Income Differences in Persons Seeking Outpatient Treatment for Mental Disorders:  A Comparison of the United States With Ontario and the Netherlands FREE

IMPROVING EQUITY in mental health care within a country—guaranteeing that those in need receive services independently of the person's ability to pay¹—is increasingly relevant as national policies seek to modify disease outcomes.² However, there is considerable cross-national variation in how the organization and financing of mental health services are designed to provide equity in access and quality of treatment. Some countries, such as Turkey, have failed to develop a special health care program for the poor,³ while others, such as the United States, have developed separate health care programs for those unable to pay.⁴ Still other countries, such as Canada, have established universal health insurance plans explicitly designed to provide equality in the opportunity for treatment regardless of ability to pay.⁵

The US system has been criticized for being inherently inequitable.^6- 7 People in different income strata are seen as segregated into separate sectors believed to differ in type and quality of care.^6- 7 Consistent with this criticism, research has shown that low-income people in the United States have comparatively low rates of treatment for mental disorders^8- 9 and are less likely than their economically advantaged counterparts to be treated by mental health specialists.^10- 11 These findings are especially disturbing given that need for treatment is higher among low-income people than in the rest of the population,^12- 13 and that those with reported mental disorders seem to experience significant barriers to the receipt of needed medical care.¹⁴

The United States, Canada, and the Netherlands vary in how their health care systems are organized and financed to provide equitable access in health and mental health services. In the United States, approximately 60% of the population was covered by private insurance during the early 1990s.¹⁵ Insurance coverage differed widely¹⁶ and tended to be associated with income, making the effects of income difficult to disentangle from those of coverage. The poor received mental health care either through state public mental health programs in community clinics or through independent specialty providers paid through Medicaid coverage.¹⁷

Ontario¹⁸ has universal health insurance coverage available to residents through a government-sponsored single-payer system explicitly designed to equalize access to treatment. Mental health services are provided by physicians¹⁹ (such as family physicians²⁰) reimbursed by government in a fee-for-service modality or by psychologists and social workers employed by government-funded mental health centers.²¹ The system places no limits on inpatient stays or outpatient visits for mental health care, but does apply supply-side controls to limit use.²² Despite Canada's universal access system, recent evaluations suggest problems of access^22- 23 and greater use of specialist services for medical care among high-income vs other groups.²⁴

In the Netherlands,²⁵ every person subject to Dutch income taxation is covered by the Exceptional Medical Expenses Act, funded by a compulsory premium; benefits include comprehensive psychiatric care. The Social Health Insurance Act, a statutory insurance for employees earning less than $40 000 per year, covers normal medical expenses not included by the Exceptional Medical Expenses Act. In the Netherlands, for the 35% of the people who are not covered either by the Social Health Insurance Act or by specific plans for public employees, private health insurance is an available option. Patients getting outpatient mental health pay a fixed fee per visit. Thus, a fee-for-service structure for the higher-income population and a capitated system for the lower-income population coexist.²³ Social and psychiatric services are stratified so that specialty mental health care usually requires referrals from general practitioners, who retain an integral role in the mental health care of the patient.²⁶ The Netherlands also has a broad concept of mental health that encompasses psychosocial problems,²⁷ and additional policies regarding health promotion and disease prevention, such as outreach programs.²⁶

The purpose of the current report is to present comparative results on the relationship between income and mental health treatment in 3 countries that have fairly different systems of organization and financing of health care: the United States, the Netherlands, and the province of Ontario, Canada. This report takes advantage of the general-population epidemiological surveys conducted in these 3 sites, using parallel sample designs and instruments. Drawing from this comparative analysis of mental health care use across 3 countries may lead us to observations of how national policies may achieve equity in mental health care use.

The present study addresses 2 questions: (1) Are there site differences in the probability of receiving any outpatient mental health treatment by income among people with mental disorders assessed in these surveys? We expected that in the United States (before managed care) mental health service use would be higher by persons at the extreme ends of the income distribution, equal for the poor and nonpoor in Ontario, and lower for the wealthy in the Netherlands. (2) Are there cross-national differences in the association between income and sector in which treatment is received? Several differences were expected. In the United States, we expected proportional use of the specialty sector to be lower among low-income people. In Ontario, sector of care was not expected to vary by income strata. The Netherlands was expected to have a higher use of the general health sector for all income groups except for the high-income group.

