Volume 25
Number 1
2004

[Table of Contents]

Refining the measurement of the
economic burden of chronic diseases
in Canada

John Rapoport, Philip Jacobs, Neil R Bell and Scott Klarenbach

Abstract

This article presents an analysis of the economic burden of a number of chronic diseases in Canada. In the analysis, we adjusted our measure of utilization of physician and hospital services for co-existing chronic diseases, which we found to be widely prevalent and to have an impact on resource use. Using data from the 1999 National Population Health Survey, we developed resource use rankings for several chronic conditions and decomposed these measures into prevalence and per-person utilization components. Our results indicate that, for the diseases with the greatest impact, resource use measures are driven more by disease prevalence than intensity of resource use. The diseases with the highest overall degree of resource use are back pain, arthritis or rheumatism, high blood pressure and migraines for people under 60; and arthritis or rheumatism and high blood pressure for people over 60. Our methods can be used to forecast the overall relative impact of resource use due to disease prevalence and per-person resource use intensity for various conditions.

Key words: chronic disease; economics; utilization

Introduction

Numerous studies in the published and gray literature* have reported on the economic burden of chronic conditions. Several of these allowed for a comparison between different conditions,¹ but most authors focus on one specific chronic condition and confine their focus to the services that are particular to that disease.^2-5 Few of these economic burden studies identify a cost per patient,⁶ which is important if the estimates are to be used for projecting expenditures or for assessing the impact of interventions (i.e., the usual purposes given for conducting these studies). Furthermore, although guidelines relating to disease costing have been available for a long time,^7,8 investigators do not often use common methods or data sources. Most importantly, they seldom adhere to a concept that incorporates comorbid disease. Not accounting for the additional or attributable effect of comorbidities on utilization and cost may lead to bias in estimates of resource utilization.

This study takes a different approach to estimating chronic disease burden: we look at person-level data from a nationwide population survey and, using a common metric, examine the relation between chronic disease and utilization of physician and hospital services, adjusting for comorbidities. We use the National Population Health Survey (NPHS), a national, population-based survey that provides information on the presence of a number of different chronic conditions and the characteristics of individuals with and without these diseases.

Method

All data in the analysis were obtained from the NPHS, a general health survey conducted by Statistics Canada in 1998-1999. We used the general health component of the survey, which included 17,244 individuals. The NPHS asks a series of questions on self-reported chronic disease, defined as conditions that have lasted or are expected to last six months or more and that were diagnosed by a health professional. The following chronic diseases were investigated in our analysis: asthma, arthritis or rheumatism, back problems (excluding arthritis), high blood pressure, migraine headaches, chronic bronchitis or emphysema, sinusitis, diabetes, epilepsy, heart disease, cancer, stomach or intestinal ulcers, effects of a stroke, urinary incontinence, bowel disorder such as Crohn's disease or colitis, Alzheimer's disease or any other dementia, cataracts, glaucoma, and thyroid condition. We created dummy variables that indicated the presence or absence of each of these conditions. We also created a separate variable that indicated the total number of chronic diseases reported by each respondent.

The utilization of physician and hospital services was measured by the number of physician consultations per year and the number of nights spent as a hospital patient per year. In the NPHS public use data file, physician visits by respondents with over 30 encounters per year are combined in an open-ended category; we assigned a value of 31 to these. The same was done with hospital nights, which were also reported in the NPHS data file as an open- ended upper category of above 30. The following additional demographic variables were included as control variables: age, sex, household income, and education as a dummy variable indicating post-secondary educational status.

Analysis was confined to people over age 20. Descriptive statistics on utilization were calculated within four age strata: 20-39, 40-59, 60-79 and over 80. Chronic disease prevalence and regression analyses were performed separately for people under age 60 and people aged 60 and over.

Four utilization variables were used as dependent variables. Physician services were measured by the number of visits per year. Because the highest category for this variable was open-ended and because very high users can have a disproportionate effect on overall utilization, an additional dependent variable, a dummy variable indicating more than 12 annual visits, was also used. Similarly, hospital utilization was captured by means of a dummy variable indicating any hospitalization during the year, as well as by a dependent variable representing the number of nights in the hospital.

For each dependent variable two regressions were performed. The first included as independent variables the demographic control variables and the number of chronic diseases. In the second regression the number of chronic diseases was replaced by the group of dummy variables for the specific chronic diseases. Linear regression was used for the continuous dependent variables, and logistic regression was performed for the dummy dependent variables. Observations were weighted using sampling weights from the NPHS data file. The regression with nights in the hospital as a dependent variable was computed only for those patients who reported a hospitalization.

