Data source, subject selection, and target population
Public-use data from the Joint Canada/United States Survey of Health (JCUSH), 2002–03, were accessed and analyzed [18]. The JCUSH was a one-time stratified random sample telephone survey of non-institutionalized adult (aged 18 or greater) residents of Canada (N = 3505) and the U.S. (N = 5183) conducted between November 2002 and June 2003. Households were selected via a random digit dialing (RDD) process, and all interviews were conducted from the regional offices of Statistics Canada using the Computer-Assisted Telephone Interviewing (CATI) method [19]. The interview was administered in either French or English for Canadian respondents, and in either Spanish or English for U.S. respondents. The survey response rates were 66% and 50% in Canada and the U.S., respectively [19]. The target population was persons 18 years old or older residing in private dwellings with land-line telephones in Canada and the U.S., excluding persons living in the territories of Canada or the U.S. The target population sizes in Canada and the U.S. were 24,046,837 and 206,417,185, respectively [20, 21].
Chiropractic and general practitioner utilization
Respondents were queried about the number of visits to or contacts with various types of health professionals, including family doctors or general practitioners and chiropractors. Numbers of provider-specific visits per year were capped at 31 or more. Respondents within each country were categorized according to their type of reported health-care utilization in the past 12 months: any chiropractic care (DC), family doctor or general practitioner care only (GP), both DC and GP care, and DC without GP care.
Socioeconomic and demographic factors
Socioeconomic and demographic variables were age (18–44, 45–64, >64), sex, race/ethnicity (white only, other/multiple), marital status (married/with partner, widowed, separated/divorced, single), highest level of school completed or highest degree received (no high school degree, high school degree, some college, 4-year college degree), main source of income (employment vs. other), and amount of household income (adjusted for household size and placed in quintiles).
Health status and reported chronic conditions
Current general health status was assessed with several measures, including a 5-point measure of self-rated general health (excellent, very good, good, fair, poor), and presence of one or more chronic conditions (e.g., conditions that had lasted or were expected to last 6 months or more and had been diagnosed by a doctor or other health professional). Conditions included asthma, osteoarthritis, rheumatoid arthritis, hypertension, emphysema or chronic obstructive pulmonary disease, diabetes, heart disease, coronary heart disease, angina pectoris, and heart attack history. The reported chronic conditions were summed to create a chronic condition index ranging from 0 (no reported conditions) to 10.
Mental health status, specifically depression, was assessed with a subset of questions from the Composite International Diagnostic Interview (CIDI) [22], which covers depressive disorder symptoms itemized in the Diagnostic and Statistical Manual of Mental Disorders (DSM-III-R) and produces diagnoses according to the Diagnostic Criteria for the Research of the ICD-10 [23]. Responses were transformed into a probability estimate of a diagnosis of major depressive episode (MDE) in the past 12 months. Respondents with estimates reflecting a 90% or greater certainty of a positive diagnosis (0.9 or greater) were classified as having had major depressive episodes in the past year [22–24]. Visits to mental health professionals were also tabulated.
Health-related quality of life was measured with the well validated Health Utility Index (HUI) [25]. The HUI includes components related to vision, hearing, speech, mobility, dexterity, emotions, cognition, and pain and discomfort [25].
Activity restrictions and lifestyle factors affecting health
The JCUSH included queries on restriction of activities (sometimes, often, never) due to one or more chronic health conditions, and activity limitations due to pain (no pain, pain but no activity limitations, pain prevents a few activities, pain prevents some activities, pain prevents most activities). Respondents were asked about the specific conditions or health problems responsible for any difficulties with performing activities of daily living, with back or neck problems among the specific response options. Respondents reporting pain rated the usual intensity of their pain or discomfort as being mild, moderate, or severe.