The data come from the 1990-1992 National Comorbidity Survey (NCS) (n=5384),¹³ the 1990-1991 Mental Health Supplement to the Ontario Health Survey (the Supplement) (n=6321),^28- 29 and the 1996 Netherlands Mental Health Survey and Incidence Study (NEMESIS) (n=6031).^30- 31 Each of these surveys carried out face-to-face interviews with a probability sample of the general population and obtained parallel information on the lifetime and recent prevalence of DSM-III-R disorders and use of mental health services. The focus of this report is on disorders and outpatient treatment during the 12 months previous to the interview.

The sample for each of the 3 surveys was based on a multistage area-probability design in which a single respondent was selected for interview in each sampled household. Weights were imposed on the data to adjust for differential probabilities of selection and, in the NCS, for differential response. Although rates of psychiatric morbidity in the NCS were higher among nonresponders, the overall morbidity rates were not substantially affected by the adjustment for nonresponse. Response rates were 82.4% in the NCS, 67.4% in the Supplement, and 69.7% in NEMESIS. Given that the age ranges of the samples differed, all the analyses reported herein are limited to respondents in the age range of 18 to 54 years.

Diagnoses are based on the Composite International Diagnostic Interview (CIDI),³² a fully structured research diagnostic interview designed for use by trained interviewers who do not have clinical experience. Twelve CIDI disorders were considered: mood disorders (major depression, dysthymia, and mania); anxiety disorders (generalized anxiety disorder, panic disorder, simple phobia, social phobia, and agoraphobia); and substance use disorders (abuse of or dependence on either alcohol or drugs). The World Health Organization CIDI Field Trials³³ and an NCS clinical reappraisal study³⁴ documented good reliability and validity for all of these diagnoses other than mania.³⁵

All respondents, independent of symptoms, were asked whether they had ever seen a professional for problems with their mental health. Positive responses were followed by inquiries on the recency of treatment (use of outpatient services in the last year) and the types of professionals seen. The NCS and the Supplement assessed service use by asking, "Did you ever in your lifetime go to see any of the professionals in this list [the lists were identical except that the NCS added programs provided in jails, prisons, and drop-in centers as options] for problems with your emotions, nerves, or your use of alcohol or drugs?" The NEMESIS asked, "I will list a number of persons or agencies where people can go for help. Did you ever in your lifetime go to see [providers were assessed by separate yes/no questions] for psychological problems or alcohol or drug-related problems of your own?" The wording of questions varied, but was sufficiently similar to allow parallel coding of information as to whether treatment was obtained in each of 3 service sectors: general medical, specialty, and human services. The main differences between the lists were that NEMESIS excluded cardiologists, gynecologists, and other physicians and instead asked about company physicians, added crisis centers as another option, and combined psychiatrists, psychologists, and psychotherapists under 1 question.

Any mental health treatment was defined as 1 or more visits to a professional for a mental health problem. General medical sector was defined as mental health treatment by a medical physician other than a psychiatrist, regardless of setting, or as treatment by an ancillary health professional (eg, nurse), excluding social worker and counselor, in any non–mental health setting. Specialty sector was defined as mental health treatment by a psychiatrist, psychologist, or psychotherapist, or as treatment by any professional in a specialty mental health setting (eg, psychiatric outpatient clinic). Human services sector was defined as mental health treatment received in a social or welfare service agency or by a social worker or counselor in any setting other than a specialty setting.

Total family income was assessed by parallel questions in the 3 surveys. The NCS asked, "Please tell me which [of a series of categories presented on a card] represents your total household income before taxes last year, including salaries, wages, social security, welfare, and any other income?" The Supplement asked, "What is your best estimate of your household income from all sources in 1990 before income tax deductions?" The NEMESIS phrased the question as, "We would like to compare the results of this survey with the incomes of the Dutch population. That is why it is important for us to have information about the total income of your household per month [transformed to annual household income]. Can you indicate which category corresponds with the total income per month of your household?"