A summary measure of resource use, by condition, was derived for physician services. To estimate the number of physician consultations attributable to each chronic disease we multiplied the regression coefficient for the disease's dummy variable by the number of people who reported having the disease. All analyses were performed using SPSS® (Statistical Package for the Social Sciences: SPSS Inc., Chicago, Illinois) version 10.

Results

Table 1 presents a stratification by the number of chronic diseases and compares the percentage of people in each stratum with the percentage of physician and hospital use. As seen in this table, chronic disease comorbidities are commonplace, and with the move to higher age groups, their prevalence grows. In the lowest age group, 20-39 years, people with one or more chronic diseases use more than "their share" of services, and those with no chronic disease use less than "their share". In the older age groups the percentage of services used exceeds the percentage of people only for two or more chronic diseases

TABLE 1
Number of chronic diseases and utilization of physician and hospital services in the 1998-1999 National Population Health Survey

Prevalence by specific chronic condition is shown in Table 2. In the under 60 age group three chronic diseases are found in 10% or more of people: back problems (15%), arthritis or rheumatism (12%) and migraine headaches (10%). In people aged 60 and over, there are seven chronic diseases with 10% prevalence or higher. The prevalence of arthritis or rheumatism (46%) and high blood pressure (35%) is about twice as high as the next most prevalent disease. Table 3 shows the regression analysis using number of chronic diseases in the equations for physician services. The number of chronic diseases is a highly significant predictor of utilization (p < 0.001) in all the regressions. In the younger age group an additional chronic disease is associated with 1.74 more physician visits per year, and in the older age group the increase in physician visits predicted is 1.29. The number of chronic diseases is also a statistically significant predictor of very high physician use (more than 12 visits per year). The odds ratios in the logistic regressions reported in Table 3 suggest that an additional chronic disease is associated with a 76% increase in the chance that a person under 60 is a high user of physician services and with a 51% increased chance in a person over 60.

TABLE 2
Prevalence of specific chronic conditions* in the 1998-1999 National Population Health Survey

* Listed in order of prevalence in the >=60 age group.

TABLE 3
Regression results for physician services

The confidence intervals shown in Figures 1 through 4 are for the coefficients or odds ratios of the chronic disease dummy variables when the number of chronic diseases is replaced with the set of specific chronic disease variables in the regressions. Only those variables whose coefficient confidence interval excludes 0 in the linear regression or whose odds ratio confidence interval excludes 1 in the logistic regression are shown. Despite fairly large confidence intervals, these data indicate that some diseases consistently have larger effects on utilization of physician services than others. Heart disease has a large effect in both age groups. Cancer in the older age group and bowel disorders in the younger age group seem noteworthy for their large effects on utilization.

FIGURE 1
Confidence intervals for coefficients of chronic disease variables*
(among those aged < 60) for physician consultations

* Variables listed in order of size of coefficient; omitted chronic diseases did not have statistically significant coefficient.

FIGURE 2
Confidence intervals for coefficients of chronic disease variables*
(among those aged >=60) for physician consultations

* Variables listed in order of size of coefficient; omitted chronic diseases did not have statistically significant coefficient.

FIGURE 3
Confidence intervals for odds ratios of chronic disease variables*
(among those aged < 60) in logistic regression, > 12 physician consultations

* Variables listed in order of size of odds ratio; omitted chronic diseases had prevalence < 1% or a confidence interval including 1.

FIGURE 4
Confidence intervals for odds ratios of chronic disease variables*
(among those aged >= 60) in logistic regression, > 12 physician consultations

* Variables listed in order of size of odds ratio; omitted chronic diseases had prevalence < 1% or a confidence interval including 1.

Regression analysis of hospital utilization is presented in Table 4, with confidence intervals for regression coefficients and odds ratios shown in Figures 5 through 8. An additional chronic disease raises the probability of any hospitalization in the previous year by 44% in the younger age group and by 27% among people over age 60. Although the explained variation is quite low (adjusted R² < 10%) in the regression for hospital nights, the number of chronic diseases has a statistically significant coefficient in both age groups. An additional chronic disease is associated with 0.77 more hospital nights in the younger age group and 0.60 hospital nights in the older age group.

