COMPARATIVE PROGRAM ON HEALTH AND SOCIETY 2001/2 WORKING PAPER WORKING PAPER

Transcription

1 COMPARATIVE PROGRAM ON HEALTH AND SOCIETY 2001/2 WORKING PAPER WORKING PAPER Access to Home Care Services in Ontario: The Role of Socio-economic Status Audrey Laporte, Ph.D.* Lupina Fellow Munk Centre for International Studies and Department of Health Policy, Management and Evaluation University of Toronto Peter C. Coyte, Ph.D. Professor of Health Economics and CHSRF/CIHR Health Services Chair Department of Health Policy, Management and Evaluation University of Toronto Ruth Croxford, B.Sc., M.Sc., M.Sc. Home and Community Evaluation Research Centre University of Toronto October, 2002 *Dr. Laporte gratefully acknowledges the financial support of the Comparative Program on Health and Society at the Munk Centre for International Studies at the University of Toronto, the Lupina Foundation and the Canadian Health Services Research Foundation (CHSRF). The views expressed in this paper are those of the authors and do not necessarily reflect those of the funding agencies.

2 1. Introduction The provision of health care has always involved a mix of individual or household non-financial resources, primarily time and effort, and purchased inputs, primarily professional care, hospital resources and pharmaceuticals, which traditionally have had to be purchased with individual or household financial resources. This has meant that the level of financial and non-financial resources would always impose a constraint on the amount of care an individual could receive, and would also play a major role in determining the type of care received. Some individuals or households might be better able to supply non-financial resources such as the time input of a household member, while others, might be more likely to make more intensive use of purchased inputs for the care of a household member. Canadian Medicare was introduced in order to ensure that no one was denied access to necessary medical care because of a lack of financial resources. Public medical and hospital insurance coverage were designed in the 1950s and 60s with the intention of easing the constraint imposed by financial resources. Certain services, which had previously been available only when purchased on the medical market, were now available at no out-of-pocket cost to the patient. Hospital building programs in particular also increased the total availability of inpatient services. The combination of increased availability and reduced price unquestionably increased access to care, especially among lower income Canadians [1-4], but it also created an incentive to substitute inpatient for ambulatory or home-based care. This tendency to favour inpatient care was strengthened by the fact that, during the early years of Medicare - until the shift to Established Programmes Financing in the mid 1970s - care provided in a doctor's office, and care provided in hospital, were eligible for federal cost- 2

3 sharing funds while care provided in a patient's home was not. The physician's home visit fee was a cost-sharable item but costs incurred by the family were not. The results of Coyte and Landon [5] suggest that provincial governments were quite sensitive to the incentives built into the cost-sharing system. From the point of view of provincial Ministries of Finance and of Health, more care could be provided for each provincial dollar spent on hospital care than on home care programs. The result was a system weighted heavily in favour of inpatient care, to the point where one study found that as many as 30% of all hospital days of care might be medically unnecessary [6]. Those hospital days were being received by patients who could not be treated on a strictly ambulatory basis but who did not have available to them any alternative, intermediate level of care. They were being hospitalised because there was nowhere else for them to receive treatment. Recently, there has been increased interest in the potential role of home care 1 as a substitute for hospital care, and some figures have been produced suggesting that the average cost of a day of home care is considerably less than the average total cost of a day of in-patient care [7]. There is also evidence that, for a significant number of patients, reduced hospital stays combined with increased home care treatment may produce health outcomes fully comparable with those achieved by hospital care alone [6]. Further, there is evidence that many patients would prefer to receive some of their treatment at home [8,9]. Combined with the budgetary costs of hospital treatment and the magnitudes of the estimates of the proportion of hospital days 1 Home care services include: nursing, physiotherapy, occupational therapy, social work, home making, personal support, meals on wheels etc. that are provided in the care recipient s home, school, or workplace. 3

4 which are rated as medically unnecessary, home care has increasingly come to be seen by some as an attractive alternative to hospitalisation. Canadian home care expenditures have increased at an annual rate of approximately 20% since 1975 [10]. In spite of this growing trend towards home care, Canadians over 65 years of age are four times as likely to receive institutional long term care than they are to receive care in their homes [11]. This contrasts with Scandinavian countries, particularly Sweden, Finland and Denmark, where those over 65 years of age are more than three times as likely to receive home care than institutional long term care [11]. The pervasive use of home care in Scandinavia is partially attributable to their demographic profile. Specifically, the very elderly (those over 80 years of age) and the elderly (those over 65 years of age) account for 4% and 15.5% of the Scandinavian population, respectively. Equivalent Canadian figures are 3% and 12.5%. The greater use of home care in Scandinavia is also based on a clear recognition that aging and the use of health care must be considered in a broad systems framework that operates in concert with cultural norms and shared beliefs about how health and social care ought to be provided to the elderly [12,13]. One problem with the current configuration of home care services in Canada is that it places greater demands on a household s non-financial resources than does care associated with in-patient care [14,15,16]. And, to the extent that medication and supplies have to be paid for by the individual, rather than being financed out of hospital budgets, it may also place heavier demands on personal financial resources. 4