Careful consideration was given to select the most appropriate way to define income groups for purposes of comparison. Simply translating the 3 currencies into a common currency (such as dollars) on the basis of exchange rates does not take into account the variable purchasing power of a dollar in each site. A comparison of percentages ignores the fact that the lowest 10% in one site may be richer than the lowest 10% in another site. A comparison of the standardized income distributions has the undesirable effect of assuming that the means and variances of all sites are equal. Sensitivity analyses were performed with several income definitions. Given that our primary purpose was to identify people in other countries who were comparable with people in the 4 quartiles in the United States, we established the quartiles based on the US income distribution. We also investigated the effect of income across service use by defining income groups based on the quartiles of income in each country separately, guaranteeing 25% of the population in each group in each country. The results were uniformly in the direction of diluting the differences between sites by income. We believe this is because the lowest income quartile in Ontario or the Netherlands contains many "nonpoor" and the highest quartile many "nonrich" in these countries with more egalitarian distributions of income. This set of results led us to use absolute rather than relative income as the method for comparing Ontario and the Netherlands with the United States.

Total family income was therefore coded in 4 categories based on the income distribution in the United States. For the United States, these categories were formed from the quartile distribution points (≤$18 750, $18 751-30 000, $30 001-60 000, ≥$60 001). For Ontario and the Netherlands, income was converted to US dollars and 4 categories were created using the same income cutpoints as in the United States. While this constrains the distribution of income across categories in the United States, it places no constraints on the distribution in Ontario and the Netherlands.

Three sociodemographic control variables available across data sets and found to be associated with mental health service use^36- 37 were considered: age, sex, and marital status. Age was coded categorically in the ranges 18 to 24, 25 to 34, 35 to 44, and 45 to 54 years. Sex was coded dichotomously (male or female). Marital status was coded categorically as married, never married, and previously married (ie, separated, divorced, or widowed). Given the differences in health care structures, insurance data were not comparable across sites.

Differences in psychiatric morbidity and severity of illness by country may account for differences in the response of utilization patterns to income. Therefore, psychiatric morbidity was operationalized as the presence or absence of any of the assessed psychiatric disorders and as the recency of such disorders, ranging from no long-term disorder to 2 or more disorders lasting at least 12 months.

Site-specific cross-tabulations were computed to study psychiatric disorders and treatment use and the gross associations between income and treatment. Logistic regression analysis was then used to adjust for age, sex, marital status, and psychiatric morbidity in estimating the net associations between income and treatment. Differential willingness to report income across surveys (refusal rate was 8.0% in the NCS, 15.4% in the Supplement, and 4.7% in NEMESIS) was controlled by adding a dummy variable to distinguish respondents who refused to report their income from other respondents. Regression-based imputation³⁸ was used to assign income values to NCS respondents with missing scores. Missing values were not imputed in the other 2 surveys. Imputing values for missing data influences only the SEs and does not introduce bias into the point estimates. Pooled analyses with interaction terms were used to test the significance of between-site differences in the associations of income with treatment.

The SEs of prevalence estimates and of logistic regression coefficients were computed using the method of jackknife repeated replications to adjust for the design effects introduced by clustering and weighting of observations.³⁹ The significance of differences between pairs of coefficients was evaluated with z tests based on these corrected SEs. The significance of differences among larger sets of coefficients was evaluated with Wald χ² tests based on jackknife repeated replication coefficient variance-covariance matrices. All evaluations of significance are based on 2-sided tests with P≤.05.

As shown in Table 1, there were significant between-site differences in the 12-month prevalence of having at least 1 psychiatric disorder assessed (29.1% in the United States, 24.4% in the Netherlands, and 19.9% in Ontario). There were also significant differences across sites in the estimated probabilities of any 12-month mental health treatment (referred to as any treatment in this article), as shown in the table. The estimated probabilities of receiving any treatment were higher in the Netherlands (31.7% among respondents with a disorder and 7.6% among those with no disorder) than in the United States (22.3% and 6.3%, respectively) or Ontario (21.8% and 3.4%, respectively).

