Obtaining Long-Term Disease Specific Costs of Care: Application to Medicare Enrollees Diag-nosed with Colorectal Cancer.’’ Medical Care 37 (12): 1249–59.

Burton, L. C., D. M. Steinwachs, P. S. German, S. Shapiro, L. J. Brant, T. M. Richards,and R. D. Clark. 1995. ‘‘Preventive Services for the Elderly: Would CoverageAffect Utilization and Costs under Medicare?’’ American Journal of Public Health85 (3): 387–91.

Coburn, A. F., R. G. Keith, and E. J. Bolda. 2002. ‘‘The Impact of Rural Residence onMultiple Hospitalizations in Nursing Facility Residents.’’ Gerontologist 42 (5):661–6.

Coleman, E. A., S. J. Min, A. Chomiak, and A. M. Kramer. 2004. ‘‘Posthospital CareTransitions: Patterns, Complications, and Risk Identification.’’ Health ServiceResearch 39 (5): 1449–65.


Cooper, G. S., Z. Yuan, A. Chak, and A. A. Rimm. 2000. ‘‘Patterns of EndoscopicFollow-Up after Surgery for Nonmetastatic Colorectal Cancer.’’ GastrointestinalEndoscopy 52 (1): 33–8.

Feng, Z., P. R. Katz, O. Intrator, J. Karuza, and V. Mor. 2005. ‘‘Physician and NurseStaffing in Nursing Homes: The Role and Limitations of the Online SurveyCertification and Reporting (OSCAR) System.’’ Journal of American Medical Di-rector Association 6 (1): 27–33.

Fries, B. E., D. P. Schneider, W. J. Foley, M. Gavazzi, R. Burke, and E. Cornelius. 1994.‘‘Refining a Case-Mix Measure for Nursing Homes: Resource UtilizationGroups (RUG-III).’’ Medical Care 32 (7): 668–85.

Gambassi, G., F. Landi, L. Peng, C. Brostrup-Jensen, K. Calore, J. Hiris, L. Lipsitz,V. Mor, and R. Bernabei. 1998. ‘‘Validity of Diagnostic and Drug Data in Stan-dardized Nursing Home Resident Assessments: Potential for Geriatric Pharma-coepidemiology. SAGE Study Group. Systematic Assessment of Geriatric DrugUse via Epidemiology.’’ Medical Care 36 (2): 167–79.

Gornick, M., A. McMillan, and J. Lubitz. 1993. ‘‘A Longitudinal Perspective on Pat-terns of Medicare Payments.’’ Health Affairs (Millwood) 12 (2): 140–50.

Grabowski, D. C., J. J. Angelelli, and V. Mor. 2004. ‘‘Medicaid Payment and Risk-Adjusted Nursing Home Quality Measures.’’ Health Affairs (Millwood) 23 (5):243–52.

Hawes, C., J. N. Morris, C. D. Phillips, V. Mor, B. E. Fries, and S. Nonemaker. 1995.‘‘Reliability Estimates for the Minimum Data Set for Nursing Home ResidentAssessment and Care Screening (MDS).’’ Gerontologist 35 (2): 172–8.

Hirdes, J. P., D. H. Frijters, and G. F. Teare. 2003. ‘‘The MDS-CHESS Scale: A NewMeasure to Predict Mortality in Institutionalized Older People.’’ Journal of Amer-ican Geriatrics Society 51 (1): 96–100.

Intrator, O., K. Berg, V. Hiris, V. Mor, and S. T. G. Miller. 2003. ‘‘Development andValidation of the Medicare-MDS Residential History File.’’ Gerontologist 43: 30–1.

Intrator, O., D. C. Grabowski, J. Zinn, M. Schleinitz, Z. Feng, S. Miller, and V. Mor.2007. ‘‘Hospitalization of Nursing Home Residents: The Effects of States’ Med-icaid Payment and Bed-Hold Policies.’’ Health Services Research 42 (4): 1651–71.

Iwashyna, T. J. 2003. ‘‘On the Detection of Nursing Home Use in Medicare Claims.’’Health Services and Outcomes Research Methodology 4: 187–96.

Katz, P. R., J. Karuza, O. Intrator, and V. Mor. 2009. ‘‘Nursing Home PhysicianSpecialists: A Response to the Workforce Crisis in Long-Term Care.’’ Annals ofInternal Medicine 150 (6): 411–3.

Levinsky, N. G., W. Yu, A. Ash, M. Moskowitz, G. Gazelle, O. Saynina, and E. J.Emanuel. 2001. ‘‘Influence of Age on Medicare Expenditures and Medical Carein the Last Year of Life.’’ Journal of American Medical Association 286 (11): 1349–55.