5 This raises the possibility that greater reliance on home care may reintroduce significant financial barriers to access to care. As noted above, a primary objective of Canadian Medicare was to ensure that no one be denied access to care for financial reasons. To a large extent, this objective has been achieved [1-4, 17,18], although there is evidence that some socio-economic gradients in utilization of various types of care remain [19-24]. If home care is to play an increasing role in the Canadian health care system, it must be integrated in a manner that does not disadvantage the poor. Some efforts are already in place to do this - the single entry point programs in Ontario, Manitoba, New Brunswick and Newfoundland are designed to assess the adequacy of family resources in general before patients are admitted to home care programs. Despite these efforts, a socioeconomic gradient in the use of home care may exist, in which case simply extending existing structures to the population at large would not achieve Medicare's fundamental equity objective. The purpose of this paper was to evaluate the degree to which an individual's financial resources, as reflected by socio-economic status (SES), are likely to affect access to home care in Ontario. We considered access to home care services along two dimensions: propensity and intensity. Propensity refers to the probability that an individual receives service, intensity to the amount of service received, conditional on receipt. In particular, we assessed the comparative relationship between socioeconomic status and both the propensity and intensity of home care service use. 5

6 2. Data Sources All individuals (342,309) who received at least one home care visit in calendar year 1998 were identified from the Ontario Home Care Administration System (OHCAS) database, obtained from the Ontario Ministry of Health and Long Term Care. OHCAS provides information about the type and amount of publicly funded home care services provided to care recipients in Ontario. The Registered Persons Data Base (RPDB) was used to identify the age, sex, date of death and place of residence for 11,583,921 Ontario residents who had Ontario Health Insurance Plan (OHIP) coverage for all of calendar year Analyses incorporated all individuals who received at least one home care visit during calendar year 1998, along with a 10 percent subset of non-home care recipients, selected at random from the RPDB 2. Morbidity information was derived from Hospital Discharge data, provided by the Canadian Institutes of Health Information, and Physician Claims data, provided by the Ontario Ministry of Health and Long Term Care. Place of residence was characterized by the Forward Sortation Area (FSA - the first three elements of the postal code) and was linked to 1996 Statistics Canada Census data in order to obtain ecological socio-economic descriptors. Place of residence was also characterized by census subdivision (CSD) in order to link observations to a rurality index, described later. All individuals, whether they received home care or not, were assigned to one of the 43 mutually exclusive and exhaustive home care regions (Community Care Access Centres) on the basis of their Forward Sortation Area 3. Those who: i) resided outside of Ontario, 2 A stratified random sample of non-home care users was taken in which the elderly were oversampled. 3 See Appendix 1 for a listing of the CCACs. 6

7 ii) iii) died prior to the end of the year, did not have OHIP coverage for the entire year (i.e. with lapses of no more than 30 days) or, iv) could not be assigned a home care program by a match between their postal code and the home care program in the postal code look up file, were excluded from the analysis. 4 The final sample contained 1,385,265 individuals of whom 297,497 received home care. Episodes of Care We distinguished between short- and long-term home care services in the analysis because short-term service use is generally associated with acute care follow-up (i.e. care following hospitalization) while long-term service use is generally associated with chronic care (i.e. assistance with activities of daily living). The literature gives no guidance as to what constitutes an episode of home care, nor to the specific length of time or number of visits that ought to constitute short- as opposed to long-term care. Our heuristic definition is described below. An 'intermediate' category was included, to ensure good separation between short- and long-term. The first home care visit of calendar year 1998 was identified for all recipients who received at least one visit in An episode of home care was terminated by a gap of at least 5 weeks (35 days) with no home care visits. The episode to which this first home care visit belonged was classified as follows: 4 Of 335,918 individuals who received at least one home care visit in calendar year 1998, 31,761 (9.5%) died during the year, and were excluded from the analysis. A further 4,913 did not have a valid Ontario postal code for at least part of the year, and 1,747 had a postal code which could not be linked with a CCAC. The final dataset contained 297,497 home care recipients (88.6% of the original home care population). 7

8 i) If the episode lasted no more than 91 days (13 weeks), it was termed shortterm. ii) If the episode lasted 119 days (17 weeks) or more then the episode was termed long-term. iii) Otherwise, the episode was termed intermediate. At least 153 days of home care history were used to categorize an episode. In some cases, this involved the use of information about visits received in calendar years 1997 and For example, if the first visit of 1998 occurred early in the year, data from as far back as September 1, 1997 might have been used in order to characterize the home care episode. If the first home care visit of 1998 occurred on October 31, 1998, information from as far forward as March 31, 1999 might have been used. March 31, 1999 is the last day for which data are available (See Figure 1). Thus, only individuals whose first home care visit in 1998 occurred before November 1, 1998 could be classified. This left 266,767 individuals, or 89.7% of the original sample. Figure 1 Classification of Home Care Episodes Date of First Home Care Visit in CY 1998 If first visit was early in 1998, may have to search back in time, examining home care visits from 1997, in order to classify the episode. Search forward in time to find the end of the episode. If first visit was late in 1998, may have to examine home care visits from 1999 in order classify the episode. Sept 1, 1998 Jan 1 Dec 31 Mar 31, 1999 (start of CY) (end of CY) (end of data) Based on the above definition, these home care clients were categorized as follows: 48.6% short-term, 47.6% long-term, 3.8% intermediate. The discussion which follows focuses on the characteristics of short and long-term care episodes. 8

