High health care utilization and costs associated with lower socio-economic status: results from a linked dataset.
Lemstra, Mark ; Mackenbach, Johan ; Neudorf, Cory 等
Persistent socio-economic inequalities are a costly economic
deadweight in terms of higher expenditures on health care, income
assistance, social services, correctional services and lost tax revenue.
(1) Two reports from Canada and the European Union have concluded that
disparities in socio-economic status account for 20% of total health
care resources. (2,3) The concern, however, with using estimates of
self-report health care utilization through telephone surveys is that
the recall of 'number of contacts' with health care services
does not demonstrate good validity. (4)
The primary purpose of this paper was to use a linked dataset to
compare actual health care utilization rates, high health care
utilization patterns and overall costs between income groups in
Saskatoon, Canada. The second purpose was to use regression analysis to
determine which covariates were independently associated with high
health care utilization after controlling for disease prevalence.
METHODS
The Canadian Community Health Survey (CCHS) is administered by
Statistics Canada with the central objective of collecting self-report
health-related data at the level of health regions. (5) The CCHS
consists of cross-sectional surveys in 2000/01, 2003 and 2005. The
methodology of the CCHS has been documented in detail previously. (5)
Income status was based on the Low Income Cut-Off (LICO) developed
by Statistics Canada. (6) Cut-off points are adjusted for family size,
population of area of residence, urban/rural differences and consumer
price index. For example, a single adult in Saskatoon with an income
less than $18,000, and a family of four with an income of less than
$33,000, fall below the LICO and are therefore classified as low-income
earners. High-income earners were those who made more than $80,000 per
year. The remainder were classified as middle-income earners.
The review of health care utilization included hospitals (including
emergency room and day surgeries), physicians (including specialists)
and prescription medications. Saskatchewan has universal health coverage
for all residents, with a centralized administrative database that
collects information on all hospital separations, physician visits and
medication usage. The positive predictive value of a primary diagnosis
from hospital administrative data in Saskatchewan (for stroke) is 90%.
(7) At the time of the CCHS survey, each respondent was asked to consent
to having their self-report survey information linked with their
provincial health records. The respondents' name and Saskatchewan
Health Services number were collected at the time of interview.
Saskatchewan Health completed the data linkage and provided the
de-identified dataset to the researchers. The overall counts of
utilization were collected for the year in which the survey was
completed (i.e., health care utilization for 2005 if CCHS survey was
completed in 2005) and then merged into one larger sample in order to
increase precision of the estimates.
Hospital costs based on ICD-10 separation codes were calculated by
Strategic Health Information Planning Services and Finance of the
Saskatoon Health Region. The costs provided were direct departmental
costs and do not include overhead (i.e., administration) or support
costs (i.e., lab or medical imaging). The cost of physician visits was
provided by Saskatchewan Health. For medications, the average cost per
drug within each class was calculated using the Saskatchewan Drug
Formulary.
The first comparison was one-year incidence counts of health care
utilization for hospitals, physicians and medications by income group.
The second comparison was to review high health care utilization of
hospitals, physicians and medications by income group by calculating the
upper 20th percentile of utilization overall. Once determined, each
income group was reviewed independently to determine what percentage of
users fell above or below the upper 20th percentile of overall users of
hospitals, physicians and medications. The third comparison was the cost
of hospitals, physicians and medications by income group in order to
determine mean cost per user of hospitals, physicians or medications.
Three separate binary logistic regression models were built to
describe the relationship between the outcome variables of a) high
utilization of hospitals, physicians or medications (upper 20th
percentile) in comparison to b) lower 80th percentile usage of health
care. The covariates within the regression models included the
demographics of age (12-39, 40-59 and 60 and above), gender and cultural
status (Caucasian and Aboriginal). Health outcomes included self-report
health (good/fair/poor compared to excellent/ very good) and self-report
heart disease prevalence, diabetes prevalence and lifetime suicide
ideation. Socio-economic status included family income (described above)
and individual education (less than high school grad, high school grad
and university education). Disease intermediaries included blood
pressure (yes/no) and body mass index (obese/overweight versus normal/
underweight). Behaviours included physical activity (regular/
occasional/infrequent), smoking (daily versus other), alcohol usage (5
or more drinks at a time at least once per week or not in the past year)
and consumption of fruits and vegetables (5 servings per day or not).