The remainder of Table 1 shows data on the proportional distribution of any treatment in the general medical, specialty, and human services sectors. A significantly higher proportion of patients was treated in the general medical sector in the Netherlands (78.8% of all mental health service users with 1 or more disorders and 67.7% of those without a disorder) and in Ontario (65.9% and 64.6%, respectively) than in the United States (35.6% and 31.4%, respectively). There were no significant between-site differences in the proportional use of specialty sector services, either in the subsample with a disorder (52.5%-55.3% across sites) or in the subsample with no disorder (42.3%-46.8% across sites). Proportional use of the human services sector was significantly higher in the United States (38.5% of all service users with 1 or more disorders and 40.9% of those without a disorder) than in either Ontario (15.3% and 16.0%, respectively) or the Netherlands (18.6% and 14.9%, respectively).

The distributions of annual household income are presented in Table 2. There was a significant between-site difference in these distributions, largely because of the higher proportion of US respondents with incomes in the lowest category (24.2%, compared with 6.9% in Ontario and 10.7% in the Netherlands).

The results in Table 2 document a consistent and significant inverse relationship in each of the surveys between income and the prevalence of having at least 1 psychiatric disorder. The odds ratios (ORs) comparing disorder prevalence in the lowest vs highest income categories vary only modestly across sites, from 2.2 in the Netherlands to 1.5 in Ontario. Results also show that there was no significant association between income and receiving any outpatient psychiatric treatment in any of the sites when differences in demographics and psychiatric morbidity were controlled.

Findings in Table 3 examine the between-site differences in more detail by distinguishing subsamples of respondents who did not meet criteria for any of the disorders compared with those who did meet criteria for any of these disorders in each site. Results in the United States show marginally significant elevated relative odds of treatment among respondents with no disorder in the highest income category. There was no significant income-treatment association in either subsample (no disorder or any disorder) in Ontario. In the Netherlands, a negative and significant association between income and any treatment was observed among those with no disorder. However, there was no significant income-treatment association in the subsample with any psychiatric disorder for any of the sites. There was a significant between-site difference in the association between income and any treatment for respondents with no disorder.

Associations between income and the proportional distribution of treatment across service sectors are presented in Table 4. As shown, the between-site difference in the association between income and proportional use of the general medical sector was not significant. However, there was a significant inverse association in Ontario, a significant U-shaped association in the Netherlands, and no meaningful association in the United States. Subsample analyses (results not shown) found that these associations existed both among respondents who met criteria for any psychiatric disorder and among those who did not meet criteria for any of these disorders.

There was a significant between-site difference in the association between income and proportional use of the specialty sector because of a significant positive association in the United States, a significant U-shaped association in the Netherlands, and no meaningful association in Ontario. Subsample analyses (results not shown) found that these between-site differences hold up in subsamples defined by the presence or absence of the DSM-IV disorders assessed in the surveys.

Finally, there was no significant between-site difference in the association between income and proportional use of the human services sector. A significant inverse association was present in both the United States and the Netherlands but not in Ontario. Subsample analyses (results not shown) found that these between-site differences hold up for those assessed with DSM-III-R disorders and for those who did not meet criteria for any of these disorders.

The results reported are limited by several methodologic considerations. These include differences in the 3 survey conditions, the lack of refined measures to establish receipt of mental health treatment rather than mental health visits, and the study's inability to account for heterogeneity across the sites in attitudinal values related to mental health care. Therefore, some unknown part of the difference in the results may have been because of differences in the types of instruments, training procedures, or field quality control used in the 3 surveys. Furthermore, the CIDI's lack of measurements of syndrome severity, distress, or impairment could explain some of the between-site differences in rates of treatment. Research has shown that these dimensions are more important than dichotomous diagnostic classifications in determining help-seeking behaviors.^40- 43 Finally, while we used income as a proxy for economic access, it may be a proxy for other factors, such as differential perceived need for treatment,⁴⁴ differential readiness to use treatment,⁴⁵ and differential attitudes about physicians.³⁶

Within the context of these limitations, however, many of the results are in agreement with previous studies and highlight the remarkable pattern of consistency across sites. The findings document that 1 of every 4 or 5 people in the general population met criteria for a psychiatric disorder during a year,^{13,31,46- 48} that low-income people had a higher prevalence of mental disorders than others in the population,^12- 13,49 and that only a minority of people with mental disorders obtained treatment within a given year.^42,46,50

