Measuring chiropractic use
To identify visits to chiropractors in the CMS Medicare claims, we examined two sources of information. The first involved
Current Procedural Terminology (CPT) codes for the United States [
15]. For the entire period under study, the CPT code that was supposed to be used for all subluxation procedures performed by chiropractors was A2000 ("manual manipulation of the spine to correct a subluxation"). For the 4,310 AHEAD subjects, we found 13,340 line entries containing the A2000 CPT code over the four-year period. Our second source of information was the specialty type code associated with the Unique Physician Identifier Number (UPIN) in the United States. We found that 18,016 line entries contained UPIN specialty codes for chiropractors. When cross-classified, only 284 (2%) of the A2000 CPT code entries were not associated with a chiropractor's UPIN. The majority of these (86%) were associated with ambulatory surgical centers (specialty code 49). Because our focus is on the use of chiropractors, we relied solely on the UPIN specialty codes for chiropractors at the line level in the CMS Medicare claims. We then aggregated up from the line level, defining any visit that included a chiropractic line charge as a visit to a chiropractor. This approach is consistent with CMS Medicare policy in the United States (Title 42, Part 410; 51 FR 41339, 64 FR 59439, and 66 FR 55328).
Covariates
To model the use of chiropractic among older adults, we chose variables traditionally used in studying the demand for health care [
16], namely sociodemographics, socioeconomics, lifestyle, disease history, functional health status, prior health services use, and the supply of providers in the county. All of these data were obtained from the baseline interviews, except for the supply of chiropractors in the county, which was taken from archival sources. Sociodemographic characteristics included age, sex, race, and living arrangements. Given the potential for nonlinear age effects, we used a set of four dummy variables, contrasting those aged 75–79 years old, 80–84 years old, and 85 years old or older with those aged 70–74 years old (the reference group). Sex was a simple contrast of men (coded 1) vs. women (coded 0). Race was measured with a set of three dummy variables contrasting African Americans and Hispanics with Whites (the reference group). Living arrangements were reflected by a marker coded 1 for living alone vs. 0 for living with others.
Socioeconomic status was measured by education, income, veteran status, and having private insurance. Given the age of our subjects, education was coded as a set of dummy variables contrasting only having attended grade school or having some college, with having a high school education (the reference group). Household income was measured with a set of five dummy variables reflecting income quintiles, with the middle one as the reference group. Veteran status was coded 1 for veterans and 0 for nonveterans. We included it because veterans in the United States have access to the Veterans Health Administration (VHA) in addition to Medicare. Having private health insurance, in addition to Medicare coverage, was coded 1 for yes and 0 for no. We included it because those with access to private health insurance might have their chiropractic visits paid for this way rather than from Medicare.
Lifestyle was measured by cigarette smoking, alcohol consumption, their interaction, body mass, and ever having had a valid motor vehicle (driver's) license. Both cigarette smoking and alcohol consumption may be considered coping mechanisms, and thus are quite relevant to the use of chiropractic, which is commonly used in response to pain. Each of these substance use measures were coded 1 if the subject had ever smoked cigarettes or drank alcohol, and 0 if not. We also included the interaction between these two measures (smoking and drinking) to determine whether there was a synergistic effect of such substance use on the demand for chiropractic. Body mass was measured using a set of dummy variables contrasting being overweight or obese with being of normal or underweight status (the pooled reference group) based on established body mass index (BMI) cut-offs. Driving status was a binary variable contrasting those who had never had a valid driver's license (coded 1) with those who at one time had had one (coded 0), because many members of this cohort in the United States never did.
Disease history was obtained by asking each respondent whether they had ever been told by a physician that they had arthritis (affirmative responses mostly reflected osteoarthritis), cancer (excluding minor skin cancer), diabetes, hypertension, lung disease (affirmative responses mostly reflected chronic obstructive pulmonary disease), a heart condition (affirmative responses mostly reflected congestive heart failure or a myocardial infarction), a hip fracture, or a psychological condition (including emotional, nervous, or psychiatric problems). Subjects were also asked if they were often bothered by pain. Each of these was reflected in a binary marker coded 1 for yes and 0 for no. In addition, we included a set of dummy variables to capture the extent of comorbidity, by contrasting having none or two or more of the above diseases vs. having only one (the reference category).
Functional limitations were measured in numerous ways. The first three were simple counts (0–5) of whether the subject reported having any difficulty in performing activities of daily living (ADLs) such as bathing or dressing, performing instrumental ADLs (IADLs) such as money management or taking their medications, or lower body limitations such as stooping, kneeling, or crouching. The next four measures of functional limitations involved binary markers for whether the subject reported fair or poor (as opposed to excellent, very good, or good) responses to questions assessing their hearing, vision, and memory acuity, as well as their overall health. A binary marker was used to reflect whether the subject currently drove a motor vehicle. We also used two multiple item scales to tap depressive symptoms and cognitive function. For depressive symptoms, we used the sum of eight common depressive symptoms taken from the well-established Centers for Epidemiologic Studies Depression (CES-D) scale [
17]. These sums were then recoded into a set of dummy variables contrasting having no or three or more symptoms with having 1–2 symptoms (the reference group). For cognitive status, we used the well-established Telephone Interview for Cognitive Status (TICS-7) battery [
18]. The TICS-7 score was than recoded into a set of dummy variables contrasting 0–10 (low performance) and 14–15 (high performance) with normal performance (11–13) as the reference group.
The two final categories of covariates were the use of health services, and the supply of chiropractors in the community. There were two measures of self-reported health services use – the number of physician visits in the year prior to baseline, and whether or not the subject had continuity of care. The latter was defined as having no more than 8 months between visits to the same physician during the two years prior to the baseline interview [
13]. The supply of chiropractors was taken from a well-established archival data source for area (geo-political) markers in the United States, known as the
Area Resource File. The supply of chiropractors per thousand persons in the county was coded into tertiles, with the middle tertile used as the reference group.