Life stress was measured by current amount of stress in daily life.
A hierarchal well-formulated step-wise modeling approach was used
instead of a computer-generated stepwise algorithm.8 In the final model,
the unadjusted effect of each covariate was determined and then entered
one step at a time based on changes in the -2 log likelihood and the
Wald test. (9)
Ethics approval was obtained from the University of Saskatchewan
Behavioural Research Ethics Board.
RESULTS
Over three cycles in 2001, 2003 and 2005, 4,103 residents of SHR
were asked to complete the CCHS with 3,867 agreeing to participate
(94.2%) and complete data available on 3,688 participants (89.9%). Of
these 3,688 participants, 3,433 agreed to the data linkage with health
records (83.7% overall). By individual cycle, the sample sizes were
1,174, 1,082 and 1,177. With all three cycles merged, the mean age was
46.3 (SD 20.32), females represented 55.2% of the sample and Caucasians
represented 73.4% of the sample. In comparison to 2001 census data for
SHR, the sample had a statistically significant difference in age
(p=0.01) but not gender or cultural status. Based on the definitions of
income discussed above, there were 785 low-income, 1,793 middle-income
and 855 high-income participants; which is also consistent with the 2001
census.
At the cross-tabulation level, low-income residents were 27-33%
more likely to be hospitalized and 36-45% more likely to receive a
medication but were 5-7% less likely to visit a physician over a
one-year period in comparison to middle- and high-income earners (Table
1).
The upper 20th percentile for health care utilization over one year
was determined to be greater than 2 hospital visits, 32 physician visits
and 29 medications. In comparison to middle-income residents, low-income
residents had 56% more high users of hospitals, 166% more high users of
physicians and 90% more high users of medications. In comparison to
high-income residents, low-income residents had 28% more high users of
hospitals, 226% more high users of physicians and 73% more high users of
medications (Table 1).
The average cost of hospitals, physicians and medications over a
one-year period for low-income residents who accessed health care was
$7186. The average costs for middle- and high-income residents who
accessed health care were $5266 and $5478. Low-income residents who used
health care had 24-27% higher costs in comparison to middle- and
high-income residents (Table 2). If we calculate health care costs for
all low-income residents combined (regardless of access to health care),
the average cost for all low-income residents is $4489 in comparison to
$2964 and $2923 for all middle- and high-income residents; which equates
to 34-35% higher health care costs overall.
After cross-tabulation, it was found that low-income residents have
higher prevalence rates of high blood pressure, heart disease and
diabetes (Table 3a). After stratification, those with higher disease
prevalence were more likely to have higher health care utilization.
However, in most cases, low-income residents were still more likely to
have high health care utilization even after controlling for disease
prevalence (Table 3b).
After multivariate logistic regression, high hospital utilization
was independently associated with the covariates of heart disease
prevalence, lower self-report health and higher age. High physician
utilization was independently associated with the covariates of heart
disease prevalence, lower self-report health, higher age and low income.
High medication utilization was independently associated with the
covariates of high blood pressure, diabetes, heart disease, lower
self-report health and higher age. The results are presented in Table 4.
DISCUSSION
In 2005, Saskatchewan residents paid $528,759,380 for physician
services, $1,875,752,000 for health regions to provide mainly hospital
services and $184,020,000 for prescription medications for a sum of
$2,588,531,380 out of a total health care budget of $2,990,625,000 for a
population of 1,020,966.10,11 In other words, every Saskatchewan
resident consumed an average of $2929 health care costs in 2005. This is
very similar to the $2964 average health care costs for all
middle-income earners and $2923 average health care costs for all
high-income earners calculated in this study.
According to the 2001 Canadian census, low-income earners in
Saskatoon represented 17.1% of the entire population and, as such,
should consume $511,396,875 of health care costs.12 However, this study
demonstrates that low-income residents consume 35% more costs overall
than anticipated in comparison to middle- and high-income residents. In
other words, low-income residents in Saskatchewan consume an extra
$178,988,906 in health care costs than if they were middle income.