Respondents were asked about their smoking status, height and weight, and physical activity levels. Current smokers were those individuals who reported having smoked at least one whole cigarette and at the time of the survey smoked cigarettes every day. Each person's body mass index (BMI) was computed by dividing weight in kilograms by the square of height in meters. The World Health Organization's categories for classifying persons according to BMI are used [26]: underweight (<18.5), normal weight (18.5 – <25), overweight (25 – <30), and obese (> = 30). Activity-specific metabolic equivalent task (MET) scores [27, 28] and responses to questions about the frequency and duration of participation in leisure time physical activities in the past 3 months were used to compute total daily energy expenditure [29], which were then used to classify persons as being physically active (> = 3), moderately active (1.5 – <3), or inactive (<1.5). Respondents were also classified according to their frequency of physical activity lasting more than 15 minutes in the past 3 months (regular [> = 12], occasional [4 – <12], infrequent [<4]) [30].
Health-care utilization, perceived unmet needs, and satisfaction with care
The JCUSH included several items on hospitalizations and the use of and visit frequency to medical doctors and other health care professionals in the past 12 months; prescription medication use in the past month and number of medications taken in the past 2 days; and in the U.S., health insurance status during the past 12 months. Respondents were also asked if they needed a health-care service in the past 12 months but didn't receive it because of lack of access, cost, or other reason (unmet health care need); and about their satisfaction with the overall quality of health care in the past 12 months, with physician care during their most recent visit (excellent, good, fair, poor), and the way health care services and physician care were provided (very satisfied, somewhat satisfied, neither satisfied nor dissatisfied, somewhat dissatisfied, very dissatisfied).
Statistical methods
Distributions and percentages weighted to reflect the complex survey design and nonresponse were produced according to each variable, stratified by pattern of health-care use and country. The survey weight corresponds to the number of persons represented by the respondent for the target population [19]. Post-stratification using age, sex, region (Canada only), and race/ethnicity (U.S. only) was performed to ensure that the final weights sum to the population estimates, based on Canada's 1996 Census of Population [20] and the United States' October 2002 Current Population Survey [21]. SAS was used for data management and preliminary statistical analysis [31]. Because of the need to account for the complex survey design when estimating variances and confidence intervals, SUDAAN was used in modeling of associations and variance estimation [32]. SUDAAN uses the Taylor series method for estimation of variances.
Logistic regression modeling was employed to estimate crude and adjusted associations (odds ratios and 95% confidence intervals) of each factor on care seeking from chiropractors vs. general practitioners. Specifically, two sets of crude and adjusted models were run: one to estimate associations of potential predictors with any chiropractic care vs. general practitioner care only, and another to estimate associations of potential predictors with chiropractic care only vs. both chiropractic and general practitioner care. Sets of one or more binary (dummy) variables were used to model the effects of all predictors, with the exceptions of the health utility, chronic condition, and prescription medication indices, which were modeled as continuous variables, and age, which was modeled as both a categorical and a continuous predictor. Potential between-country differences in effect estimates were evaluated by the inclusion of interaction terms in additional sets of multivariable models.
Potential confounders were identified a priori as those measured variables thought to be predictors of type of health-care use and possibly associated with one or more of the selected variables of interest. Because of inherent multidirectional relations (e.g., variables acting as both causes and consequences of certain predictors and/or health-care use) and lack of longitudinal data, two sets of separate multivariable logistic regression models were built to estimate adjusted associations. In addition to the selected variable (e.g., potential predictor), one set of models included only age, sex, and race/ethnicity. The second set included these variables plus education, main source of income, health-related quality of life (HUI), and health insurance status (U.S. only).
To assess the sensitivity of estimates to method of health-status measurement and to avoid multicollinearity, alternative models replaced the HUI with self-rated general health, chronic condition index score, and activity limitations due to pain. Because replacing the HUI with each of these variables, alone and in combination, did not materially change the odds ratios, these effect estimates are not presented.
Human subjects
The JCUSH design, questionnaires, and informed consent, interview and all other survey-related protocols were reviewed and approved by the Institutional Review Boards of Statistics Canada and the United States National Center for Health Statistics [19]. Because all direct identifiers, plus any characteristics that could possibly lead to the identity of any individual respondent, are removed from the public-use data files, which can only be used for statistical research and data analysis, and cannot be linked to any other individually identifiable data, the project qualified for exemption from coverage by the Human Subjects regulations.