Directory UMM :Journals:Journal of Health Economics:Vol19.Issue6.Dec2001:

www.elsevier.nlrlocatereconbase

Costs and outcomes associated with alternative

discharge strategies following joint replacement

surgery: analysis of an observational study using

a propensity score

Peter C. Coyte

a,b,c,d,e,f,)

_{, Wendy Young}

a,c

_{, Ruth Croxford}

a,c

a

Institute for Clinical EÕaluatiÕe Sciences in Ontario, Toronto, ON, Canada

b

Department of Health Administration, UniÕersity of Toronto, Toronto, ON, Canada

c

Home Care EÕaluation and Research Center, UniÕersity of Toronto, Toronto, ON, Canada

d

Hospital Management Research Unit, UniÕersity of Toronto, Toronto, ON, Canada

e

Institute for Policy Analysis, UniÕersity of Toronto, Toronto, ON, Canada

f

Arthritis Community Research and EÕaluation Unit, Wellesley Hospital Research Institute,

Toronto, ON, Canada

Received 24 June 1998; received in revised form 25 March 1999; accepted 3 December 1999

Abstract

We estimated the impact of alternative discharge strategies, following joint replacement

ŽJR surgery, on acute care readmission rates and the total cost of a continuum of care..

Following surgery, patients were discharged to one of four destinations. Propensity scores were used to adjust costs and outcomes for potential bias in the assignment of discharge destinations.

We demonstrated that the use of rehabilitation hospitals may lower readmission rates, but at a prohibitive incremental cost of each saved readmission, that patients discharged with home care had longer acute care stays than other patients, that the provision of home care services increased health system costs, and that acute care readmission rates were

)_{Corresponding author. Department of Health Administration, McMunich Building, 12 Queen’s}

Park Crescent West, University of Toronto, Toronto, Ontario, Canada, M5S 1A8. Tel.:q 1-416-978-8369; fax:q1-416-978-7350.

Ž .

E-mail address: coyte@chass.utoronto.ca P.C. Coyte .

0167-6296r00r$ - see front matterq_{2000 Elsevier Science B.V. All rights reserved.}

Ž .

greatest among patients discharged with home care. Our study should be seen as one important stepping stone towards a full economic evaluation of the continuum of care for patients.q2000 Elsevier Science B.V. All rights reserved.

JEL classification: I10; C10

Keywords: Costing episodes-of-care; Joint replacements; Home care; Rehabilitation; Readmission rates

1. Introduction

Musculoskeletal disorders affect a significant proportion of the population ŽStatistics Canada, 1993 , their prevalence increases with age Badley et al., 1994. Ž .

Ž

and they are a leading cause of disability McAlindon et al., 1992; Verbrugge, .

1991 Once arthritis affects the joints, patients suffer pain, they often lose function Ž

and their quality of life is adversely affected Hadler, 1985; Kramert et al., 1983; Liang et al., 1984; McAlindon et al., 1992; Moskowitz et al., 1992; Peyron and

.

Altman, 1992; Pincus et al., 1989; Verbrugge, 1991 . In individuals for whom Ž .

medical therapy has failed, joint replacement JR surgery has been shown to be a cost-effective intervention resulting in pain relief, enhancing function and

improv-Ž

ing the quality of life Harris and Sledge, 1990; Jonsson and Larsson, 1991; . Laupacis et al., 1993; Liang et al., 1986; Rorabeck et al., 1994; Scott et al., 1988 .

In the last 5 years, the number of JRs performed in Ontario has increased at an average annual rate of 9.0%, with hip replacement surgery increasing at a slower

Ž .

annual rate, 5.0%, than knee replacement surgery, 14.6% Coyte et al., 1996a,b . While JRs are cost-effective from the viewpoint of patients and the health system

ŽLaupacis et al., 1993; Liang et al., 1984, 1986; Rorabeck et al., 1994 , they.

represent a significant expense to hospitals. Thus, as hospitals strive to contain costs, limits may be placed on access to JRs by budget restrictions for prosthetic devices, reductions in operating room time and by bed closures.

Besides restrictions to the availability of JRs, health system restructuring has significantly reduced the number of days of inpatient care for JR patients, and, simultaneously, has altered discharge patterns. Indeed, the propensity to imple-ment early discharge programs for JR patients and to discharge these patients to

Ž .

either a rehabilitation hospital or to home with or without home care services has increased. However, little is known about the effect of these modified patterns of practice on health system costs, health outcomes and the extent to which the burden of care has shifted from acute care institutions to patients, families and

Ž .

community agencies Coyte and Young, 1997a,b; Parr, 1996; Welch et al., 1996 . Ž While it is clear that expenditures on community-based care have increased Parr,

.

1996; Welch et al., 1996 , this increase has occurred without compelling evidence Ž

of service cost-effectiveness FederalrProvincialrTerritorial Working Group on

Home Care, 1990; Hollander, 1994; Parr, 1996; Price Waterhouse, 1989; Weissert, .

perception that unless these services are targeted, they will not represent a Ž

cost-effective alternative to acute care Weissert, 1985, 1991; Weissert et al., .

1980, 1988; Welch et al., 1996 .

Ž

This paper employs an episode-of-care approach Gonnella et al., 1984; Horn-.

brook, 1995; Hornbrook et al., 1985 to estimate the impact of alternative

discharge strategies following JR surgery on the total cost of a continuum of care and on acute care readmission rates after adjusting for potential assignment bias. The paper contributes to the literature by developing a continuum of care methodological framework for cost–outcome studies that explicitly adjusts for

Ž

potential assignment bias by using a propensity score Rosenbaum and Rubin, .

1983, 1984 . The potential to control for possible selection biases in a non-rando-mized comparison of the costs and outcomes of clinical practice expands the scope of health services research.

2. Methods

All separations from Ontario hospitals between fiscal years 1991 and 1994

Ž_{hereafter, FY91 and FY94 for patients receiving JR surgery, comprising knee}.

and hip replacements, were acquired from the Canadian Institute for Health

Ž .

Information CIHI . These separations exclude residents of Ontario who had insured JRs performed outside Ontario. These non-Ontario JRs comprised less

Ž .

than 0.5% of all procedures performed on Ontario residents Coyte et al., 1994 . JRs were identified as Canadian Classification of Diagnostic, Therapeutic and

Ž . Ž . Ž

Surgical Procedures CCP Statistics Canada, 1994 code 93.41 Total Knee

Ž . Ž .. Ž

Replacement geomedic polycentric , 93.51 Total Hip Replacement with use of

. Ž .

methyl methacrylate or 93.59 Other Total Hip Replacement , in any procedure field. Our data extraction algorithm identified and selected 48,826 JR hospitaliza-tions over the study period.

2.1. Data cleaning

Various case eligibility criteria were applied sequentially to all identified JR hospitalizations to better define the study population. Hospitalizations were

ex-Ž .

cluded if the patient: was a non-resident of Ontario 471 JR hospitalizations ; had

Ž .

an invalid residence code 23 JR hospitalizations ; had surgery performed in a

Ž . Ž

non-acute facility six JR hospitalizations ; was less than 20 years of age 62 JR

. Ž

hospitalizations ; if the patient had a missing residence postal code 599 JR

. Ž

hospitalizations ; or if the patient had a missing unique health care number 459

. Ž .

JR hospitalizations . These exclusions amounted to 3.3% or 1620 of the original set of 48,826 JR hospitalizations.

The resulting and cleaned JR hospitalization file, which comprised 47,206 JR hospitalizations, was converted to a unique patient-level file by taking only the

Ž . Ž first or index occurrence of a JR hospitalization during the study period 41,257

.

unique JR patients . Moreover, to improve the homogeneity of the JR patient population, four further criteria were applied. First, as JR patients present with an array of different diagnoses, we developed a mutually exclusive and exhaustive

Ž . 1

diagnosis hierarchy and included only osteoarthritic OA patients thereby exclud-ing 8947 JR patients. Second, to reduce the extent of clinical variation among JR

Ž

patients, we only included typical JR patients Hospital Medical Records Institute, . Ž

1993 such patients experience a complete course of treatment in contrast to

.

atypicals who suffered an inpatient death, discharged themselves, etc. who were Ž

assigned one of the following four Case Mix Group categories Hospital Medical .

Records Institute, 1993 : CMG350 — Multiple or Bilateral Joint Replacement;

CMG352 — Hip Replacement with Co-existing Condition andror Complication;

CMG353 — Hip Replacement with No Co-existing Condition andror

Complica-tion; and CMG354 — Knee Replacement. This criterion resulted in the exclusion of 2225 JR patients. Third, we excluded JR patients if they were discharged from their index JR hospitalization to either another acute care facility, which may

Ž

signal medical instability at the time of discharge Rehabilitation Planning Com-.

mittee, 1995 , or where their discharge destination was not a rehabilitation hospital, home care or home with self-care, thereby excluding 910 JR patients. Finally, we excluded patients if they were discharged to home within 3 days of

Ž . Ž .

surgery 44 JR patients . These exclusions amounted to 29.4% or 12,126 of the original set of unique and heterogeneous JR patients.

Application of the exclusion criteria resulted in a patient-level file comprising 29,131 typical JR patients with osteoarthritis. Each patient was categorized according to their discharge to one of four destinations: a rehabilitation hospital,

Ž .

and subsequently, discharged to home without home care RS ; a rehabilitation

Ž .

hospital, and subsequently, discharged to home with home care RH ; home with

Ž . Ž .

home care HC ; and home with self-care or informal-care SC . Receipt of home

Ž .

care within 30 days of final discharge Kenney, 1993a was determined from the

Ž .