TABLE 4
Regression results for hospital services

FIGURE 5
Confidence intervals for odds ratios of chronic disease variables
(among those aged < 60) in logistic regression, hospitalization

* Variables listed in order of size of odds ratio; omitted chronic diseases had prevalence < 1% or a confidence interval including 1.

FIGURE 6
Confidence intervals for odds ratios of chronic disease variables*
(among those aged < 60) in logistic regression, hospitalization

* Variables listed in order of size of odds ratio; omitted chronic diseases had prevalence < 1% or a confidence interval including 1.

FIGURE 7
Confidence intervals for coefficients of chronic disease variables*
(among those aged < 60) for hospital days

* Variables listed in order of size of coefficient; omitted chronic diseases did not have statistically significant coefficient.

FIGURE 8
Confidence intervals for coefficients of chronic disease variables* (among those aged >= 60) for hospital days

* Variables listed in order of size of coefficient; omitted chronic diseases did not have statistically significant coefficient.

In Table 5 the product of regression coefficient times number of people with the disease is calculated to estimate the total physician consultations attributable to the disease. In the younger age group, four conditions (back problems, arthritis or rheumatism, high blood pressure and migraine) each account for more than twice as many consultations as the other conditions. Except for high blood pressure, this is largely a result of the frequency with which these conditions occur, rather than the resource impact factor. The resource use coefficient is high for diabetes, heart disease, and bowel disorder, but the frequency of people with these conditions is not high in people under 60. Arthritis or rheumatism and high blood pressure are also at the top of the list in the older age group, in both cases because of the numbers with the disease. Fewer people have heart disease, but its relatively high resource factor (2.3) results in a higher overall measure of visits saved if the disease were eliminated.

TABLE 5
Total physician consultations as derived from the product of the regression coefficient and the number of people with the disease

Discussion

In this article we developed an analysis of the economic burden of chronic diseases using a common measure of burden for all conditions. This measure was decomposed into two separate components - per-person utilization and disease prevalence - and adjusted for the numbers and types of chronic disease comorbidities. We concentrated our measure on the use of physician services using data from the NPHS, a population based Canadian survey, because of small samples for hospital services.

Given the frequency with which we observed concurrent chronic diseases in individuals of all age groups as well as the influence of multiple diseases on utilization, adjustment for comorbidity is appropriate. Our results demonstrate that, after such adjustment, chronic diseases differ in the extent to which their presence is associated with increased utilization. The order of magnitude of the variation in per capita effect is three to four times. For example, in the younger age group sinusitis increases per capita physician use by about one consultation per year whereas bowel disorder increases it by about four consultations per year. In the older age group, thyroid disease increases per capita physician use by about one day while stroke increases it by about three days.

As we did not specifically study disease characteristics and their effects on physician utilization, generalizations regarding this are speculative. Our data did not include the specific reasons for physician visits. However, our findings of the ordering of diseases by regression coefficients in Figures 1 and 2 can be explained by presumptive drivers of utilization for specific diseases. Disorders that typically require minimal monitoring and are unlikely to progress once appropriately diagnosed and treated, such as thyroid disorders, are associated with less frequent physician visits. Urinary incontinence and migraine, which may be accompanied by troublesome symptoms, but are not typically associated with dire consequences or the need for frequent monitoring, are also ranked lower. Disorders that are identified as being associated with heavy physician utilization, such as cancer, diabetes mellitus and heart disease, may be progressive in nature despite treatment, may have dire consequences including death, and may require frequent revisions to therapy. High blood pressure, by itself an asymptomatic disease, requires observation and possible alterations to therapy over time, and appears in the middle of the list.

An important finding from the perspective of composite resource utilization is that the overall effect on utilization of a specific disease seems more driven by the number of people who have it than by the per capita effect. At the top of the list for total effect are disorders such as musculoskeletal disease and high blood pressure, which have high prevalence although their per capita effect on utilization is moderate or low.

Decomposing utilization into prevalence and a coefficient of use may allow a clearer evaluation of the potential impact of numerous changes on resource use. For example, the prevalence of diabetes is low compared with other chronic diseases examined here. However its widely predicted rise in the coming decade, combined with the high coefficient of utilization found in the present study, may substantially increase its future ranking.

Typically, the regression coefficient for number of chronic diseases as well as for specific chronic diseases is smaller in the over 60 age group than in the younger group. This is true for both physician and hospital utilization. Possible reasons are that these diseases are treated more aggressively in the younger population, or that older people have lived with them longer and are better at self-care or at using alternative health care services. Another possibility is that since younger people have relatively few chronic diseases the presence of one creates much anxiety and thus a tendency to seek added care. In the older group, in contrast, the incremental effect of an additional disorder on the demands for physician care may be less when several other chronic diseases are already present.