9 3. Methods We considered both the propensity and intensity of home care utilization, in recognition of the fact that provision of service is a function of two distinct processes. Once an individual qualifies for home care, the amount of service received may, for example, depend on resource availability in the individual's region. Also, factors such as health status may have a differential impact on the probability of receiving care, as opposed to on the amount of care received. As a consequence, the analysis was conducted in the context of a two-part model structure. In the first part, the propensity or probability of receiving home care was estimated using a multi-nomial logit equation. The base case for the model was receives no home care. Coefficient estimates are interpreted as log odds-ratios. The model shows how the probability of receiving short-term home care versus none, intermediate-term home care versus none and long-term home care versus none, changes for different groups in the population. In the second part of the analysis we modelled the log of service intensity and estimated a separate ordinary least squares (OLS) equation for short-term and longterm care as a function of the same set of explanatory variables used in the first part. Since we were estimating the log of service intensity, for each unit increase in the explanatory variable of interest, service intensity changes by 10 β (where β is the parameter estimate associated with the variable). 9

10 Explanatory variables (i) Demographic Characteristics Age was collapsed into six categories: 0-19 years, years, years, years, 75 to 84 years, and 85 years and over. An age-sex interaction term was included in the analyses, to test whether the effect of age differed for males and females. (ii) Co-Morbidity It has been shown that differences in the observed utilization of medical services between socio-economic groups may be attributable to differences in underlying health status [17]. We used the Adjusted Clinical Group Case-Mix System (ACG) developed at Johns' Hopkins University [25] to characterize each individual s level of morbidity. The ACG classification system has been validated in the United States [26], Manitoba, and British Columbia [27] as a means of accounting for and predicting individual health care expenditures. The System uses ICD-9 diagnostic codes from physician billing and hospital discharge records, to characterize individuals using 12 Collapsed Ambulatory Diagnostic Groups (CADGs) 5 which reflect the individual s health status and probable level of health care expenditure, given the degree of illness. Individuals were assigned to as many of the relevant CADG categories as were applicable based on the diagnoses contained in their hospital discharge records and fee for service physician claims data for calendar year (iii) Regional variables In recognition of the possibility that the likelihood and amount of service provision may be different across regions, we included dummy variables denoting the region in which home care services were received. Because some regional differences may be 5 See Appendix 2 for a list of the CADGs. 6 This approach assumes that the effects of each CADG on health status is additive. 10

11 due to differences in the availability of medical resources or regional economic characteristics, we incorporated a rurality index developed by Kralj [28] for Ontario as a means of controlling for variation in health care resources and some social indicators across regions. The index is calculated for Census subdivisions (CSD) 7 and is an attempt to go beyond the dichotomous urban/rural designation which is based solely on population size and density. This is particularly relevant in Ontario, as many small cities, such as Guelph and Windsor, have been classified as under-served in terms of the availability of certain types of medical services. A CSD is allotted points if, relative to the Ontario average, it: i) lacks any of the measured services, ii) iii) iv) experiences extreme weather conditions, residents must travel greater distances to services, lacks educational institutions or an airport. v) has a high unemployment rate. The higher the overall point-score, the more rural the CSD. Scores were transformed so that they range between a low of zero and a high of 100. Appendix 3 reports the formula used to generate the Rurality Index (RIO) and the potential scores and weights associated with each of its components. (iv) Socio-economic Status The population health literature has adopted a broader definition of SES than that used in economics, which tends to focus on income-based measures, such as median 7 The index was also calculated for unorganized areas, aboriginal reserves and settlements and excludes CSDs with a population of less than 500. As a result, there were some missing values for the rurality index (1.3% of sample observations). All subsequent analysis was conducted on those observations for which there were no missing data. 11

12 income. A more recent strand of the population health literature has identified the distribution of income the earnings gap between the rich and poor - as measured, for example, by the Gini coefficient, as a potentially important determinant of health and as a consequence, of health care service utilization [29,30]. Some studies have focused on the impact of non-monetary dimensions of SES. Grossman [31] argued that a more educated person (independent of his/her income level) is better able to make use of health information and can therefore make more efficient use of health care resources, compared to an otherwise identical individual. The Whitehall studies [32,33] purported to show that an individual s rank in the occupational hierarchy had an effect on his/her health status independent of observed differences in health behaviours (i.e. smoking, drinking) and income; those working in non-professional jobs faced a significantly greater risk of premature mortality. SES was measured as an ecologic variable; individuals were assigned the values of the SES indicators associated with the neighbourhood (Forward Sortation Area) in which they lived 8. The assumption underlying this approach is that the neighbourhoods are fairly homogeneous and, as a result, an individual is probably well characterized by the neighbourhood in which he/she lives. While there is a degree of measurement error associated with the accuracy of individuals FSA, as reported in the RPDB, this approach has been validated elsewhere [34]. In an attempt to measure the different aspects of SES identified in the literature, the following FSA-level variables were explored: 8 Ontario is composed of 503 FSAs each corresponding to approximately 7000 dwellings. 12