Ontario Home Care Administrative System OHCAS database, which records all utilization of home care services by residents of Ontario. Admission to a rehabili-tation hospital was determined from the CIHI database, which records all admis-sions to hospitals, including rehabilitation hospitals, in Ontario. The purpose of our study was to assess acute care readmission rates and variations in the total cost

1 _{Ž .}

Patients were classified as fracture patients if there was an ICD-9 Kenney, 1991 diagnostic code

) _{Ž .}

of 820. Fracture of neck of femur in any diagnostic field. Rheumatoid arthritis patients were so

) _Ž

classified if there was a diagnostic code of 714. Rheumatoid arthritis and other inflammatory

.

polyathropathies in any diagnostic field. Traumatic arthritis patients were so categorized if there was a

Ž .

diagnostic code of 716.1 Traumatic arthropathy , while osteoarthritic patients were those remaining

) _{Ž .}

patients with a diagnostic code of 715. Osteoarthrosis and allied disorders . All other patients were assigned to a miscellaneous category.

of a continuum of care for JR patients assigned to each discharge destination after adjusting for potential assignment bias.

2.2. Determination of readmission rates

For each discharge destination, acute care readmission rates were computed based on criteria developed by the US Health Care Financing Administration

ŽHCFA. ŽRiley et al., 1993 . The diagnoses and procedure codes used by HCFA.

were determined by expert panels that identified three types of adverse events leading to readmissions: complications related to a particular procedure; failure of the procedure to achieve its therapeutic goals; and untoward events associated with the history of the disease. An array of time windows from the index event were specified for each readmission, with a 1-year window being the longest.

2.3. Cost of a continuum of care

A partial health system perspective was taken to assess the four components to the total cost of a continuum of JR care. First, acute care case costs associated with the index JR hospitalization were estimated using audited patient-specific

Ž . Ž

cost data acquired from the Ontario Case Cost Project OCCP for FY94 Joint . Planning and Policy Committee, 1995; Ontario Case Costs Subcommittee, 1991 .

Ž

These data were obtained from hospitals that were MIS compliant The MIS

. Ž .

Group, 1985 . Since index case cost estimates were available for 18.7% or 1466 of all JR hospitalizations in FY94, multiple regression analysis was used to assess the relationship between these JR case costs and various independent variables. Once this estimated relationship was derived, it was applied to all index JR hospitalizations to yield acute care case cost estimates. While several factors may

Ž

account for variations in case costs Anderson et al., 1989; Cromwell et al., 1987; Feigenbaum et al., 1992; Hendricks and Cromwell, 1989; Lave and Leinhardt,

.

1976 , these factors may be collapsed into four broad classes of hypotheses: Ž

patient factors comprising demographic and socio-economic characteristics of

. Ž .

patients ; clinical factors comprising procedure type and comorbidity ; provider Ž

factors comprising hospital characteristics, such as teaching status and the annual

. Ž

JR volume at the index hospital ; and spatial and temporal factors measuring .

access to JRs, its timing and area characteristics . The relationship between these factors and acute care case costs were derived. Age, gender, length of stay, whether a revision was performed, CMG, comorbidities, hospital teaching status

and annual JR volume, and urbanrrural classification of patient’s residence, were

used in the final regression model.

To assess the comorbidity of patients during their index event, a medical

Ž .

record-based comorbidity index, the Charlson index Charlson et al., 1987 , was Ž

adapted for use with ICD-9 diagnosis and CCP procedure codes Deyo et al., .

applied to medical record data to predict the relative risk of 1-year mortality, it has Ž

been used to predict other outcomes in claims data Deyo et al., 1992; Melfi et al., .

1995; Moskowitz et al., 1992; Romano et al., 1993a,b .

Hospitals were classified as: teaching, that is, were members of the Ontario

Ž .

Council of Teaching Hospitals OCOTH ; or non-teaching hospitals.

Since JRs may be either primary or revision procedures, an algorithm was

Ž .

developed to identify revision JRs Heck et al., 1998 . Revisions were identified through the joint occurrence of one of the previously specified CCP codes and any

Ž .

one of the following ICD-9 World Health Organization, 1977 diagnostic codes in Ž

any diagnostic field: 996.4, 996.6, or 996.7 complications peculiar to certain .

specified procedures . All other JRs were classified as primary JRs.

Second, for patients discharged to a rehabilitation hospital, such costs were estimated as the product between the number of days of care and an estimated per

Ž .

diem cost Rehabilitation Planning Committee, 1995 . Such case cost estimates

Ž .

included direct treatment costs and fully allocated indirect or overhead costs. Third, all home care service encounters between the first home care service date and the home care discharge date, censored at 90 days from the first service encounter, were assessed for patients who received services within 30 days of their final inpatient discharge date. Utilization initiated beyond 30 days or extending beyond 90 days from the first service date is unlikely to be related to the index

Ž . Ž

hospitalization Hollander, 1994; Kenney, 1993a . The direct and indirect or .

overhead cost of home care and home-making services was derived as the product Ž between the number of services rendered by various health professionals

physio-. Ž

therapists, nurses, etc. and the fully allocated cost of these services Drummond et

. Ž .

al., 1987 , originally estimated by Hollander Hollander, 1994 and updated by the Metropolitan Toronto Home Care Program. Equipment rental costs were estimated separately using a standard cost for each month the patient was registered with the home care program to a maximum of 3 months.

Fourth, for patients discharged from their index hospitalization, but subse-quently readmitted to an acute care institution, the cost of that acute care readmission episode was estimated. Readmission hospitalizations were identified

Ž .

according to criteria developed by HCFA Riley et al., 1993 . The resource

Ž .

intensity weight methodology Pink and Bolley, 1994a,b was then applied: each Ž

hospitalization was assigned to an individual CMG Hospital Medical Records .

Institute, 1993 based on the most responsible diagnosis, principal procedure, age group and presence of comorbidity or complications. For each CMG, case costs were estimated by combining the resource intensity weight appropriate to the CMG with hospital-specific costs per weighted case. In this way, case costs for each acute care readmission were estimated.

The four components to the cost of a continuum of JR care were weighted by Ž

utilization rates and adjusted for growth in the consumer price index Statistics .

Canada, 1996 to yield total resource costs by discharge destination in FY94 Canadian dollars.

2.4. Adjustment for assignment bias

The randomized controlled trial is the ‘gold standard’ for comparing treatment outcomes. Properly performed, randomization should produce treatment groups, which are comparable with respect to both observed and unobserved covariates, allowing differences in outcomes to be attributed to treatment differences. How-ever, it is not feasible or ethical to study many interesting questions in health care using random treatment assignment. Furthermore, there is concern about the generalizability of the results of randomized trials, due to their narrow inclusion criteria, strict treatment protocols, the fact that they are typically carried out in larger centers, and the potential for bias in subjects who elect to participate. The present study is an example of one that addresses an important question, but would

Ž .

be neither ethical nor feasible both in terms of cost and time required to carry out as a randomized trial.

When treatment assignments are not randomized, outcomes may be biased by differences in the characteristics which influence a physician to select a particular treatment for the patient. One method of adjusting an analysis of treatment outcomes for the effects of confounding covariates is to perform a post hoc stratification of the patients. The determination of treatment outcomes is

per-Ž .

formed within each subgroup or stratum of similar patients. The results are then combined to obtain an estimate of the treatment effect for the entire population.

Ž .

The propensity score Rosenbaum and Rubin, 1983, 1984 is a means of defining strata of comparable patients.

The propensity score, which is the probability that a patient will be assigned to a specific treatment group, conditional on the values of the observed covariates, is typically calculated using logistic regression, although other models may be used. The covariates used in our logistic regression were patient age, gender, comorbid

conditions, CMG, diagnosis, length of stay on the index admission, urbanrrural

characteristic of the patient residence, whether a revision JR was performed, teaching status of the hospital and year. In addition to these main effects, two-way interaction terms were included in the logistic regression model.

When there are more than two treatments being compared, a separate propen-Ž

sity score is calculated for each pair of treatments e.g. HC and RH, HC and RS,

. Ž .

HC and SC, etc. Rubin, 1997 . This is an important strength of the propensity

score methodology compared to other approaches, since it allows different propen-sity score models in order to adjust appropriately for different comparisons. The propensity score model for the comparison of one pair of treatments generally needs to be different from that for a comparison of a different pair of treatments. The propensity score summarizes the information in all available covariates as a single number: the patient’s propensity to be discharged to a given destination. Subclassifying patients on the basis of the propensity score produced strata of patients in which both treatment groups were balanced with respect to all of the observed covariates; the effects of unobserved covariates are also controlled for, to

the extent that they are correlated with the observed ones. Five strata suffice to Ž

remove over 90% of the bias for each of the variables Rosenbaum and Rubin, .

1984 . Patients were, therefore, assigned to strata based on the quintiles of the propensity score. Univariate analyses of the covariates were performed to ensure that within each stratum, there was no appreciable difference between the patient groups, by discharge strategy. Tests of significant differences for continuous data were based on t-tests and analysis of variance. Fisher’s exact tests and chi-square tests were used for binary and nominal variables. Statistical analyses were

Ž

performed on a Sun Sparc Server 1000 using SAS release 6.11 SAS Institute, .

Cary, NC .

The treatment cost and patient outcome due to each discharge strategy where calculated within each stratum. To estimate the treatment costs and outcomes for the entire patient population, weighted sums of the stratum-specific results were

Ž .

calculated, using standard methods for stratified sampling Jessen, 1978 . The end-product of the propensity score analysis was an estimate, for each discharge strategy, of the average cost and the readmission rate had all patients in the province received that treatment following discharge. Standard deviations of the estimates were also obtained, allowing the calculation of P-values in order to test for significant differences between the discharge destinations. To adjust for multiple comparisons of discharge destinations, individual pairwise comparisons

Ž .

considered P-values -0.0083 (0.05r6 to be statistically significant. This

adjustment yields an overall significance level of 0.05.