The cross-tabulation finding that on average low-income residents
access physicians less often, and hospitals more often, in comparison to
other income groups has been demonstrated previously. (2,13) Low-income
groups have more complex needs while at the same time have less
continuous and comprehensive health care; which results in more usage of
expensive services like hospitals. (3) In fact, the highest-income
groups are the most likely to receive optimal primary care and obtain
more referrals to specialists, which widens health disparities. (2)
In our study, low-income status was also associated with high
health care utilization (upper 20th percentile) of hospitals, physicians
and medications at the cross-tabulation level. However, after
multivariate adjustment, low-income status had a reduced association
with high health care utilization after controlling for disease
prevalence. After controlling for the prevalence of high blood pressure,
heart disease, diabetes, low self-report health and age, the odds of
high health care utilization dropped for low-income residents by 73% for
hospitals, 60% for physicians and 58% for medications. The results
suggest that most (but not all) of the disparity in high health care
utilization for lower-income residents is associated with higher disease
prevalence, and not merely a difference in utilization behaviour. This
finding is consistent with the literature. (14,15) In these studies, the
increased use of family physician and hospital services in lower
socio-economic groups corresponded to higher needs resulting from poor
health. (14,15)
The finding that high health care utilization is associated with
higher age and lower self-report health is supported by a linked study
from Nova Scotia. (16) Another linked study from Manitoba found
high-cost users of medications were more likely to be low income, older
in age and more likely to have a chronic condition; all of which are
consistent with our results. (17)
The dataset is believed to be valid. First, the overall
participation rate and consent to the data linkage was 83.7% in the
sample with only a slight bias in age. Second, the utilization rates are
believed to be accurate. For example, 84.6% of middle-income respondents
within the sample visited a physician within one year while the annual
statistical report for 2005 states that 83.6% of the Saskatchewan
population accessed a physician within that year. (10) Third, the health
care cost information from the sample is very similar to the costs from
the annual statistical report presented above.
The limitation of the study is that it is cross-sectional, and not
prospective, and as such cause and effect cannot be determined.
The Health Disparity Task Groups of the Federal/Provincial Advisory
Committee on Population Health and Health Security concluded that the
most appropriate and effective way to improve overall population health
status in Canada is to improve the health of those in lower
socio-economic groups. (2) The results from this study demonstrate that
residents from lower socio-economic status are responsible for
disproportionate usage of hospitals, physicians and medications; due
mainly to differences in disease prevalence.
Received: August 25, 2008
Accepted: February 27, 2009
REFERENCES
(1.) Hay DI. Economic arguments for action on the social
determinants of health. Ottawa, ON: Public Health Agency of Canada,
2006.
(2.) Health Disparity Task Groups of the
Federal/Provincial/Territorial Advisory Committee on Population Health
and Health Security. Reducing health disparities--role of the health
sector: discussion paper. Ottawa: Health Disparity Task Groups of the
Federal/Provincial/Territorial Advisory Committee on Population Health
and Health Security, 2004.
(3.) Mackenbach JP, Meerding WJ, Kunst AE. Economic implications of
socioeconomic inequalities in health in the European Union. European
Commission, 2007.
(4.) Statistics Canada. Measurement of healthcare utilization in
Canada: Agreement between surveys and administrative records.
Proceedings of Statistics Canada Symposium 2003: Challenges in survey
taking for the next decade. Ottawa: Statistics Canada, 2003.
(5.) Statistics Canada. Canadian Community Health
Survey--Methodological overview. Health Rep 2002;13(3).
(6.) Statistics Canada. Income Research Paper Series: Low Income
Cutoffs from 1994-2003 and Low Income Measures from 1992-2001.
(Catalogue no. 75F0002MIE--no. 0002). Ottawa: Statistics Canada, 2004.
(7.) Liu L, Reeder B, Shuaib A, Mazagri R. Validity of stroke
diagnosis on hospital discharge records in Saskatchewan, Canada:
Implications for stroke surveillance. Cerebrovascular Dis 1999;9:224-30.
(8.) Rothman KJ, Greenland S. Modern Epidemiology, 2nd ed.
Philadelphia, PA: Lippincott Williams and Wilkins, 1998.
(9.) Hosmer DW, Lemeshow S. Applied Logistic Regression. New York,
NY: Wiley, 1989.
(10.) Government of Saskatchewan. 2005-2006 Annual Statistical
Report. Regina, SK: Saskatchewan Health, 2006.