13 i) proportion of the population living in low-income households i.e. below the low income cut-off 9. (high values = low SES); ii) iii) median household income (high value = high SES); income distribution as measured by the Gini coefficient (high value = low SES); iv) proportion of population aged 15 years or older with at least a college/university degree (high values = high SES); v) proportion of population working in occupations which were not classified as white collar (professional) (high values = low SES). vi) Modified deprivation score [35] The deprivation score was calculated using three measures: proportion of people living in low-income households, male unemployment rate and proportion of males employed in non-professional (blue-collar) occupations. Each of the three measures was standardized (i.e., ((value-average value)/standard deviation) and the average of the three standardized values was calculated. We also considered the proportion of the population who are recent (within 5 years) immigrants to Canada since Ontario receives a large influx of immigrants each year. These newcomers may have greater difficulty accessing services due to linguistic or cultural differences or lack of information about the structure of the health care system. There is also evidence to suggest that recent immigrants, because they are screened for illnesses before entering the country, may have higher than average 9 The low income cut-off is a measure of poverty developed by Statistics Canada, based on the proportion of household income spent on food, clothing and shelter and is adjusted for household size and urban/rural characteristics. 13

14 (compared to long-term immigrants or native born Canadians) health status [36] and may therefore use fewer health care resources. The discussion to follow will focus on the results which were based on low-income, median household income and the deprivation index since these measures of SES were significant in the regression analysis 10. For each investigation (of propensity to receive home care, and of short- and long-term intensity of use), three models were estimated. Each model included age, sex, CCAC, Health Status, the Rurality Index, recent immigrant and one of the SES measures. 4. Results A. Probability of Receiving Home Care Age, sex, CCAC, and health status (H) were significant (p<0.0001) predictors of the probability of home care, no matter what other variables were included in the model. Individuals with co-morbid conditions were more likely to receive home care, a finding which is consistent with that reported in Hall and Coyte [37], based on the National Population Health Survey (NPHS). Females aged 85 years and older were more likely to receive home care than any other group. The probability of receiving home care increases with age (Figure 2). Except for children, males were less likely toreceivehomecarethanwerefemalesinthesameagegroup,andwhenmalesdid receive home care, they were more likely than females in the same age group to receive short-term care. Among adult recipients of home care, the chances that the home care episode was short-term (rather than long-term) decreased with increasing age (Figure 3). Until age 65 for women and age 75 for men, an individual who received home care was more likely to receive short-term care, rather than long-term 10 The significance threshold was set at the 5% level. The Gini coefficient was dropped from the analysis because it was highly correlated (r= 0.9) with median income. 14

15 care (i.e. the odds-ratio was > 1). After that, the odds-ratio was less than 1: individuals receiving home care were more likely to get long-term care. The detailed results for the CCAC, age, sex and health status variables are reported in Appendix 4 (Tables A to C). Figure 2 Probability of Receiving Home Care by Age Group 0.25 Probability of Home Care age group female short-term male short-term female long-term male long-term 15

16 Figure 3 Probability of Short-Term vs. Long-Term Home Care Odds Ratio Age Group females males The coefficient estimates for the rurality index (RIO), recent immigrant and each of the SES variables are reported in Table 1, below. For each variable, we report an overall p-value, which tests the hypothesis that the variable in question was correlated with the probability of receiving home care. More specifically, it tests the null hypothesis that the parameter estimates associated with each of the alternatives (shortterm vs. none, long-term vs. none, intermediate vs. none), were all equal to zero. We also report a p-value for the three individual odds-ratios, testing the hypotheses that the variable was a significant predictor of the particular probability being calculated. No matter how SES is measured, the probability of receiving home care increased with lower SES, once other factors, including age, sex, health status, region and recent immigrant were held constant. Furthermore, SES had a larger effect on the probability of long-term home care than it did on the probability of short-term home care (intermediate home care looked similar to short-term home care). 16

17 The effect of proportion of recent immigrants in the neighborhood was stable across the three types of home care (short-term, intermediate-term, long-term). For each type of home care, the probability of receiving home care decreased with increased immigrant population, and it decreased fairly evenly for all three types of home care. Table 1 Deprivation model Rurality Index (RIO), per 10 point increase in the index p = Recent immigrant, per 1% increase p < Deprivation Index, per 1 unit increase p < Low income model Rurality Index, per 10 point increase in the index p = Recent Immigrant, per 1% increase p < Low income, per 1% increase p < Median income model Rurality Index (RIO), p = Recent Immigrant, per 1% increase p < Median income, per $10,000 increase p < Probability of Receiving Home Care Odds ratios (95% confidence interval) p-value Short-term home care vs. none 0.98 (0.97, 0.99) p = (0.97, 0.99) p < ( ) p < (0.98, 1.01) p = (0.97, 0.98) p < (1.01, 1.02) p < (0.98, 0.99) p < (0.899, 0.935)) p < Long-term home care vs. none 0.99 (0.98, 1.01) p = (0.97, 0.98) p < (1.19, 1.26) p < (1.00, 1.03) p = (0.97, 0.98) p < (1.02, 1.03) p < Intermediate-term home care vs. none 0.97 (0.93, 1.02) p = (0.95, 0.99) p= (1.05, 1.29) p = (0.94, 1.04) p = (0.95, 0.99) p = (1.001, 1.027) p = (0.98, 0.99) p < (0.842, 0.878) p < (0.96, 1.00) p = (0.833, 0.954) p =