3. Results

Between FY91 and FY94, the number of JR hospitalizations in Ontario increased at an average annual rate of 5.5%, which was less than half the annual

Ž

growth rate of 11.5% recorded between FY81 and FY91 Cohen et al., 1994a,b; .

Coyte et al., 1996b; Naylor and De Boer, 1996 . Moreover, after adjusting for population growth, the annual rate of growth in JR hospitalization rates was 3.8%

Ž .

between FY91 and FY94 compared to 9.8% between FY81 and FY91 Fig. 1 .

3.1. Discharge strategies

Study exclusion criteria resulted in a unique patient-level file comprised of 29,131 JR patients who were discharged to one of four destinations between FY91 and FY94: a rehabilitation hospital, and subsequently, to home without home care ŽRS ; a rehabilitation hospital, and subsequently, to home with home care RH ;. Ž .

Ž . Ž .

home with home care HC ; and home with self-care or informal-care SC . Despite deceleration in the growth of JR rates since FY91, there have been

Ž .

Fig. 1. JR surgery rates in Ontario, FY1981–FY1994.

patients fell from 40.8% in FY91 to 28.8% in FY94, while the percentage of HC patients increased from 34.7% to 42.8%. Since an additional 7.3% of all JR patients received home care services following discharge from a rehabilitation hospital in FY94, more than half of all JR patients received home care services in

Ž .

FY94 following their episode s of inpatient care.

While home care utilization by JR patients increased at an average annual rate of 10.8% and 12.5% between FY91 and FY94 for HC and RH patients, respec-tively, there was also an increase in the utilization of rehabilitation hospitals. Specifically, the average annual rate of growth in the use of rehabilitation hospitals

Table 1

Number of JR patients by discharge destination, FY1991–FY1994

RSsRehabilitation and Self-Care; RHsRehabilitation and Home Care; SCsSelf-Care; HCsHome Care.

Fiscal year Discharge destination

RS RH SC HC All

FY91 1337 405 2902 2473 7117

FY92 1429 551 2250 2799 7029

FY93 1501 503 2177 2955 7136

FY94 1646 576 2264 3363 7849

Ž . Ž . Ž . Ž . Ž .

was 7.2% and 12.5% for RS and RH patients, respectively. Thus, the average annual rate of change in the use of home care services, rehabilitation hospitals, and

self-care since FY91 were 11.0%, 8.5%, andy7.9%, respectively. These results

demonstrate a shift towards greater reliance on home care services and inpatient rehabilitation, and less reliance on self-care following an acute care JR hospitaliza-tion.

3.2. Characteristics of JR patients

Table 2 reports JR patient characteristics by discharge destination. RH patients

Ž .

were significantly P-0.0001 more likely to be female, older, exhibit a

rela-tively high comorbidity index and be discharged from a non-teaching hospital than other patients. These same characteristics distinguish patients discharged with

Ž . Ž

home care services HC and RH from those discharged without such services SC .

and RS . Moreover, SC patients form the antithesis to RH patients; they were

Ž .

significantly P-0.0001 more likely to be younger male patients, with relatively

few comorbid conditions, who were discharged from a teaching hospital. There

Ž .

was a significant P-0.0001 trend towards JR surgery on younger male patients,

with more comorbid conditions, who were discharged from non-teaching hospitals, with this trend common to all discharge destinations.

3.3. Index acute care hospitalizations

Between FY91 and FY94, there was a dramatic decline in the duration of

Ž .

inpatient acute care for JR patients Table 3 . Specifically, the average length of stay fell by 21.0% from 12.4 days in FY91 to 9.8 days in FY94. This decrease, which was similar for all discharge strategies, represents an average annual decrease in the duration of inpatient care of 7.5%. Patients discharged to home

Ž . Ž .

with home care HC experienced a significantly P-0.0001 longer average

Table 2

Characteristics of JR patients by discharge destination, FY1991–FY1994

RSsRehabilitation and Self-Care; RHsRehabilitation and Home Care; SCsSelf-Care; HCsHome Care.

Patient characteristics Discharge destination

RS RH SC HC All Chi-square

P-0.0001 63.2 72.9 45.7 62.3 57.7 P-0.0001

Ž .

Age means 69.3 73.7 67.1 70.7 69.4 P-0.0001

Ž .

Charlson)0 % 12.4 17.0 11.9 14.4 13.3 P-0.0001

Ž .

Revision JR % 2.3 3.0 2.2 2.2 2.3 Ps0.134

Ž .

Teaching hospital % 31.9 29.9 40.0 34.8 35.6 P-0.0001

Ž .

Table 3

Acute care length of stay for JR patients by discharge destination, FY1991–FY1994

RSsRehabilitation and Self-Care; RHsRehabilitation and Home Care; SCsSelf-Care; HCsHome Care.

Fiscal year Discharge destination

RS RH SC HC All

FY91 12.2 10.3 12.2 12.9 12.4

FY92 10.9 11.4 11.3 12.2 11.6

FY93 10.3 10.9 10.2 11.3 10.7

FY94 9.7 9.5 9.3 10.3 9.8

length of stay than other JR patients even after controlling for potentially

confounding variables, comprising patient age, gender, urbanrrural residence,

comorbidity, case mix group, type of procedure performed and hospital teaching status. After controlling for potential confounders, RH patients experienced a

Ž .

significantly P-0.0001 shorter average length of stay than other JR patients.

3.4. Costs and outcomes after adjusting for assignment bias

Ž .

Fig. 2 reports the total cost in FY94 Canadian dollars of a continuum of JR care and acute care readmission rates by discharge destination over the study period after controlling for various confounding variables. Standardized total costs,

Ž

comprising the cost of acute care hospitalizations both index and subsequent

.

readmissions , the cost of rehabilitation hospitalizations and the cost of home care services, varied from a low of $8166 for SC patients to a high of $13,569 for RH patients.

Ž .

Patients discharged to rehabilitation hospitals RH and RS patients incurred

Ž .

health system costs that were almost $5000 or between 50% and 60% greater than either HC or SC patients. However, acute care readmission rates were 20.0% lower for RH patients than for HC patients, 266.5 per 10,000 JRs vs. 333.3, and 25.5% lower for RS patients than for SC patients, 201.9 per 10,000 JRs vs. 271.1. Thus, increased continuum of care costs associated with the use of rehabilitation hospitals were associated with better health outcomes as measured through lower acute care readmission rates.

Patients discharged to home with home care either directly from an acute care

Ž . Ž .

hospital HC or after a rehabilitation hospitalization RH reported health system

Ž .

costs that were between $1000 and $400 or between 12.1% and 3.4% greater than either SC or RS patients. Acute care readmission rates were, 18.7% lower for SC patients than for HC patients, 271.1 per 10,000 JRs vs. 333.3, and were 24.2% lower for RS patients than for RH patients, 201.9 per 10,000 JRs vs. 266.5. Thus,

Ž

despite increased expenditures, adverse health outcomes measured by acute care .

readmission rates were greater for patients discharged with home care than for patients without home care. While readmission rates were higher for some patients, the distribution of the reasons for readmission were similar for all JR patients irrespective of their discharge destination.

Average health system costs fell at an average annual rate of 2.7% over the

Ž .

study period. This annual reduction in costs was greatest for RS patients 4.2%

Ž . Ž .

followed by SC patients 3.7% , HC patients 3.5% , and finally, RH patients Ž1.9% . The main reason for these reductions stem from the shorter duration of the. index acute and rehabilitation hospitalizations.

Fig. 3 depicts the distribution of costs for JR patients by discharge strategy. The share of index acute care costs ranged from a low of 56.1% for RH patients to a high of 97.9% for SC patients. However, there was very little variation in the average cost of the index acute care hospitalization, which ranged from a low of $7616 for RH patients to a high of $8062 for RS patients. This lack of variation in the cost of the index acute care hospitalization was also reported for rehabilitation hospital costs and home care costs. Rehabilitation costs were $4756 for RH patients and $4881 for RS patients, while the cost of home care services were $908 for RH patients and $896 for HC patients. While acute care readmission costs for all discharge strategies represented approximately 2.0% of total costs, these costs represented a larger share of such costs for JR patients discharged with

Ž . Ž

home care services HC and RH patients than patients without such services SC .

and RS patients . Our results suggest that variations in costs were primarily attributable to variations in the propensity to utilize different packages of rehabili-tation care, rather than in variations in the cost of the services provided by these rehabilitation programs.

Fig. 3. Components of costs for JR patients by discharge strategy, FY1991–FY1994.

3.5. Costs and outcomes

Propensity scores were used to control for potential bias in the assignment of discharge destinations. Although patients who were assigned to different discharge destinations had quite different characteristics, there were no appreciable differ-ences within each stratum. Fig. 4 reports the total cost of a continuum of care and

Ž .

acute care readmission rates. There were significant P-0.0001 differences in

costs among the four discharge destinations. Costs ranged from $14,081 for RH patients to $8220 for SC patients. The acute care readmission rate per 10,000 JRs for the relatively small number of RH patients was 373.6 and was not significantly different than those for other discharge strategies. Significant differences in acute care readmission rates were recorded for comparisons between RS and HC

Ž . Ž .

patients Ps0.002 , and RS and SC patients P-0.0001 . In ascending order,

these readmission rates per 10,000 JRs were 203.6 for RS patients, 284.4 for SC patients, 344.4 for HC patients.

Incremental cost–outcome analysis was performed to measure the potential increase in continuum of care costs in order to lower acute care readmissions. Since RS patients recorded the lowest stratum-adjusted acute care readmission rates, all comparisons included the RS discharge strategy. Moreover, as RH

Ž .

patients were associated with significantly P-0.0001 higher costs and greater

Ž .

