The 4P Study was conducted from July 2002 to March 2003 at 12 academic medical centers in the United States. Patients were recruited from the primary care practices of each center, including both resident and attending physician practices. The Institutional Review Boards at all centers approved the study protocol.
Patient Selection and Recruitment
Patients were eligible for enrollment if they were at least 18 years of age and reported a history of pain for at least 3 consecutive months. We excluded pregnant females, patients not fluent in English or Spanish, incarcerated individuals, patients being treated for active cancer, and patients with documented cognitive impairment such as dementia.
During the course of routine clinic check-in procedures, the research assistant, triage nurse, or treating physician identified eligible patients. The research assistant then evaluated patients for exclusion criteria, obtained informed consent, and administered the 82-item survey verbally in a secluded location.
The initial patient survey included questions about demographic information, general health and pain perceptions, treatment choices, and attitudes toward provision of medical care. The survey also included the numeric rating scale (NRS-11),16
a standardized pain assessment tool used to measure pain intensity, where pain scores range from 0 (no pain) to 10 (unbearable pain). Patients were shown the NRS-11 and responses to the question “How severe is your pain today?” were recorded.
Pain Medication Use
Patients were specifically asked the types of medicines that they were currently taking for their pain and whether by prescription, over the counter, or by illicit means. The research assistants were instructed to ask patients, “What medicines are you taking now for your pain?” The research assistant provided each patient with a complete list of examples from each class: (1) nonsteroidal anti-inflammatories (NSAIDS) such as Motrin, Advil, Celebrex, Goody's Powders, and Aspirin, (2) Acetaminophen or Tylenol, (3) weak opioids such as Darvocet, Tylenol III, and Vicodin, (4) strong opioids such as Demerol, Morphine, Percocet, Oxycodone, and Dilaudid, (5) long-acting opioids such as Oxycontin, MS Contin, and Duragesic patches, (6) adjuvant medications such as Elavil or Gabapentin, (7) the “atypical” analgesic Ultram, and (8) substance-abuse “medications” such as alcohol, marijuana, heroin, and cocaine. Patients could choose multiple classes, and for purposes of analysis we considered patients to be taking opioid medications if they were taking Schedule II, III, and IV drugs as defined by the Drug Enforcement Agency (classes 3, 4, and 5 above).
Patients were asked to self-report their racial status from a list of 16 groups. The 3 major groups were classified as (1) white, (2) black, African American or Negro, or (3) Spanish, Hispanic, or Latino. Because of insufficient numbers, Hispanic patients and other races such as Indian, Chinese, Filipino, Japanese, or Korean were excluded from this analysis.
As a measure of other treatment choices, patients were asked whether they had previously seen different types of specialists or nonprimary care providers for their pain such as a rheumatologist, orthopedist, pain specialist, chiropractor, podiatrist, reflexologist, acupuncturist, or herbalist. Patients were also asked whether they had undergone physical therapy as a modality to improve their pain.
Other medical conditions reported by patients included any of 15 common chronic diseases including diabetes, hypertension, arthritis, stroke, or emphysema. The type of health insurance, the current number of hours per week the patient worked, whether the patient received a disability check, annual income, and the years of schooling completed were also documented as elements of socioeconomic status. The research assistant also documented whether a resident or attending physician was the primary care provider for the patient on the study day.
Statistical analyses were performed using Stata 7.0 (Stata Corp., College Station, Tex). Descriptive statistics for the demographic variables, the use of opioids, the different treatment modalities, and other covariates as described above were performed. Race was analyzed as black versus white, physician level as resident versus attending, and the presence of comorbidities, insurance status, and disability check status were dichotomized as yes or no. Age, duration of pain, educational level, and number of current hours worked were analyzed as continuous variables.
The primary outcome variable was any opioid use by the patient. We also performed additional analysis considering the 3 classes of opioids (weak, strong, and long acting). In these analyses, persons reporting the use of more than one class of opioid were classified hierarchically as follows: long acting, strong, weak, and none. Thus, a person reporting both long-acting and weak opioid use would be classified in the long-acting group. Four additional outcome variables included patients' use of other treatment modalities, including (1) NSAIDS, (2) physical therapy, (3) pain specialists, and (4) other specialists. The use of orthopedists, rheumatologists, chiropractors, podiatrists, reflexologists, acupuncturists, and herbalists were all grouped into the category of other specialists because of the low overall percentage of referrals for the last five groups.
We performed bivariate analyses of race versus other demographic data, level and duration of pain, comorbidities, and other socioeconomic-related covariates. We then compared race with the prevalence of opioid use and the other 4 treatment choices. We applied 2-sample t tests, χ2, and Wilcoxon rank sum tests as appropriate. Variables that were significant (P<.05) in bivariate analysis and variables of particular interest that might predict opioid use or treatment decisions were included in the final regression models. We used multiple logistic regression to assess the relationship between race and each outcome while adjusting for potential confounding by age, sex, pain severity and duration, comorbidities, disability status, and elements of socioeconomic status. Additionally, in order to examine any effect of physician experience on race and opioid utilization, we ran separate models stratified by physician level of training. We used 2-sided P values of less than .05 to indicate statistical significance. In the logistic regression analysis, robust variance estimates were used to account for clustering by site.