18 The rurality index (RIO) was significant when SES was measured using the deprivation index (p-value for RIO = ), and when SES was measured using the proportion of low-income households in the neighbourhood (p-value for the RIO = ), but not when SES was measured using median income (p-value for the RIO = ). However, when SES was measured using the deprivation score, the probability of short-term home care was lower in more rural areas, but when SES was measured using low income, the rurality index was significant only in predicting longterm home care, and the probability of home care was higher in more rural areas. In order to understand why the significance of the rurality index depended on which measure of SES was used, we divided Ontario by the first letter of the postal code, and obtained median values for the various measures of SES (See Table 2). Table 2 Region L ( 905 area) N (southwestern Ontario) K (eastern Ontario) M (Toronto) P (northern Ontario) Median Values of SES and Rurality by Region Deprivation %low Index income Rurality Index (RIO) Median income ($000 s) %blue collar %unemployment %Recent Immigrant % 55 60% 3.9% 2.4% % 43 70% 4.2% 1.4% % 41 58% 6.4% 0.8% % 40 52% 7.7% 13.4% % 41 71% 8.0% 0.3% The so-called 905 area surrounding Ontario (named after the area code for phone numbers in the region) had the lowest deprivation, while Northern Ontario had the most deprivation. Toronto, too, had high deprivation, but it also stood out as having a 18

19 high percentage of households classified as low income. In terms of median income, the 905 area was clearly well off, but Toronto was not significantly different than other areas (living in Toronto, one needed more income to avoid being classified as low income). The deprivation index is composed of three measures: the same low income variable used on its own in the low income model plus the proportion of men in occupations classified as blue collar, and the unemployment rate. While low income and unemployment reflect monetary aspects of SES, blue collar relates to lower social status. Northern Ontario has high deprivation because it has high unemployment, a large proportion of the men employed in blue-collar occupations, and low income. Toronto has high deprivation due to high unemployment and a large number of lowincome people. Southwestern Ontario is low deprivation, despite the high proportion of men with blue-collar occupations, because it has low unemployment and high incomes. The rurality index, or the lack of health care services, is high in Northern Ontario. Eastern and Southern Ontario are more rural than Toronto and the 905 area but much less rural than Northern Ontario; and Toronto and the 905 area are non-rural. The contradictory results for the rurality index in the deprivation model appear to be a function of the degree of correlation between the rurality index (RIO) and the components of the deprivation index. The lack of significance of the rurality index in the median income model seems to indicate that medical rurality is positively correlated with the absolute standard of living in the CCAC. The fact that the probability of receiving long-term home care increased with the rurality index in the 19

20 low income model may suggest that home care services are used as a substitute for institutional long-term care in medically rural areas, if, the lack of institutional beds is correlated with a lack of other resources (i.e. hospitals) as measured by the index. The Role of SES in the Prediction of Short- Versus Long-term Home Care To determine whether the effect of a particular variable on the probability of shortterm home care (versus no home care) was the same as its effect on the probability of long-term care (versus no home care), contrasts were calculated to compare the two odds-ratios. For example, the odds-ratio for receipt of short-term home care was 1.12 for each one-unit increase in the deprivation index (Appendix 4, Table A). The corresponding odds-ratio for receipt of long-term care was The p-value of <0.0001, reported in the last column of the first row in Table 3, indicates that the two odds ratios are significantly different from one another. We observe from Table 3 that no matter how SES was measured, the odds ratio for short-term home care differed significantly from the odds ratio for long-term home care. Examination of the parameter estimates indicates that SES had a larger effect on the probability of receiving long-term home care than on the probability of receiving short-term care. Table 3. Contrasts of Short- and Long-term Model Rurality Index (RIO) Immigration SES Deprivation Model p= p= p< Low income Model p= p= p< Median Income Model p= p= p< Interpretation of the Model Results In order to help interpret the odds-ratios, we report, in Table 4, the range of values for rurality, immigration, and SES observed for the population of Ontario. Using the 20