Fig. 4. Costs and outcomes of alternative discharge strategies for JR patients, FY1991–FY1994.

assessed. Switching JR patients from a HC discharge strategy to a RS strategy results in increased continuum of care costs of $4058 per patient, a reduction in readmissions of 140.8 per 10,000 JRs, and therefore, yields an incremental cost of each saved readmission of $288,210. Furthermore, switching from a SC discharge strategy to a RS strategy results in increased continuum of care costs of $4942 per patient and a reduction in readmissions of 80.8 per 10,000 JRs. The incremental cost of each saved readmission associated with this switch in practice was $611,634.

4. Discussion

Ž

An episode-of-care approach Gonnella et al., 1984; Hornbrook, 1995; Horn-.

brook et al., 1985 was used to estimate the impact of discharge strategies on acute care readmission rates and the total cost of a continuum of care for JR patients, after controlling for potential assignment bias. The results address a common informational concern for policy makers pertaining to the costs and outcomes from alternative practices.

Because discharge assignments were not randomized in this study, any compar-ison of costs and outcomes must adjust for potential selection bias. Even within our relatively homogeneous JR patients, there were significant differences among patients assigned to each of the four distinct discharge destinations, with those

discharged to self-care tending to be younger, with few comorbid conditions than those receiving additional care after discharge. By employing the propensity score to account for those variables that affect the assignment of patients to each discharge strategy, we were able to operationalize our adjustment for potential assignment bias.

The propensity score methodology was used to adjust for the effects of confounding factors on treatment costs and outcomes in the absence of

random-Ž .

ized assignment. As this study demonstrates, the outcome s of interest may be

Ž . Ž

discrete e.g. whether or not a patient is readmitted or continuous e.g. total cost .

of a continuum of care . In fact, any type of outcome measure can be studied using a propensity score, making this method of analysis fully generalizable. The data must include the covariates needed to group subjects into strata such that subjects within each stratum are comparable and differences in outcomes can be attributed to the treatment. As well, each stratum must contain subjects from both treatment groups, although they will, by the nature of the stratification, appear in unequal numbers.

The aim of the propensity score analysis was to answer questions about the entire population, e.g. What would happen if the entire population was treated the same way — if everyone was discharged to the same destination? What would the average cost per patient be if all patients received home care? If they all were sent to a rehabilitation hospital? Are the costs significantly different, and if so, by how much? How would patient outcomes differ under each scenario?

When treatment assignment and treatment outcome are confounded,

instrumen-Ž . Ž .

tal variables IVs analysis Kmenta, 1971 is another method that can be used to adjust for unobserved covariates that affect treatment assignment. The attractive-ness of IV analysis is that, unlike other methods of adjustment, the researcher does

Ž not need to identify and measure the sources of treatment assignment bias Sturm,

.

1997 . However, finding IVs is likely to depend on an understanding of underlying relationships.

Ž .

The technique of IVs relies on the identification of variable s which satisfy two conditions: they are strongly correlated with the choice of treatment; but they

Ž

have no direct effect on the outcome their effect on the outcome is exerted only .

through their influence on the selection of treatment . If these two conditions are Ž .

not met, the IV technique is not useful. When the IV s are only weakly correlated Ž .

with treatment, then 1 there is a substantial loss in statistical power relative to Ž .

that obtained with standard regression models and 2 even a weak correlation Ž .

between the IV s and the outcome variable results in a large inconsistency in the

Ž .

estimators the very thing the IV estimation method was designed to avoid

ŽBound et al., 1995; Newhouse and McClellan, 1998; Sturm, 1997 . The identifi-.

cation of a valid IV, one which has a strong effect on the choice of treatment, but is not associated with the outcome, can require some ingenuity.

Both the propensity score approach and the IV approach require assumptions, which cannot be fully tested and which rely on the researchers’ insight in the

subject area. The propensity score adjusts for treatment assignment bias in terms of the identified covariates; it is presumed that this is sufficient to also adjust for unobserved covariates. IV analysis presumes that one is able to identify one or

Ž .

more variables, which are: 1 strongly correlated with treatment assignment and

Ž .2 uncorrelated with the outcome other than through their effect on the treatment

assignment. These requirements make IV analysis inapplicable to this study, since we do not believe that we have any IVs in our data. It would be difficult to convince anyone that patient age, gender, comorbid conditions, or CMG were not predictors of outcome as well as of discharge destination. Revision surgery is more costly than non-revision JR; treatment in a teaching hospital costs more than treatment in a non-teaching hospital. The most likely candidate for an IV is urbanrrural patient residence. This variable is a significant predictor of discharge

Ž

destination in particular, 47.4% of rural residents were discharged to home care and 10.8% to a rehabilitation hospital, compared with 38.7% and 21.7%,

respec-.

tively, of urban residents . However, urbanrrural residence turns out also to be a

good predictor of the amount of rehabilitation a patient will receive and therefore directly affects one of the outcome variables — total cost.

Ž One other technique bears mentioning: the Heckman selection model

Heck-.

man, 1976 . This technique is not appropriate in the current study, because is was not designed to address the question being investigated. It is included because a number of people asked why it had not been used. In the current study, there are no individuals for whom the cost of care is unknown. In a study intended to predict the cost of JR surgery for individuals who, for whatever reason, were eligible for but did not receive the surgery, it might have been an appropriate method of analysis.

Irrespective of the JR discharge destination, standardized acute care readmis-sion rates increased from 170.3 per 10,000 JRs in FY91 to 222.5 in FY94. Despite the obvious adverse consequences to a patient’s quality of life associated with a preventable acute care readmission, they were infrequent events. Moreover, while the cost–outcome analysis demonstrated that modifications to discharge strategies may lower readmission rates, the incremental cost of each saved readmission varied from $288,210 to $611,634 for all JR patients to $254,477 to $630,668 for low-risk JR patients. For society to be willing to engage in policies that lower readmission rates, it is necessary that the adverse consequences to a patient’s quality of life as well as society’s willingness to pay to prevent such readmissions be sufficiently high. Health planners and policy makers must recognize that while modified patterns of practice may lower readmission rates, such modified patterns of care may also result in non-trivial cost increases.

The funding and delivery of home care services, either as an alternative to institutional care or as a complementary delivery modality, have become

promi-Ž .

nent political and economic issues Coyte and Young, 1997a,b; Williams, 1996 . Since JR patients frequently experience extended periods of hospitalization despite low intensity service needs, home rehabilitation may result in both cost-savings to

Ž

the health system through shorter inpatient stays Kenney, 1991, 1993b; Kenney

. Ž

and Dubay, 1992; Parker et al., 1991 and enhanced patient satisfaction Parker et .

al., 1991; Scott et al., 1988; Sikorski and Senior, 1993 . However, our results demonstrate that patients discharged with home care services have longer, not shorter, acute care hospitalizations than other patients, that the provision of home care services increases the total cost of a continuum care, and that adverse health

Ž .

outcomes measured by acute care readmission rates were greater for patients discharged with home care services than other patients. These results complement a recent national study concerning the use of home health care by US Medicare

Ž .

beneficiaries across metropolitan areas Welch et al., 1996 , and suggest that great care needs to be taken in any proposed restructuring exercise.

There is an urgent need to comprehensively evaluate the costs and conse-quences, from a societal perspective, of variations in clinical practices for common

Ž .

episodes of care Jacobs et al., 1995; Parr, 1996 . Such episodes encompass the care continuum from community-based services, offered by health professionals and informal care-givers, through institutional care, and eventually, back into the community following discharge from a hospital. By adopting an episode-of-care

Ž .

approach Gonnella et al., 1984; Hornbrook, 1995; Hornbrook et al., 1985 , the emphasis shifts away from discrete interventions to consideration of the costs incurred and benefits derived through the provision of a whole range of treatments. Our cost–outcome analysis of alternative discharge strategies for JR patients was limited in a number of respects. First, a partial health system perspective was taken in this study instead of either a comprehensive analysis or a societal perspective. The societal perspective would have resulted in a more complete analysis of the costs and outcomes associated with a continuum of care as all costs and consequences would have been assessed regardless of the payor or the

Ž .

beneficiary Lave and Leinhardt, 1976 . To the extent to which modified clinical practices shift costs towards patients and their families, a societal perspective would have been the appropriate approach to measure the magnitude of such cost-shifting. Primary data collection would be required to operationalize this societal perspective. Moreover, since the only non-institutional costs assessed were those borne through the provision of home care and home-making services, we may have underestimated the full cost of an episode of JR care. This underestima-tion may have resulted in a systematic downward bias in the relative cost of those discharge destinations where outpatient rehabilitation is commonly received. Un-fortunately, a provincial database is not available to reliably assess the use of

Ž .

outpatient rehabilitation Williams and Young, 1996 .

Second, since acute care readmission rates were the only health outcome assessed in this study, the inclusion of other potential outcomes, such as quality-adjusted life years, may have yielded a more complete assessment of the impact of alternative discharge strategies on health outcomes. In the absence of a primary data collection exercise, the outcomes assessed in this paper were those validated

Ž .

Third, our study was based on hospital discharge data acquired from the CIHI and home care data obtained from the Ontario Ministry of Health. The main concern associated with the use of such data pertains to its validity and reliability

Ž_{Williams and Young, 1996 . Studies that have compared administrative data to}.

hospital chart data have overwhelmingly concluded that major events, such as surgical procedures and mortality, patient demographics, and primary diagnosis

Ž .

are coded accurately Harris and Sledge, 1990; Hawker et al., 1997 . However, Ž

complications and comorbid conditions are often miscoded Harris and Sledge, .

1990 . While these results may allay some concerns with respect to the CIHI data, there has yet to be an equivalent comprehensive assessment of the home care database. Moreover, to overcome potential concerns with the discharge destination field in the CIHI data, discharge destination assignment in our study was deter-mined by linking the index hospitalization to subsequent rehabilitation hospital

admissions andror the receipt of services documented in the OHCAS database.