(11.) Government of Saskatchewan. Public Accounts 2005-06 Volume 2:
Details of Revenue and Expenses. Regina, SK: Government of Saskatchewan,
2006.
(12.) Statistics Canada. CANSIM Table 109-0200, 2001.
(13.) Nabalamba A, Millar WJ. Going to the doctor. Health Reports
2007;18:23-35.
(14.) Brownwell M, Lix L, Okechukwu E, Derksen S, De Haney S, Bond
R, et al. Why is the health status of some Manitobans not improving? The
widening gap in the health status of Manitobans. Report. Manitoba Centre
for Health Policy, 2003.
(15.) Veugelers PJ, Yip AM. Socioeconomic disparities in health
care use: Does universal coverage reduce inequalities in health? J
Epidemiol Community Health 2003;57:424-28.
(16.) Yip AM, Kephart G, Rockwood K. Linkage of the Canadian Study
of Health and Aging to provincial administrative health care databases
in Nova Scotia. Int Psychogeriatrics 2001;13:147-58.
(17.) Kozyrskyj A, Lix L, Dahl M, Soodeen R. High-cost users of
pharmaceuticals: Who are they? Manitoba Centre for Health Policy,
University of Manitoba, 2005.
Mark Lemstra, PhD, [1] Johan Mackenbach, MD, PhD, [2] Cory Neudorf,
MD, MHSc, FRCPC, [1] Ushasri Nannapaneni, MA [1]
Author Affiliations
[1.] Saskatoon Health Region, Saskatoon, SK
[2.] Erasmus University, Netherlands
Correspondence and reprint requests: Mark Lemstra, Saskatoon Health
Region, 101-310 Idylwyld Drive North, Saskatoon, SK, S7L 0Z2, Tel:
306-230-3911, E-mail: marklemstra@shaw.ca
Acknowledgements: The research was supported by a grant from the
Canadian Institutes for Health Research. The authors thank Terry Dunlop,
Lynne Warren, Tanis Kershaw and Ceal Tournier for their assistance, and
Saskatchewan Health and Statistics Canada for the provision of data.
Table 1. Overall Health Care Utilization in One Year--Including Rates
of High Utilization
Health Care Utilization Low Income Middle Income High Income
in One Year (n) (%) (n) (%) (n) (%)
# of people who 387 (49.2) 692 (38.5) 315 (36.8)
attended hospital
# of people who 618 (78.7) 1517 (84.6) 707 (82.6)
visited MD
# of people who 415 (52.8) 696 (38.8) 310 (36.2)
received RX
High health care utilization Low Income Middle Income High Income
in one year* (n) (%) (n) (%) (n) (%)
# of people who 92 (23.8) 105 (15.2) 58 (18.5)
attended hospital
# of people who 169 (27.4) 298 (10.3) 126 (8.4)
visited MD
# of people 123 (29.7) 108 (15.6) 57 (17.1)
who received RX
Health Care Utilization Low vs. Middle Income Low vs. High Income
in One Year RR (95% CI) RR (95% CI)
# of people who 1.27 (1.16-1.25) 1.33 (1.19-1.49)
attended hospital
# of people who 0.93 (0.89-0.97) 0.95 (0.91-1.01)
visited MD
# of people who 1.36 (1.25-1.49) 1.45 (1.30-1.62)
received RX
High health care utilization Low vs. Middle Income Low vs. High Income
in one year* RR (95% CI) RR (95% CI)
# of people who 1.56 (1.21-2.00) 1.28 (0.96-1.72)
attended hospital
# of people who 2.66 (2.26-3.13) 3.26 (2.66-4.00)
visited MD
# of people 1.90 (1.51-2.39) 1.73 (1.31-2.28)
who received RX
Total Sample size (N): 3433
Sample by Income status (n): Low Income = 785;
Middle Income = 1793; High Income = 855
* > 80th percentile for the overall group
Table 2. Total Costs of Health Care Utilization by Income Group
Mean Costs for Mean Costs for Mean Costs for
Low-income All Low-income Middle-income
Users of Health People [dagger] Users of