21 deprivation model as an illustration, the predicted effect on the probability of shortterm home care, of moving from the neighbourhood with the lowest deprivation to the neighbourhood with the highest deprivation, is That is, an individual living in a neighbourhood with the highest deprivation score is approximately three times more likely to receive short-term home care than someone living in a neighbourhood with the lowest deprivation score, all else being constant. Likewise the predicted effect on the probability of short-term home care of moving from the neighbourhood with the 25 th percentile deprivation to the neighbourhood with the 75 th percentile deprivation is ,or1.11. Table 4 Descriptive Statistics for SES and Rurality Range Median 25 th percentile 75 th percentile Low income 0 to 53.4% 15.6% 11.2% 22.9% Deprivation index to Median income 0 to (000 s) Immigration 0 to 29.6% 2.4% 0.7% 8.2% As a way of quantifying the predictive ability of each variable, a series of logistic regressions were performed. The R 2 values are reported in Table 5. Health status was the best single predictor of whether or not someone would receive short-term home care (middle column), with age being the next best predictor. Age was the best predictor of whether or not someone would receive long-term home care, with health status also being a good predictor. Age was also the best predictor of whether the first home care episode would be short- or long-term. Once age was accounted for, adding sex produced very little improvement in any of the models. We observed that rurality, 11 From Table 4 we first calculate the difference in deprivation, which is 9.4. The odds ratio for a oneunit change in deprivation is Thus, the odds ratio for a change of 9.4 in deprivation is ,or

22 SES, and immigrant status played relatively minor roles in terms of predicting an individual s likelihood of receiving home care 12. Table 5 Variable Univariate Analyses from Logistic Regression Adj-R 2 Adj-R 2 Short- vs. Longterm Short-term vs. none Age 7.5% 8.4% 23.9% Age + Sex 8.0% 8.4% 24.2% Age/Sex interaction 8.2% 8.6% 24.5% Health Status 4.3% 10.5% 20.2% (CADGs) CCAC 1.0% 1.0% 1.1% Rurality Index (RIO) 0.0% 0.2% 0.2% Recent Immigrant 0.0% 0.5% 0.2% SES: deprivation low income median income 0.1% 0.3% 0.3% 0.0% 0.0% 0.4% Adj-R 2 Long-term vs. none 0.2% 0.1% 0.9% B. Service Intensity (i) Determinants of Short-term Service Intensity Service intensity for short-term episodes was defined as the number of visits during the episode. 13 During the index episode, 1,600,426 visits were made to 129,755 shortterm care clients. 11.0% of the clients received only a single day of home care, and 27.1% had episodes of a week or less. The average length of a short-term episode was28days(standarddeviation24.5days)andthemedianwas21days. The25 th percentile was 7 days, the 75 th percentile was 44 days. The 90 th percentile was 67 days. The distribution of short-term clients and visits by type of service is presented in Table Likelihood ratio tests were conducted and indicated that all the models fit well. 13 The intermediate category contained 10,021 clients, who received 360,360 visits during their index episodes, and a total of 446,441 visits during the whole of the calendar year. The 25 th percentile episode length was 98 days, the median was 104 days, and the 75 th percentile was 111 days. 22

23 Table 6 Distribution of Short-Term Care Clients and Visits Visits Care Recipients Nursing 71.06% 66.32% Personal Support and/or 12.52% 15.00% Homemaking Physiotherapy and/or 13.31% 34.18% Occupational Therapy Other Services (e.g. Social Work, Speech Language Pathology etc.) 3.12% 8.91% In terms of service mix, Table 7 shows the proportion of people who received at least one visit of each type. Almost half the short-term episodes involved only nursing, and one-fifth involved only physio- /occupational therapy. Table 7 Nursing Personal Support and/or Homemaking Service Mix for Short-term Clients Physiotherapy Other and/or Services Occupational Therapy % (of 129,755 people) X 4.82% X 21.58% X X 0.84% X 3.28% X X 0.08% X X 2.92% X X X 0.15% X 50.55% X X 1.94% X X 4.98% X X X 0.50% X X 4.94% X X X 0.40% X X X 2.83% X X X X 0.38% In terms of service intensity, the median number of short-term visits was 6; the 25 th and 75 th percentiles were 3 and 15 visits, respectively. The 10 th percentile was 1 visit and the 90 th percentile was 29 visits. Figure 4 depicts the distribution of short-term clients by number of visits. 23

24 Figure 4 Distribution Of Short-Term Clients by Number of Visits Percentage of Clients (%) 14.00% 12.00% 10.00% 8.00% 6.00% 4.00% 2.00% 0.00% ' Number of Visits Empirical Model Results Age, sex, CCAC and health status (H) were significant (p< ) predictors of short-term service intensity in all three SES models 14. Their coefficient estimates are reported in Appendix 5. Among short-term home care recipients, children received the fewest services, and boys received fewer services than girls. The most elderly (aged 85 years and older) received the next fewest number of services, followed by adults aged 75 to 84 years. Generally, males received more services than females of the same age. Health status as measured by the vector of co-morbidity groups (CADGs) was significant. The results from adding rurality, immigration, and each of the three SES measures to the base model (age, sex, CCAC, Health status) are reported in Table 8. The first row indicates that recent immigration was not a significant predictor of short-term service intensity in any of the models. In contrast, the rurality index was significant in all three models. Short term service intensity decreased with increasing rurality. In the 14 In order to remove extreme outliers from the data, short-term service intensity was capped at 100 visits, which retained 99.55% of the observations and long-term care service intensity was capped at 730 visits (an annualized rate of 2 visits per day), which retained 98.87% of the observations. Both short- and long-term service intensity were skewed, but their logs were reasonably close to a normal distribution. 24