Fourth, the health system costs derived in this study were based on cost estimates from a variety of sources. Specifically, index acute care costs were based

Ž

on the OCCP data Joint Planning and Policy Committee, 1995; Ontario Case .

Costs Subcommittee, 1991 and applied to CIHI administrative data. While we were able to account for over half of the variation in OCCP case costs, the cost estimates were not validated for non-OCCP hospitalizations. The cost of rehabilita-tion hospitalizarehabilita-tions were based on an estimated per diem cost and the length of the inpatient stay. Since such estimates reflect average instead of incremental costs, the cost-savings from shorter lengths of stay may overestimate the cost-sav-ings, particularly for patients with relatively low service intensity needs. Home care costs were based on the number of services rendered and fully allocated unit cost estimates derived from the Metropolitan Toronto Home Care Program. If these unit costs vary systematically in Ontario, home care costs in some regions may be over- or underestimated. Thus, our ability to generalize our episode-of-care costs may be limited.

Lastly, while several alternative analyses were conducted to control for poten-tial bias in the assignment of patients to various discharge destinations, we cannot rule out the possibility that our adjustments were deficient in some respects. For example, since our analyses were confined to administrative data, many

unmea-Ž

sured characteristics, such as the intensity of co-existent disease Greenfield et al., .

1993 , living arrangements, etc., were not included. Since these characteristics have been shown to be associated with the assignment of discharge destination and postoperative outcomes, our analyses did not control for all potential confounding

Ž

variables. However, wide regional and institutional variation Coyte and Young, .

1997c in the use of home care services and rehabilitation hospitals make it implausible that the relationship between discharge destination and both cost and outcomes was primarily an artifact of differential case mix.

In conclusion, despite the prevalence of musculoskeletal disorders, the high propensity to intervene surgically and the relevance of costs, there is a paucity of

cost–outcome studies associated with the continuum of care for JR patients. While

Ž .

economic and outcome data collected as part of a randomized clinical trial RCT

Ž .

are the ‘gold standard’ for comparing treatments Drummond and Davies, 1991 , an RCT would have been expensive in time and research funds, and for some groups may be neither feasible nor ethically appropriate. Our aim was more modest. We measured health system costs and readmission rates, after adjusting for potential assignment bias, as they represent an essential ingredient in any subsequent attempts to analyze service cost-effectiveness. Our results demon-strated that while modifications to discharge strategies resulting in the increased use of rehabilitation hospitals may lower readmission rates, the incremental cost of each saved readmission may be prohibitive. Moreover, patients discharged with home care services had longer acute care stays than other patients, that the provision of home care services increased health system costs and that acute care readmission rates were greatest among patients discharged with home care ser-vices. As such, our study should be seen as one important stepping stone towards a full economic evaluation of the continuum of care for JR patients.

Acknowledgements

We thank Dr. C. David Naylor and members of our Home Care Research Advisory Group for their advice on this paper. This research was supported by the Institute for Clinical Evaluative Sciences in Ontario. Dr. Coyte is supported by grants from the Ontario Ministry of Health to the Institute for Clinical Evaluative Sciences, to the Hospital Management Research Unit and to the Arthritis Commu-nity Research and Evaluation Unit. The opinions expressed are those of the authors and do not necessary reflect the opinions of any funding agency, institu-tion, or the corporate policy of the Institute for Clinical Evaluative Sciences in Ontario.

References

Anderson, G.F., Lave, J., Russe, C., Neuman, P., 1989. Providing Hospital Services: The Changing Financial Environment. Johns Hopkins Univ. Press, Baltimore, MD.

Badley, E.M., Arnold, J., Webster, G., 1994. Arthritis in Ontario: A Study of Intra-Provincial

Ž .

Variation. Toronto: Arthritis Community Research and Evaluation Unit. Working paper 94-1. Bound, J., Jaeger, D.A., Baker, R.M., 1995. Problems with instrumental variables estimation when the

correlation between the instruments and the endogenous explanatory variable is weak. J. Am. Stat. Assoc. 90, 443–450.

Charlson, M.E., Pompei, P., Ales, K.L., MacKenzie, C.R., 1987. A new method of classifying

Ž .

prognostic comorbidity in longitudinal studies: development and validation. J. Chronic Dis. 40 5 , 373–383.

Cohen, M.M., De Boer, D.P., Young, W., 1994a. Total hip replacement. In: Naylor, C.D., Anderson,

Ž .

G.M., Goel, V. Eds. , Patterns of Health Care in Ontario. 1st edn. Canadian Medical Association, Ottawa, pp. 72–76.

Cohen, M.M., De Boer, D.P., Young, W., 1994b. Total knee replacement. In: Naylor, C.D., Anderson,

Ž .

G.M., Goel, V. Eds. , Patterns of Health Care in Ontario. 1st edn. Canadian Medical Association, Ottawa, pp. 76–79.

Coyte, P.C., Young, W., 1997a. An Overview of Ontario’s Home Care Program and Suggestions for a National Home Care Research Agenda. Institute for Clinical Evaluative Sciences, North York.

Ž .

Coyte, P.C., Young, W., 1997b. Applied home care research. Leadership Health Serv. 10 1 , 1–4. Coyte, P.C., Young, W., 1997c. Variations in the Use of and Reinvestments in Home Care Services

Following an Acute Care Hospitalization. Institute for Clinical Evaluative Sciences, North York. Coyte, P.C., Wright, J.G., Freund, D.A., Bombardier, C., Heck, D.A., Hawker, G.A., 1994. Regional

variation in the utilization of knee replacement procedures and the supply of orthopaedic surgeons

Ž .

in Ontario, Canada. In: Boan, J.A. Ed. , Proceedings of the Fifth Canadian Conference on Health Economics. Canadian Plains Research Center, University of Regina, Regina.

Coyte, P.C., Hawker, G.A., Wright, J.G., 1996a. Variations in knee replacement utilization rates and

Ž .

the supply of health professionals in Ontario, Canada. J. Rheumatol. 23 7 , 1214–1219. Coyte, P.C., Young, W., Williams, J.I., 1996b. The potential for reallocation of funds from teaching

Ž .

hospitals to community hospitals by devolving hip and knee replacements. Can. J. Surg. 39 5 , 373–378.

Cromwell, J., Mitchell, J., Calore, K., Iezzoni, L., 1987. Sources of hospital cost variation by

Ž .

urban–rural location. Med. Care 25 9 , 801–829.

Deyo, R.A., Cherkin, D.C., Ciol, M.A., 1992. Adapting a clinical comorbidity index for use with

Ž .

ICD-9-CM administrative databases. J. Clin. Epi. 45 6 , 613–619.

Drummond, M.F., Davies, L.M., 1991. Economic analysis alongside clinical trials: revisiting the methodological issues. Int. J. Tech. Assessment Health Care 7, 561–573.

Drummond, M.F., Stoddart, G.L., Torrance, G.W., 1987. Methods for the Economic Evaluation of Health care Programmes. Oxford Univ. Press, Oxford.

FederalrProvincialrTerritorial Working Group on Home Care, 1990. A Working Group of the FederalrProvincialrTerritorial Subcommittee on Long Term Care: Report on Home Care. Ottawa: Health Services Promotion Branch, Health and Welfare Canada.

Feigenbaum, S., Anderson, G.F., Lave, J., 1992. Medicare’s prospective payment system: the victim of aggregation bias? Rev. Econ. Stat. 74, 185–191.

Gonnella, J.S., Hornbrook, M.C., Louis, D.Z., 1984. Staging of disease: a case-mix measurement. JAMA, J. Am. Med. Assoc. 251, 637–644.

Greenfield, S., Apolone, G., McNeil, B.J., Cleary, P.D., 1993. The importance of co-existent disease in the occurrence of postoperative complications and one-year recovery in patients undergoing total

Ž .

hip replacement. Med. Care 31 2 , 141–154.

Hadler, N.M., 1985. Osteoarthritis as a public health problem. Clin. Rheum. Dis. 11, 175–185. Harris, W.H., Sledge, C.B., 1990. Total hip and total knee replacement. N. Engl. J. Med. 323,

801–807.

Hawker, G.A., Coyte, P.C., Paul, J.E., Wright, J.G., Bombardier, C., 1997. Accuracy of administrative claims data in assessing patient characteristics and course in hospital for knee replacement surgery. J. Clin. Epi. 50, 265–273.

Heck, D.A., Melfi, C.A., Mamlin, L.A., Katz, B.P., Arthur, D.S., Dittus, P.S., Freund, D.A., 1998. Revision rates after knee replacement in the United. Med. Care 36, 661–669.

Heckman, J., 1976. The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Ann. Econ. Soc. Meas. 5, 475–492.

Hendricks, A., Cromwell, J., 1989. Are rural referral centers as costly as urban hospitals? Health Serv.

Ž .

Res. 24 3 , 289–309.

Hollander, M., 1994. The Costs and Cost-Effectiveness of Continuing Care Services in Canada. Queen’s-University of Ottawa Economic, Ottawa, pp. 1–113.

Ž .

Studies. In: Grady, M.L., Weis, K.A. Eds. , Cost Analysis Methodology for Clinical Practice

Ž

Guidelines. US Department of Health and Human Services, Rockville, MD, Agency for Health

.

Care Policy and Research Publication No. 95-0001 .

Hornbrook, M.C., Hurtado, A.V., Johnson, R.E., 1985. Health care episode: definition: measurement, and use. Med. Care Rev. 42, 163–218.

Hospital Medical Records Institute, 1993. Resource Intensity Weights: Summary of Methodology. Hospital Medical Records Institute, Toronto.

Jacobs, P., Hall, E., Henderson, I., Nichols, D., 1995. Episodic acute care costs: linking inpatient and home care. Can. J. Public Health 86, 200–205.

Jessen, R.J., 1978. Statistical Survey Techniques. Wiley, New York.