Several noteworthy between-site differences were also documented. The first is a difference in the rates of obtaining treatment; the rates were similar in the United States and Ontario but much higher in the Netherlands. Since the systems of Ontario and the Netherlands are more similar, differences in the financing of health services cannot fully account for this finding. Some alternate explanations include the Netherlands' unique tradition of primary-care outreach and screening for mental health problems,²⁶ or their differential knowledge of, positive attitudes toward, and reduced stigma associated with mental health care.³⁷

The distribution of patients across the service sectors also differed by site and led to important insights into the sites' differences in mental health care organization. The dramatically higher proportion of patients treated by the general medical sector in Ontario (two thirds) and the Netherlands (three fourths) compared with the United States (one third) confirms that the primary care physician served a stronger gatekeeper function in the former 2 sites. Furthermore, the significantly lower proportion of patients seen in the human services sector in the Netherlands and Ontario suggests that family physicians in Canada²⁰ and general practitioners in the Netherlands²⁶ may more commonly undertake these social and psychiatric services. Given the greater percentage of people in poverty in the United States, there might be a greater need than in the other sites to structure a more defined human services sector.

Within the individual sites, no significant association was found between income and probability of obtaining any treatment, suggesting that, contrary to the proposed hypothesis, each system provides some type of mental health care to its citizens, regardless of their ability to pay. Because many of these services are offered outside the specialty sector, however, such findings may not hold true under more strict criteria regarding receipt of mental health treatment.

In contrast, specification analyses across sites revealed an association between income and treatment in the data from respondents who did not meet criteria for any of the assessed disorders. Their responses showed both a positive income-treatment trend in the United States and a negative trend in the Netherlands. Indeed, when care is discretionary, attitudinal and propensity characteristics of the person may interact with the organization and financing of the services to increase or reduce the likelihood of seeking mental health treatment. One plausible interpretation is that, in contrast to Ontario, barriers to treatment among people with low need for services in the United States may be primarily financial. The decreased access to treatment among high-income respondents with low need in the Netherlands probably reflects the fact that Social Health Insurance is not available to Dutch residents in the top third of the income distribution and disposable income might not be comparable with that in the United States.

In addition, a significant between-site difference was also found for income and specialty care. High-income US patients are significantly more likely to obtain specialty sector treatment, while low-income patients are significantly more likely to receive treatment in the human services sector. These results reflect financial barriers to specialty care for the poor in the United States, barriers nonexistent within the universal insurance system in Ontario. These findings raise concerns about the quality and content of care received by low-income patients, a criticism of the US system. Although more research is needed to support these claims, available evidence is consistent with this possibility.^{6- 7,10- 11,51- 52} Data show that the proportions of depressed patients in the United States and Ontario who receive adequate medication treatment are much higher among those treated by psychiatrists than among those treated by primary care physicians.⁴⁴ Our findings suggest that social class may limit the type of care patients receive in specialties other than mental health care. Future research on differential access clearly needs to address the question of whether separate is equitable in the provision of mental health services to people who differ in economic resources.

Accepted for publication December 22, 1999.

Data collection for the National Comorbidity Study was supported by grants MH48376 and MH49098 from the National Institute of Mental Health, Bethesda, Md; for the Mental Health Supplement to the Ontario Health Survey by the Ontario Ministries of Health and Community and Social Services through the Ontario Mental Health Foundation, Toronto; and for the Netherlands Mental Health Survey and Incidence Study by the Netherlands Ministry of Health, Welfare, and Sports and the Medical Sciences Department of the Netherlands Organization for Scientific Research, the Hague, and the National Institute for Public Health and Environment, Hilversum, the Netherlands. Joint data analyses of the 3 surveys and preparation of our study was supported by grant DA11121 from the National Institute on Drug Abuse, Bethesda, Md (Dr Kessler).

A complete list of collaborators and publications can be found on the International Consortium in Psychiatric Epidemiology home page at http://www.hcp.med.harvard.edu/icpe/.

Corresponding author: Magarita Alegría, PhD, School of Public Health, University of Puerto Rico, PO Box 365067, San Juan, Puerto Rico 00936-5067 (e-mail: m_algeria@rcmaca.upr.clu.edu).