25 model with SES measured using the deprivation index, short-term service intensity increased by 2% for each unit increase in deprivation (p = ) and decreased by 0.7% for each 10-point increase in the rurality index. In the low-income model, SES was not significant. In the median income model, short-term service intensity decreased by 0.9% for each additional $10,000 in median income (p = ) and decreased by 0.6% for each 10-point increase in the rurality index. Thus, in both models, service intensity increased with lower SES. The R 2 for the three models ranged from 9.45% to 9.97%, which is not unexpected given the cross-section analysis and the large sample size. Table 8 Rurality Index (RIO) Recent immigration Results of Intensity Equation for Short-term Home Care Deprivation model Low income model Median income model ($000 s) p = β = ( ) p = β= (0.0004) SES p = β = (0.0023) p = Β= ( ) p = β= (0.0005) p = β= ( ) p = Β= ( ) p = β= (0.0004) p = β= ( ) R % 9.45% 9.97% SES*H p= p< p< R % 10.00% 10.00% In model containing only the main effects of immigration, rurality, and SES, it appeared that immigration did not predict short-term service intensity, but rurality did. The effect of rurality was negative - people living in areas which are more medically rural received less service, on average. Low SES, in so far as it is measured by the deprivation index or median income, was associated with more service, on average. The SES-Health Status Connection To investigate whether SES affected service intensity differentially depending on health, we introduced an interaction between each SES measure and the co-morbidity categories (CADGs) into the regression models. The significance of the SES-health 25

26 interaction in all three models indicated that the role of SES in predicting service intensity depended on health status 15. If SES was measured using the proportion of people living in low-income households, there was a significant interaction between SES and chronic medical stable (p=0.0016), psychosocial (p=0.0004), prevention/administration (p=0.0097), and pregnancy (p=0.0270). In the first three cases, the direction of the interaction was such that someone living in a lower SES neighbourhood received fewer services than would be predicted by illness or SES alone. For pregnancy, the effect was in the opposite direction. When SES was measured using median household income, there were, again, significant interactions between SES and illness for psychosocial illness (p=0.0009), prevention/administration (p=0.0027), and pregnancy (p=0.0005). For a given SES, the pregnancy CADG carried the largest penalty, in terms of number of short-term home care visits, of the three CADGs. But visits paid to pregnant women increased most steeply as SES increased. The interaction parameter between health (H) and SES, when using deprivation as the measure, was only significant for psychosocial illness and prevention/administration, and was associated with decreased service use on the part of the person with the condition, relative to someone living in a less deprived area. The direction of the interaction effects between SES and illness, regardless of the SES measure used, were such as to reduce the amount of service for people living in low SES neighbourhoods. In comparison with a neighbour without illness, a person in a lower SES neighbourhood fared worse than a person in a higher SES neighbourhood. While the person in the lower SES neighbourhood might still receive more visits than their neighbour without illness, the incremental gain due to illness was less that that obtained by their counterpart in the high SES neighbourhood. It is important to note, however, that the main predictors in the short-term intensity modelwereage,sex,age/sexinteraction(r 2 = 7.1%). Adding health status to the model increased the model R 2 to 7.7%. Rurality, SES, the SES-Health interactions 15 The coefficients for the SES*Health interactions are reported in Appendix 6 (Tables A to C). 26

27 and immigrant status added little in the way of predictive ability. In fact the model R 2 increased to, at most, 10.2% (for a model containing CCAC, immigration, SES, and interactions between health status and both SES and CCAC). (ii) Long-term Home Care Service intensity for long-term care episodes was defined as the annualized number of visits i.e. the number of visits the client would have received had the visits continued with the same intensity for a complete year. If the client received services for at least half of 1998, then only the 1998 visits were used to calculate service intensity. Otherwise, services from 1997 or 1999, if any, were included in order to obtain a robust estimate of service intensity 16. The mean length of a home care episode used to calculate long-term service intensity was 278 days, and the median was 296 days. The 25 th percentile was 199 days and the 75 th percentile was 358 days. Table 9 presents the distribution of long-term care visits and clients across types of service, and Table 10 shows the service mix. Of those who received long-term service, the vast majority of clients received some homemaking services and over half received some nursing care. Most long-term care visits were either to provide home making services or nursing care. Table 9 Distribution of Long-term Care Clients and Visits Type of service Visits Clients Nursing 35.95% 57.54% Personal Support and/or 57.41% 75.43% Homemaking Physiotherapy and/or 4.35% 40.35% Occupational Therapy Other Services 2.00% 18.74% 16 In order to avoid edge effects for long-term care clients who were receiving visits once per week or less, if the annualized service intensity was 56 or less (indicating a visit once per week for 18 weeks), the first day of home care was moved back to the nearest Monday and the last day of home care was moved forward to the next Sunday, and the service intensity was recalculated. 27