Ž .

Joint Planning and Policy Committee, 1995. Ontario Case Costs. JPPC News 3 1 , 4.

Jonsson, H., Larsson, B.J., 1991. Functional improvement and cost of hip and knee arthroplasty in destructive rheumatoid arthritis. Scand. J. Rheumatol. 20, 351–357.

Kenney, G.M., 1991. Understanding the effects of PPS on medicare home health use. Inquiry 28, 129–139.

Kenney, G.M., 1993a. How access to long-term care affects home health transfers. J. Health Pol. Policy Law 83, 412–414.

Kenney, G.M., 1993b. Is access to home health care a problem in rural areas? Am. J. Public Health 83, 412–414.

Kenney, G.M., Dubay, L.C., 1992. Explaining area variation in the use of medicare home health services. Med. Care 30, 43–57.

Kmenta, J., 1971. Elements of Econometrics. 2nd edn. MacMillan, New York, NY.

Kramert, J.S., Yelin, E.H., Epstein, W.V., 1983. Social and economic impacts of four musculoskeletal conditions: a study using national community-based data. Arthritis Rheum. 26, 901–907. Laupacis, A., Bourne, R., Rorabeck, C., Feeny, D., Wong, C., Tugwell, P. et al., 1993. The effect of

Ž .

elective total hip replacement on health-related quality of life. J. Bone Jt. Surg. 75-A 11 , 1619–1626.

Lave, J., Leinhardt, S., 1976. The cost and length of a hospital stay. Inquiry 13, 327–343.

Liang, M.H., Larson, M.G., Thompson, M.S., Eaton, H., McNamara, E., Katz, R., Taylor, J., 1984. Cost and outcomes in rheumatoid arthritis and osteoarthritis. Arthritis Rheum. 27, 522–529. Liang, M.H., Cullen, K.E., Larson, M.G., Thompson, M.S., Schwartz, J.A., Fossel, A.H., Roberts,

W.N., Sledge, C.B., 1986. Cost-effectiveness of total joint arthroplasty in osteoarthritis. Arthritis Rheum. 29, 937–943.

McAlindon, T.E., Cooper, C., Kirwan, J.R., Dieppe, P.A., 1992. Knee pain and disability in the community. Br. J. Rheumatol. 31, 189–192.

Melfi, C.A., Holleman, E., Arthur, D., Katz, B., 1995. Selecting a patient characteristic index for the

Ž .

prediction of medical outcomes using administrative data. J. Clin. Epi. 48 7 , 917–926. Moskowitz, R.W., Howell, D.S., Goldberg, V.M., Mankin, H.J., 1992. Osteoarthritis: Diagnosis and

MedicalrSurgical Management. 2nd edn. Saunders, Toronto.

Naylor, C.D., De Boer, D.P., 1996. Total hip and knee replacement. In: Goel, V., Williams, J.I.,

Ž .

Anderson, G., Blackstein-Hirsch, P., Fooks, C., Naylor, C.D. Eds. , Patterns of Health Care in Ontario. 2nd edn. Canadian Medical, Ottawa, pp. 54–63.

Newhouse, J.P., McClellan, M., 1998. Econometrics in outcomes research: the use of instrumental variables. Annu. Rev. Public Health 19, 17–34.

Ontario Case Costs Subcommittee, 1991. Development of a Case Costing Methodology for Ontario Hospitals. Phase I: Feasibility and advisability, Final Report. Toronto: Ontario Case Costs Subcomittee.

Parker, M.J., Pryor, G.A., Myles, J.W., 1991. Early discharge after hip fracture. Prospective 3-year study of 645 Patients. Acta Orthop. Scand. 62, 563–566.

Parr, K., 1996. The Cost Effectiveness of Home Care: A Literature Review. Health Services Utilization and Research Commission, Saskatoon, Saskatchewan.

Peyron, J.G., Altman, R.D., 1992. The epidemiology of osteoarthritis. In: Moskowitz, R.W., Howell,

Ž .

D.S., Goldberg, V.M., Mankin, H.J. Eds. , Osteoarthritis: Diagnosis and MedicalrSurgical Management. 2nd edn. Saunders, Toronto.

Pincus, T., Mitchell, J.M., Burkhauser, R.V., 1989. Substantial work disability and earnings losses in individuals less than age 65 with osteoarthritis: comparisons with rheumatoid arthritis. J. Clin. Epi. 42, 449–457.

Pink, G.H., Bolley, H.B., 1994a. Physicians in health care management: 3. Case mix groups and

Ž .

resource intensity weights: an overview for physicians. Can. Med. Assoc. J. 150 6 , 889–894. Pink, G.H., Bolley, H.B., 1994b. Physicians in health care management: 4. Case mix groups and

Ž .

resource intensity weights: physicians and hospital funding. Can. Med. Assoc. J. 150 8 , 1255–1261.

Price Waterhouse, 1989. Operational Review of the Ontario Home Care Program. Final Report. Price Waterhouse, Toronto.

Rehabilitation Planning Committee, 1995. Proposal for The Metropolitan Toronto Rehab Network. Rehabilitation Planning Committee, Toronto, March.

Riley, G., Lubitz, J., Gornick, M., Mentnech, R., Eggers, P., McBean, M., 1993. Medicare

beneficia-Ž .

ries: adverse outcomes after hospitalization for eight procedures. Med. Care 31 10 , 921–949. Romano, P.S., Roos, L.L., Jollis, J.G., 1993a. Adapting a clinical comorbidity index for use with

Ž .

ICD-9-CM administrative data: differing perspectives. J. Clin. Epi. 46 10 , 1075–1079. Romano, P.S., Roos, L.L., Jollis, J.G., 1993b. Further evidence concerning the use of a clinical

Ž .

comorbidity index with ICD-9-CM administrative data. J. Clin. Epi. 46 10 , 1085–1090. Rorabeck, C.H., Bourne, R.B., Laupacis, A., Feeny, D., Wong, C., Tugwell, P. et al., 1994. A

double-blind study of 250 cases comparing cemented with cementless total hip arthroplasty. Cost-effectiveness and its impact on health-related quality of life. Clin. Orthop. Relat. Res. 298, 156–164.

Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrica 70, 41–55.

Rosenbaum, P.R., Rubin, D.B., 1984. Reducing bias in observational studies using subclassification on the propensity score. J. Am. Stat. Assoc. 79, 516–524.

Rubin, D.B., 1997. Estimating causal effects from large data sets using propensity scores. Ann. Int.

Ž .

Med. 127 8S , 757–763.

Scott, N., Rubinstein, M., Scuderi, G., 1988. Results after knee replacement with a posterior cruciate-substituting prosthesis. J. Bone Jt. Surg. 70-A, 1163–1173.

Sikorski, J.M., Senior, J., 1993. The domiciliary rehabilitation and support program: rationale, organization, and outcome. Med. J. Aust. 159, 23–25.

Statistics Canada, 1993. Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures. Second Printing. Minister of Industry, Science and Technology, Ottawa.

Statistics Canada, 1994. Health Status of Canadians. Report of the 1991 General Social Survey.

Ž .

Ottawa: Supply and Services, Canada. Catalogue No. 11-612E, No. 8.

Ž .

Statistics Canada, 1996. Consumer Price Index. Catalogue No. 62-001.

Sturm, R., 1997. Instrumental variable methods for effectiveness research. Int. J. Methods Psychiatr. Res. 7, 17–26.

The MIS Group, 1985. Guidelines for Management Information Systems in Canadian Health Care Facilities. The MIS Group, Ottawa.

Verbrugge, L.M., 1991. Physical and social disability in adults. In: Hibbard, H.H., Nutting, P.A.,

Ž .

Grady, M.L. Eds. , Functional and Health Status Measures for Primary Care. 1st edn. US Department of Health and Human Services, Washington, DC.

Weissert, W., 1985. Seven reasons why it is so difficult to make community-based long-term care

Ž .

cost-effective. Health Serv. Res. 20 4 , 423–433.

Ž .

Weissert, W.G., 1991. A new policy agenda for home care. Health Aff. 10 2 , 67–77.

Weissert, W., Wan, T., Livieratos, B.B., Pellegrino, J., 1980. Effects and costs of day-care services for

Ž .

Weissert, W.G., Cready, C.M., Pawelak, J.E., 1988. The past and future of home and community-based

Ž .

long-term care. Milbank Q. 66 2 , 309–388.

Welch, H.G., Wennberg, D.E., Welch, W.P., 1996. The use of medicare home health care services. N.

Ž .

Engl. J. Med. 335 5 , 324–329.

Williams, A.H., 1996. The development of Ontario’s home care program: a critical geographical

Ž .

analysis. Soc. Sci. Med. 42 6 , 937–948.

Williams, J.I., Young, W., 1996. A summary of studies on the quality of health care administrative databases in Canada. In: Goel, V., Williams, J.I., Anderson, G., Blackstein-Hirsch, P., Fooks, C.,

Ž .

Naylor, C.D. Eds. , Patterns of Health Care in Ontario. 2nd edn. Canadian Medical, Ottawa, pp. 339–345.

World Health Organization, 1977. International Classification of Diseases. World Health Organization, Geneva.

Third, our study was based on hospital discharge data acquired from the CIHI

and home care data obtained from the Ontario Ministry of Health. The main

concern associated with the use of such data pertains to its validity and reliability

Ž

_{Williams and Young, 1996 . Studies that have compared administrative data to}

.

hospital chart data have overwhelmingly concluded that major events, such as

surgical procedures and mortality, patient demographics, and primary diagnosis

Ž

.

are coded accurately Harris and Sledge, 1990; Hawker et al., 1997 . However,

Ž

complications and comorbid conditions are often miscoded Harris and Sledge,

.