MEPS. 2006. ‘‘Household Expenditures’’ [accessed on October 4, 2010]. Available athttp://meps.ahrq.gov/mepsweb/data_stats/tables_compendia_hh_interactive.jsp?_SERVICE=MEPSSocket0&_PROGRAM=MEPSPGM.TC.SAS&File=HCFY2006&Table=HCFY2006%5FPLEXP%5F%40&VAR1=AGE&VAR2=SEX&VAR3=RACETH5C&VAR4=INSURCOV&VAR5=POVCAT06&VAR6=MSA&VAR7=REGION&VAR8=HEALTH&VARO1=4+17+44+64&VA


http://meps.ahrq.gov/mepsweb/data_stats/tables_compendia_hh_interactive.jsp?_SERVICE=MEPSSocket0&_PROGRAM=MEPSPGM.TC.SAS&File=HCFY2006&Table=HCFY2006%5FPLEXP%5F%40&VAR1=AGE&VARR2=SEX&VAR3=RACETH5C&VAR4=INSURCOV&VAR5=POVCAT06&VAR6=MSA&VAR7=REGION&VAR8=HEALTH&VARO1=4+17+44+64&VARO2=1&VARO3=1&VARO4=1&VARO5=1&VARO6=1&VARO7=1&VARO8=1&_Debug





RO2=1&VARO3=1&VARO4=1&VARO5=1&VARO6=1&VARO7=1&VARO8=1&_Debug=

Miller, S. C., O. Intrator, L. Laliberte, and Y. Cang. 2004. ‘‘End-of-Life Care forMedicare Skilled Nursing Facility (SNF) Residents: Patterns of Care andExpenditures by Hospice Enrollment Status.’’ Journal of the American GeriatricsSociety 52 (4): S193–S93.

Mor, V., J. Angelelli, R. Jones, J. Roy, T. Moore, and J. Morris. 2003. ‘‘Inter-RaterReliability of Nursing Home Quality Indicators in the U.S.’’ BMC Health ServiceResearch 3 (1): 20.

Mor, V., K. Branco, J. Fleishman, C. Hawes, C. Phillips, J. Morris, and B. Fries. 1995.‘‘The Structure of Social Engagement among Nursing Home Residents.’’ Journalof Gerontology B Psycholical Science Society Science 50 (1): P1–8.

Mor, V., D. Grabowski, O. Intrator, and Z. Feng. 2010. ‘‘The Revolving Door ofRehospitalizations from Skilled Nursing Facilities’’ Health Affairs 29 (1): 57–64.

Mor, V., J. Zinn, J. Angelelli, J. M. Teno, and S. C. Miller. 2004. ‘‘Driven to Tiers:Socioeconomic and Racial Disparities in the Quality of Nursing Home Care.’’Milbank Quarterly 82 (2): 227–56.

OIG. 2003. ‘‘Quality Assurance Committees in Nursing Homes.’’ OEI-01-01-00090[accessed on October 4, 2010]. Available at http://www.oig.hhs.gov/oei/reports/oei-01-01-00090.pdf

Phillips, C. D., and J. N. Morris. 1997. ‘‘The Potential for Using Administrative andClinical Data to Analyze Outcomes for the Cognitively Impaired: An Assess-ment of the Minimum Data Set for Nursing Homes.’’ Alzheimer Disease AssociationDisorders 11 (Suppl 6): 162–7.

Roos, N. P., P. Montgomery, and L. L. Roos. 1987. ‘‘Health Care Utilization in theYears Prior to Death.’’ Milbank Quarterly 65 (2): 231–54.

Sood, N., M. B. Buntin, and J. J. Escarce. 2008. ‘‘Does How Much You Pay Matter?Evidence from the Inpatient Rehabilitation Care Prospective Payment System.’’Journal of Health Economics 27 (4): 1046–59.

Wu, N., S. C. Miller, K. Lapane, J. Roy, and V. Mor. 2005. ‘‘The Quality of the QualityIndicator of Pain Derived from the Minimum Data Set.’’ Health Service Research40 (4): 1197–216.

SUPPORTING INFORMATION

Additional supporting information may be found in the online version of thisarticle:

Appendix SA1: Author Matrix.

Please note: Wiley-Blackwell is not responsible for the content or func-tionality of any supporting materials supplied by the authors. Any queries(other than missing material) should be directed to the corresponding authorfor the article.



=

http://www.oig.hhs.gov/oei/reports/oei-01-01-00090.pdf

http://www.oig.hhs.gov/oei/reports/oei-01-01-00090.pdf

The Residential History File: Studying Nursing Home Residents' Long-Term Care Histories*

Documents