There is no other measure of the economic burden of the conditions identified in this study that is categorized in the same groupings that we used. Nevertheless, there are some data in the Health Canada report Economic Burden of Illness in Canada, 1998¹ that allow us to assess our results, although this report covers all disease and not just chronic conditions. In this document, the burden of physician costs for musculoskeletal disorders is ranked quite low (see Figure 8). In our analysis, musculoskeletal disorders (arthritis/rheumatism and back pain) are ranked highest for the under 60 group, and arthritis/rheumatism was ranked highest for the over 60 group. Several heart disease categories were rated as high for the over 60 group, and this corresponded with the Health Canada rankings. However, respiratory conditions were of lower rank in our analyses.

The confidence intervals reported are, of course, dependent on the sample size and the particular estimation technique used. The use of bootstrapping estimation would have produced wider confidence intervals. The larger sample size in the 2000/01 Canadian Community Health Survey (not available to us when the analysis was conducted) would have produced smaller confidence intervals.

Our analysis does not attempt to study the entire economic burden of chronic disease in Canada, but, rather, only some aspects of the direct burden due to health services utilization. One of the prime shortcomings of our analysis is the omission of hospitalization in estimation of the overall economic burden of chronic disease. The main reason for its omission was the small samples of patients in specific disease groups, with the consequent loss of statistical power. However, even just focusing on physician services, our analysis indicates that the adjustment for comorbidities will have an impact on economic burden rankings. As well, we believe that decomposing the analysis will provide a more precise approach to measuring the concepts of economic burden and attributable costs.

Acknowledgements

This paper was supported through a grant to the AIMS project of the Institute of Health Economics. This project has been funded by grants from Merck Frosst Inc. and Pharmacia Inc. through Alberta Health and Wellness. We thank Kathy Gooch, Manager, AIMS, for her support.

References

1. Health Canada. Economic burden of illness in Canada, 1998. Ottawa: Public Health Agency of Canada, Health Canada, 2002.
2. Badley EM. The economic burden of musculoskeletal disorders in Canada is similar to that for cancer, and may be higher. J Rheumatol 1995;22(2):204-6.
3. Coyte PC, Asche CV, Croxford R, Chan B. The economic cost of musculoskeletal disorders in Canada. Arthritis Care Res 1998;11(5):315-25.
4. Coyte PC, Asche CV, Elden LM. The economic cost of otitis media in Canada. Int J Pediatr Otorhinolaryngol 1999;49(1):27-36.
5. Stephens T, Joubert N. The economic burden of mental health problems in Canada. Chron Dis Can 2001;22(1):18-23.
6. Jacobs P, Blanchard JF, James RC, Depew N. Excess costs of diabetes in the Aboriginal population of Manitoba, Canada. Can J Public Health 2000;91(5):298-301.
7. Hodgson TA, Meiners MR. Cost-of-illness methodology: a guide to current practices and procedures. Milbank Mem Fund Q Health Soc 1982; 60(3):429-62
8. Rice DP, Kelman S, Miller LS, Dunmeyer S. The economic costs of alcohol and drug abuse and medial illness. San Francisco: Institute for Health and Aging, University of California, 1990.
9. Simpson SH, Corabian P, Jacobs P, Johnson JA. The cost of major comorbidity in people with diabetes milletus. CMAJ 2003;168(13):1-7.

Footnote

* Defined as "foreign or domestic open source material that is usually available through specialized channels and may not enter normal channels or systems of publication, distribution, bibliographic control, or acquisition by booksellers or subscription agents" (US Interagency Gray Literature Working Group, 1995)

Author References

John Rapoport, Department of Economics, Mount Holyoke College, South Hadley, Massachusetts, USA

Philip Jacobs, Department of Public Health Sciences, University of Alberta, Edmonton, Alberta, Canada

Neil R Bell, Department of Family Medicine, University of Alberta, Edmonton, Alberta, Canada

Scott Klarenbach, Department of Medicine, University of Alberta, Edmonton, Alberta, Canada

Correspondence: Philip Jacobs, Institute of Health Economics, #1200-10405 Jasper Avenue, Edmonton, Alberta Canada T5J 3N4; Fax: (780) 448-0018; E-mail: pjacobs@ihe.ab.ca

[Previous] [Table of Contents] [Next]