1990 . While these results may allay some concerns with respect to the CIHI data,

there has yet to be an equivalent comprehensive assessment of the home care

database. Moreover, to overcome potential concerns with the discharge destination

field in the CIHI data, discharge destination assignment in our study was

deter-mined by linking the index hospitalization to subsequent rehabilitation hospital

admissions and

r

or the receipt of services documented in the OHCAS database.

Fourth, the health system costs derived in this study were based on cost

estimates from a variety of sources. Specifically, index acute care costs were based

Ž

on the OCCP data Joint Planning and Policy Committee, 1995; Ontario Case

.

Costs Subcommittee, 1991 and applied to CIHI administrative data. While we

were able to account for over half of the variation in OCCP case costs, the cost

estimates were not validated for non-OCCP hospitalizations. The cost of

rehabilita-tion hospitalizarehabilita-tions were based on an estimated per diem cost and the length of

the inpatient stay. Since such estimates reflect average instead of incremental

costs, the cost-savings from shorter lengths of stay may overestimate the

cost-sav-ings, particularly for patients with relatively low service intensity needs. Home

care costs were based on the number of services rendered and fully allocated unit

cost estimates derived from the Metropolitan Toronto Home Care Program. If

these unit costs vary systematically in Ontario, home care costs in some regions

may be over- or underestimated. Thus, our ability to generalize our episode-of-care

costs may be limited.

Lastly, while several alternative analyses were conducted to control for

poten-tial bias in the assignment of patients to various discharge destinations, we cannot

rule out the possibility that our adjustments were deficient in some respects. For

example, since our analyses were confined to administrative data, many

unmea-Ž

sured characteristics, such as the intensity of co-existent disease Greenfield et al.,

.

1993 , living arrangements, etc., were not included. Since these characteristics

have been shown to be associated with the assignment of discharge destination and

postoperative outcomes, our analyses did not control for all potential confounding

Ž

variables. However, wide regional and institutional variation Coyte and Young,

.

1997c in the use of home care services and rehabilitation hospitals make it

implausible that the relationship between discharge destination and both cost and

outcomes was primarily an artifact of differential case mix.

In conclusion, despite the prevalence of musculoskeletal disorders, the high

propensity to intervene surgically and the relevance of costs, there is a paucity of

cost–outcome studies associated with the continuum of care for JR patients. While

Ž

.

economic and outcome data collected as part of a randomized clinical trial RCT

Ž

.

are the ‘gold standard’ for comparing treatments Drummond and Davies, 1991 ,

an RCT would have been expensive in time and research funds, and for some

groups may be neither feasible nor ethically appropriate. Our aim was more

modest. We measured health system costs and readmission rates, after adjusting

for potential assignment bias, as they represent an essential ingredient in any

subsequent attempts to analyze service cost-effectiveness. Our results

demon-strated that while modifications to discharge strategies resulting in the increased

use of rehabilitation hospitals may lower readmission rates, the incremental cost of

each saved readmission may be prohibitive. Moreover, patients discharged with

home care services had longer acute care stays than other patients, that the

provision of home care services increased health system costs and that acute care

readmission rates were greatest among patients discharged with home care

ser-vices. As such, our study should be seen as one important stepping stone towards a

full economic evaluation of the continuum of care for JR patients.

Acknowledgements

We thank Dr. C. David Naylor and members of our Home Care Research

Advisory Group for their advice on this paper. This research was supported by the

Institute for Clinical Evaluative Sciences in Ontario. Dr. Coyte is supported by

grants from the Ontario Ministry of Health to the Institute for Clinical Evaluative

Sciences, to the Hospital Management Research Unit and to the Arthritis

Commu-nity Research and Evaluation Unit. The opinions expressed are those of the

authors and do not necessary reflect the opinions of any funding agency,

institu-tion, or the corporate policy of the Institute for Clinical Evaluative Sciences in

Ontario.

References

Anderson, G.F., Lave, J., Russe, C., Neuman, P., 1989. Providing Hospital Services: The Changing Financial Environment. Johns Hopkins Univ. Press, Baltimore, MD.

Badley, E.M., Arnold, J., Webster, G., 1994. Arthritis in Ontario: A Study of Intra-Provincial

Ž .

Variation. Toronto: Arthritis Community Research and Evaluation Unit. Working paper 94-1. Bound, J., Jaeger, D.A., Baker, R.M., 1995. Problems with instrumental variables estimation when the

correlation between the instruments and the endogenous explanatory variable is weak. J. Am. Stat. Assoc. 90, 443–450.

Charlson, M.E., Pompei, P., Ales, K.L., MacKenzie, C.R., 1987. A new method of classifying

Ž .

prognostic comorbidity in longitudinal studies: development and validation. J. Chronic Dis. 40 5 , 373–383.

Cohen, M.M., De Boer, D.P., Young, W., 1994a. Total hip replacement. In: Naylor, C.D., Anderson,

Ž .

G.M., Goel, V. Eds. , Patterns of Health Care in Ontario. 1st edn. Canadian Medical Association, Ottawa, pp. 72–76.

Cohen, M.M., De Boer, D.P., Young, W., 1994b. Total knee replacement. In: Naylor, C.D., Anderson,

Ž .

G.M., Goel, V. Eds. , Patterns of Health Care in Ontario. 1st edn. Canadian Medical Association, Ottawa, pp. 76–79.

Coyte, P.C., Young, W., 1997a. An Overview of Ontario’s Home Care Program and Suggestions for a National Home Care Research Agenda. Institute for Clinical Evaluative Sciences, North York.

Ž .

Coyte, P.C., Young, W., 1997b. Applied home care research. Leadership Health Serv. 10 1 , 1–4. Coyte, P.C., Young, W., 1997c. Variations in the Use of and Reinvestments in Home Care Services

Following an Acute Care Hospitalization. Institute for Clinical Evaluative Sciences, North York. Coyte, P.C., Wright, J.G., Freund, D.A., Bombardier, C., Heck, D.A., Hawker, G.A., 1994. Regional

variation in the utilization of knee replacement procedures and the supply of orthopaedic surgeons

Ž .

in Ontario, Canada. In: Boan, J.A. Ed. , Proceedings of the Fifth Canadian Conference on Health Economics. Canadian Plains Research Center, University of Regina, Regina.

Coyte, P.C., Hawker, G.A., Wright, J.G., 1996a. Variations in knee replacement utilization rates and

Ž .

the supply of health professionals in Ontario, Canada. J. Rheumatol. 23 7 , 1214–1219. Coyte, P.C., Young, W., Williams, J.I., 1996b. The potential for reallocation of funds from teaching

Ž .

hospitals to community hospitals by devolving hip and knee replacements. Can. J. Surg. 39 5 , 373–378.

Cromwell, J., Mitchell, J., Calore, K., Iezzoni, L., 1987. Sources of hospital cost variation by

Ž .

urban–rural location. Med. Care 25 9 , 801–829.

Deyo, R.A., Cherkin, D.C., Ciol, M.A., 1992. Adapting a clinical comorbidity index for use with

Ž .

ICD-9-CM administrative databases. J. Clin. Epi. 45 6 , 613–619.

Drummond, M.F., Davies, L.M., 1991. Economic analysis alongside clinical trials: revisiting the methodological issues. Int. J. Tech. Assessment Health Care 7, 561–573.

Drummond, M.F., Stoddart, G.L., Torrance, G.W., 1987. Methods for the Economic Evaluation of Health care Programmes. Oxford Univ. Press, Oxford.

FederalrProvincialrTerritorial Working Group on Home Care, 1990. A Working Group of the FederalrProvincialrTerritorial Subcommittee on Long Term Care: Report on Home Care. Ottawa: Health Services Promotion Branch, Health and Welfare Canada.

Feigenbaum, S., Anderson, G.F., Lave, J., 1992. Medicare’s prospective payment system: the victim of aggregation bias? Rev. Econ. Stat. 74, 185–191.

Gonnella, J.S., Hornbrook, M.C., Louis, D.Z., 1984. Staging of disease: a case-mix measurement. JAMA, J. Am. Med. Assoc. 251, 637–644.

Greenfield, S., Apolone, G., McNeil, B.J., Cleary, P.D., 1993. The importance of co-existent disease in the occurrence of postoperative complications and one-year recovery in patients undergoing total

Ž .

hip replacement. Med. Care 31 2 , 141–154.

Hadler, N.M., 1985. Osteoarthritis as a public health problem. Clin. Rheum. Dis. 11, 175–185. Harris, W.H., Sledge, C.B., 1990. Total hip and total knee replacement. N. Engl. J. Med. 323,

801–807.

Hawker, G.A., Coyte, P.C., Paul, J.E., Wright, J.G., Bombardier, C., 1997. Accuracy of administrative claims data in assessing patient characteristics and course in hospital for knee replacement surgery. J. Clin. Epi. 50, 265–273.

Heck, D.A., Melfi, C.A., Mamlin, L.A., Katz, B.P., Arthur, D.S., Dittus, P.S., Freund, D.A., 1998. Revision rates after knee replacement in the United. Med. Care 36, 661–669.

Heckman, J., 1976. The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Ann. Econ. Soc. Meas. 5, 475–492.

Hendricks, A., Cromwell, J., 1989. Are rural referral centers as costly as urban hospitals? Health Serv.

Ž .

Res. 24 3 , 289–309.

Hollander, M., 1994. The Costs and Cost-Effectiveness of Continuing Care Services in Canada. Queen’s-University of Ottawa Economic, Ottawa, pp. 1–113.

Ž .

Studies. In: Grady, M.L., Weis, K.A. Eds. , Cost Analysis Methodology for Clinical Practice

Ž

Guidelines. US Department of Health and Human Services, Rockville, MD, Agency for Health

.

Care Policy and Research Publication No. 95-0001 .

Hornbrook, M.C., Hurtado, A.V., Johnson, R.E., 1985. Health care episode: definition: measurement, and use. Med. Care Rev. 42, 163–218.