28 Table 10 Nursing Service Mix for Long-term Care Clients Physiotherapy Other and/or Services Occupational Therapy Personal Support and/or Homemaking % (of 126,991 Clients) X 2.97% X 4.55% X X 1.74% X 21.55% X X 0.90% X X 9.28% X X X 1.47% X 9.71% X X 1.81% X X 2.74% X X X 1.06% X X 19.21% X X X 3.51% X X X 14.22% X X X X 5.29% Less than 10% of long-term care clients received nursing services only, almost 20% received both nursing and home making services and another 19.5% received nursing, home making and either occupational or physio- therapy. The average service intensity was 138 with a standard deviation 163. The range was from 12.4 (one visit per month) to 5,010 (the 99 th percentile was 749 visits or just over 2 per day). The median was 90 and the 25 th and 75 th percentiles were 52 (one visit per week and 159 (one visit every other day). Empirical Model Results We modeled the log of service intensity in long-term homecare, where service intensity was the annualized number of visits, as a function of age, sex, the age-sex interaction, health status, rurality, CCAC, immigration, SES and the SES-health interaction. Age, sex, CCAC and health status were significant (p<0.0001). Females, other than children, received fewer services than males in the same age group. 28

29 Relative to males aged 85 years and over, males aged received more, males aged 0-19 years received less and males aged years received the same amount of long-term services. The coefficient estimates for age, sex, health status and CCAC are reported in Appendix 6 (Tables A to C). Table 11 reports the estimates and p- values for the remaining variables. Table 11 Rurality Index (RIO) Recent immigration Results of Intensity Equation for Long-Term Home Care Deprivation model Low income model Median income model ($000 s) p < β = ( ) p < β= (0.0003) SES p < β = (0.0016) p < β= ( ) p < β= (0.0003) p = β= ( ) p < β= ( ) p = β= ( ) p = β= (0.0001) R % 16.6% 16.6% SES*H p= p= p= R % 16.6% 16.6% Unlike short-term intensity, immigrant status was a significant predictor of long-term service intensity. People living in neighbourhoods with a high proportion of immigrants received less service, on average. Rurality was also a significant predictor, and SES was significant when it was measured using low-income or the deprivation index, but not when it was measured as median income. As with shortterm service intensity, service was lower in more rural areas and people living in lower SES areas received on average, more service. In contrast to the model for short-term intensity, the role of SES did not appear to depend on illness when SES was measured by low-income or the deprivation index. The SES-health interaction was significant in the median income model in relation to 29

30 unstable chronic medical and psychosocial conditions but the differences between low and high SES neighborhoods were very small. Age, sex, and their interaction were the best predictor of long-term service intensity (R 2 =8.4%), followed by CCAC (R 2 =7.3%). While rurality was negatively correlated with long-term service intensity, it is not as important a predictor (R 2 =0.1%) as the region identifier (CCAC). Lower SES was associated with greater service intensity and recent immigrant with less service intensity but neither variable was an important predictor, with R 2 s of less than 0.01% for both. The full model had an R 2 of 16.8%. 5. Discussion In this paper we have explored the determinants of access to publicly-funded home care services. The two-part approach applied in the analysis was justified, since some variables had a differential effect on propensity to receive care and intensity of care. The distinction between short- and long-term care was also appropriate, given that the importance and strength of some variables were found to differ across the types of care. Whether we considered propensity and intensity of care for long-term episodes, we found that living in a neighbourhood with a high proportion of recent immigrants was associated with reduced access. We did not find a relationship between recent immigrant and short-term home care service. Long-term home care service is composed largely of home making services whereas short-term home care involves nursing care for the vast majority (66.52%) of clients. Perhaps this is an indication that recent immigrants are more likely to rely on and receive family/ informal caregiver support for long-term care due to cultural norms or linguistic/cultural 30

31 barriers to service. As mentioned earlier, there is evidence to suggest that recent immigrants are generally healthier than long-term immigrants and native-born residents of Canada [36]. However, in all the regression equations estimated, health status, age and sex were significant which indicates that recent immigrants are less likely to receive service and use less service than individuals of comparable age, sex and morbidity profile. In other words, the negative association between recent immigrant and access is independent of need and may therefore be indicative of barriers to access for that segment of the Ontario population. While our data set does not permit us to isolate the specific causes of the negative relationship between recent immigrant and access to home care services, barriers to access appear to exist, and merit further investigation in future research. Medical service availability, as measured by the rurality index (RIO), produced mixed results in terms of determining likelihood of service receipt but was clearly negatively related to service intensity. Lack of medical resources appeared to be more important in the determination of the amount of service received as opposed to determining whether or not one received service in the first place. There were differences in terms of the determinants of access across the regions that were not explained by the RIO, as evidenced by the consistent significance of the regional variable (CCAC). This may be linked to differences in referral patterns across regions [38] or simply reflect the fact that there are aspects of community-based care which do not fall into the domain of medical resources such as the availability of homemaking services or that things like distance to tertiary care facilities may not be meaningful in the context of home care, where the personnel are deployed locally. In terms of socio-economic status (SES) we found that the relationship to access both in terms of propensity and intensity is consistent with the stated intent of Medicare. 31