Care * Health Care *
Hospital $1,208.22 $594.40 $929.98
MD $2,852.24 $2,244.70 $2,016.07
RX $3,125.43 $1,650.20 $2,319.92
Total Cost $7,185.89 $4,489.30 $5,265.97
Mean Costs for Mean Costs for Mean Costs for
All Middle-income High-income All High-income
People [dagger] Users of People [dagger]
Health Care *
Hospital $358.00 $973.11 $358.10
MD $1,705.60 $2,013.59 $1,163.20
RX $900.10 $2,491.16 $901.80
Total Cost $2,963.70 $5,477.86 $2,923.10
* Average cost per user of health care per income group
([dagger]) Average cost per person (regardless of actual utilization)
per income group
Table 3a. Prevalence of Self-report Health Outcomes by
Income Group
Disease Outcome Low Income High Income P-Value
n (%) n (%)
Has high blood pressure 163 (20.8) 115 (13.5) 0.000
Has heart disease 75 (9.6) 32 (3.7) 0.000
Has diabetes 59 (7.5) 37 (4.3) 0.018
Table 3b. Stratified Analysis for High Hospital, Physician and
Medication Use by Disease Outcome and Income Group
Variable High Users + High Users + P - Value
Low Income High Income
Hospital use % %
Has high blood pressure 22.1 20.0 0.000
Does not have high 8.8 4.7
blood pressure
Has heart disease 28.0 34.4 0.000
Does not have heart 9.9 5.7
disease
Has diabetes 25.4 16.2 0.002
Does not have diabetes 10.5 6.2
Physician use
Has high blood pressure 43.5 37.1 0.000
Does not have high 23.4 13.0
blood pressure
Has heart disease 57.1 48.3 0.000
Does not have heart 24.5 15.1
disease
Has diabetes 44.6 45.7 0.003
Does not have diabetes 26.2 14.9
Medication use
Has high blood pressure 43.5 37.1 0.000
Does not have high 23.4 13.0
blood pressure
Has heart disease 57.1 48.3 0.000
Does not have heart 24.5 15.1
disease
Has diabetes 44.6 45.7 0.003
Does not have diabetes 26.2 14.9
Table 4. Crude and Adjusted Estimates for High Utilization
of Hospitals, Physicians and Medications
Variables Crude OR Adjusted 95% CI P-Value
OR
A. Covariates of high
hospital use
Has high blood pressure 3.79 1.20 0.79-1.81 0.377
Has diabetes 2.77 1.20 0.69-2.09 0.514
Has heart disease 6.38 1.66 1.02-2.72 0.041
Good/fair/poor 4.08 2.60 1.68-4.04 0.000
self-report health
Age 60 and above 10.94 7.81 4.31-14.17 0.000
Age 40-59 2.40 2.98 1.59-5.58 0.001
Low personal income 1.80 1.07 0.73-1.56 0.726
B. Covariates of high
physician use
Has high blood pressure 3.19 1.37 0.97-1.91 0.066
Has diabetes 3.27 1.44 0.90-2.30 0.127
Has heart disease 6.23 1.95 1.24-3.07 0.004
Good/fair/poor 3.04 2.14 1.60-2.86 0.000
self-report health
Age 60 and above 5.50 3.29 2.27-4.77 0.000
Age 40-59 1.90 2.15 1.50-3.09 0.000
Low personal income 1.96 1.36 1.03-1.80 0.027
C. Covariates of high
medication use
Has high blood pressure 6.18 2.87 1.90-4.35 0.000
Has diabetes 6.40 4.27 2.46-7.41 0.000
Has heart disease 5.04 2.73 1.63-4.58 0.000
Good/fair/poor 3.52 2.59 1.61-4.17 0.000
self-report health
Age 60 and above 14.23 5.04 2.15-11.80 0.000
Age 40-59 4.39 2.67 1.10-6.44 0.029
Low personal income 1.87 1.29 0.85-1.99 0.235
Reference categories for independent variables--Blood pressure--No;
Diabetes--No; Heart disease--No; Self-report health--Excellent/very
good; Age--12-39 Yrs; Income--[greater than or equal to] 80,000
A. High hospital use: [R.sup.2] = .217; Goodness-of-fit = .234
B. High physician use: [R.sup.2] = .297; Goodness-of-fit = .438;
C. High medication use: [R.sup.2] = .365; Goodness-of-fit = .640