Hospital Medical Records Institute, 1993. Resource Intensity Weights: Summary of Methodology. Hospital Medical Records Institute, Toronto.

Jacobs, P., Hall, E., Henderson, I., Nichols, D., 1995. Episodic acute care costs: linking inpatient and home care. Can. J. Public Health 86, 200–205.

Jessen, R.J., 1978. Statistical Survey Techniques. Wiley, New York.

Ž .

Joint Planning and Policy Committee, 1995. Ontario Case Costs. JPPC News 3 1 , 4.

Jonsson, H., Larsson, B.J., 1991. Functional improvement and cost of hip and knee arthroplasty in destructive rheumatoid arthritis. Scand. J. Rheumatol. 20, 351–357.

Kenney, G.M., 1991. Understanding the effects of PPS on medicare home health use. Inquiry 28, 129–139.

Kenney, G.M., 1993a. How access to long-term care affects home health transfers. J. Health Pol. Policy Law 83, 412–414.

Kenney, G.M., 1993b. Is access to home health care a problem in rural areas? Am. J. Public Health 83, 412–414.

Kenney, G.M., Dubay, L.C., 1992. Explaining area variation in the use of medicare home health services. Med. Care 30, 43–57.

Kmenta, J., 1971. Elements of Econometrics. 2nd edn. MacMillan, New York, NY.

Kramert, J.S., Yelin, E.H., Epstein, W.V., 1983. Social and economic impacts of four musculoskeletal conditions: a study using national community-based data. Arthritis Rheum. 26, 901–907. Laupacis, A., Bourne, R., Rorabeck, C., Feeny, D., Wong, C., Tugwell, P. et al., 1993. The effect of

Ž .

elective total hip replacement on health-related quality of life. J. Bone Jt. Surg. 75-A 11 , 1619–1626.

Lave, J., Leinhardt, S., 1976. The cost and length of a hospital stay. Inquiry 13, 327–343.

Liang, M.H., Larson, M.G., Thompson, M.S., Eaton, H., McNamara, E., Katz, R., Taylor, J., 1984. Cost and outcomes in rheumatoid arthritis and osteoarthritis. Arthritis Rheum. 27, 522–529. Liang, M.H., Cullen, K.E., Larson, M.G., Thompson, M.S., Schwartz, J.A., Fossel, A.H., Roberts,

W.N., Sledge, C.B., 1986. Cost-effectiveness of total joint arthroplasty in osteoarthritis. Arthritis Rheum. 29, 937–943.

McAlindon, T.E., Cooper, C., Kirwan, J.R., Dieppe, P.A., 1992. Knee pain and disability in the community. Br. J. Rheumatol. 31, 189–192.

Melfi, C.A., Holleman, E., Arthur, D., Katz, B., 1995. Selecting a patient characteristic index for the

Ž .

prediction of medical outcomes using administrative data. J. Clin. Epi. 48 7 , 917–926. Moskowitz, R.W., Howell, D.S., Goldberg, V.M., Mankin, H.J., 1992. Osteoarthritis: Diagnosis and

MedicalrSurgical Management. 2nd edn. Saunders, Toronto.

Naylor, C.D., De Boer, D.P., 1996. Total hip and knee replacement. In: Goel, V., Williams, J.I.,

Ž .

Anderson, G., Blackstein-Hirsch, P., Fooks, C., Naylor, C.D. Eds. , Patterns of Health Care in Ontario. 2nd edn. Canadian Medical, Ottawa, pp. 54–63.

Newhouse, J.P., McClellan, M., 1998. Econometrics in outcomes research: the use of instrumental variables. Annu. Rev. Public Health 19, 17–34.

Ontario Case Costs Subcommittee, 1991. Development of a Case Costing Methodology for Ontario Hospitals. Phase I: Feasibility and advisability, Final Report. Toronto: Ontario Case Costs Subcomittee.

Parker, M.J., Pryor, G.A., Myles, J.W., 1991. Early discharge after hip fracture. Prospective 3-year study of 645 Patients. Acta Orthop. Scand. 62, 563–566.

Parr, K., 1996. The Cost Effectiveness of Home Care: A Literature Review. Health Services Utilization and Research Commission, Saskatoon, Saskatchewan.

Peyron, J.G., Altman, R.D., 1992. The epidemiology of osteoarthritis. In: Moskowitz, R.W., Howell,

Ž .

D.S., Goldberg, V.M., Mankin, H.J. Eds. , Osteoarthritis: Diagnosis and MedicalrSurgical Management. 2nd edn. Saunders, Toronto.

Pincus, T., Mitchell, J.M., Burkhauser, R.V., 1989. Substantial work disability and earnings losses in individuals less than age 65 with osteoarthritis: comparisons with rheumatoid arthritis. J. Clin. Epi. 42, 449–457.

Pink, G.H., Bolley, H.B., 1994a. Physicians in health care management: 3. Case mix groups and

Ž .

resource intensity weights: an overview for physicians. Can. Med. Assoc. J. 150 6 , 889–894. Pink, G.H., Bolley, H.B., 1994b. Physicians in health care management: 4. Case mix groups and

Ž .

resource intensity weights: physicians and hospital funding. Can. Med. Assoc. J. 150 8 , 1255–1261.

Price Waterhouse, 1989. Operational Review of the Ontario Home Care Program. Final Report. Price Waterhouse, Toronto.

Rehabilitation Planning Committee, 1995. Proposal for The Metropolitan Toronto Rehab Network. Rehabilitation Planning Committee, Toronto, March.

Riley, G., Lubitz, J., Gornick, M., Mentnech, R., Eggers, P., McBean, M., 1993. Medicare

beneficia-Ž .

ries: adverse outcomes after hospitalization for eight procedures. Med. Care 31 10 , 921–949. Romano, P.S., Roos, L.L., Jollis, J.G., 1993a. Adapting a clinical comorbidity index for use with

Ž .

ICD-9-CM administrative data: differing perspectives. J. Clin. Epi. 46 10 , 1075–1079. Romano, P.S., Roos, L.L., Jollis, J.G., 1993b. Further evidence concerning the use of a clinical

Ž .

comorbidity index with ICD-9-CM administrative data. J. Clin. Epi. 46 10 , 1085–1090. Rorabeck, C.H., Bourne, R.B., Laupacis, A., Feeny, D., Wong, C., Tugwell, P. et al., 1994. A

double-blind study of 250 cases comparing cemented with cementless total hip arthroplasty. Cost-effectiveness and its impact on health-related quality of life. Clin. Orthop. Relat. Res. 298, 156–164.

Rosenbaum, P.R., Rubin, D.B., 1983. The central role of the propensity score in observational studies for causal effects. Biometrica 70, 41–55.

Rosenbaum, P.R., Rubin, D.B., 1984. Reducing bias in observational studies using subclassification on the propensity score. J. Am. Stat. Assoc. 79, 516–524.

Rubin, D.B., 1997. Estimating causal effects from large data sets using propensity scores. Ann. Int.

Ž .

Med. 127 8S , 757–763.

Scott, N., Rubinstein, M., Scuderi, G., 1988. Results after knee replacement with a posterior cruciate-substituting prosthesis. J. Bone Jt. Surg. 70-A, 1163–1173.

Sikorski, J.M., Senior, J., 1993. The domiciliary rehabilitation and support program: rationale, organization, and outcome. Med. J. Aust. 159, 23–25.

Statistics Canada, 1993. Canadian Classification of Diagnostic, Therapeutic and Surgical Procedures. Second Printing. Minister of Industry, Science and Technology, Ottawa.

Statistics Canada, 1994. Health Status of Canadians. Report of the 1991 General Social Survey.

Ž .

Ottawa: Supply and Services, Canada. Catalogue No. 11-612E, No. 8.

Ž .

Statistics Canada, 1996. Consumer Price Index. Catalogue No. 62-001.

Sturm, R., 1997. Instrumental variable methods for effectiveness research. Int. J. Methods Psychiatr. Res. 7, 17–26.

The MIS Group, 1985. Guidelines for Management Information Systems in Canadian Health Care Facilities. The MIS Group, Ottawa.

Verbrugge, L.M., 1991. Physical and social disability in adults. In: Hibbard, H.H., Nutting, P.A.,

Ž .

Grady, M.L. Eds. , Functional and Health Status Measures for Primary Care. 1st edn. US Department of Health and Human Services, Washington, DC.

Weissert, W., 1985. Seven reasons why it is so difficult to make community-based long-term care

Ž .

cost-effective. Health Serv. Res. 20 4 , 423–433.

Ž .

Weissert, W.G., 1991. A new policy agenda for home care. Health Aff. 10 2 , 67–77.

Weissert, W., Wan, T., Livieratos, B.B., Pellegrino, J., 1980. Effects and costs of day-care services for

Ž .

Weissert, W.G., Cready, C.M., Pawelak, J.E., 1988. The past and future of home and community-based

Ž .

long-term care. Milbank Q. 66 2 , 309–388.

Welch, H.G., Wennberg, D.E., Welch, W.P., 1996. The use of medicare home health care services. N.

Ž .

Engl. J. Med. 335 5 , 324–329.

Williams, A.H., 1996. The development of Ontario’s home care program: a critical geographical

Ž .

analysis. Soc. Sci. Med. 42 6 , 937–948.

Williams, J.I., Young, W., 1996. A summary of studies on the quality of health care administrative databases in Canada. In: Goel, V., Williams, J.I., Anderson, G., Blackstein-Hirsch, P., Fooks, C.,

Ž .

Naylor, C.D. Eds. , Patterns of Health Care in Ontario. 2nd edn. Canadian Medical, Ottawa, pp. 339–345.

World Health Organization, 1977. International Classification of Diseases. World Health Organization, Geneva.