None of the characteristics examined in our study explain most of the variation among counties in prevalence of prescribed opioids. Taken together, those included in the model explain only a third of the observed variation (model r2
=0.33). This is consistent with findings of other studies. For example, Zhang, Baicker and Newhouse find that variation in total spending by Medicaid patients for all drugs (not just opioids) is not driven by patient characteristics, individual health status, or insurance coverage.51
Caudill-Slosberg, Schwartz and Woloshin report that analgesic prescriptions increased between 1980 and 2000 but office visits in the U.S. for musculoskeletal pain did not.2
The primary drivers of variation in opioid prescribing must therefore be found elsewhere. In their report on state differences in prescribing, Zerzan hypothesizes that variation may stem from differences in prescriber habits, state policies regarding pain management (including utilization management procedures), policies for curbing drug diversion and abuse, preferred drug lists, and regionally specific marketing by pharmaceutical manufacturers.50
However, given the substantial variation seen within states and the lack of correlation with the existence of state prescription drug monitoring programs, it is unlikely that differences in state policies or insurers’ utilization management procedures account for much (if any) of the variation.
Several shortcomings of our data may affect our ability to account more fully for observed variation among counties. (1) The sample excludes some large mass merchandisers, which may contribute non-sampling errors because non-reporting pharmacies may have significantly different characteristics and dispensing patterns. We believe that this exclusion affects the estimates by no more than a few percent, however. (2) The data exclude clinics from which doctors dispense medications directly to patients. Pain clinics have become the focus of law enforcement efforts to battle opioid diversion. Florida has a large proportion of the nation’s dispensing doctors,17
and their exclusion of such prescriptions from our data may result in underestimating opioids dispensed there and perhaps elsewhere as well. (3) More generally, we do not include information about physicians other than their declared specialty. We do not know if they practice alone or in groups, or the extent to which managed care constraints affect their prescribing. (4) Direct measures of morbidity in the resident populations were not available. (5) Our analysis examines characteristics of resident populations in counties where prescribers practice, and not where patients live, as we lack information about patients’ residence. An undetermined number of patients obtained prescriptions from prescribers outside the county in which they lived. In a significant number of counties, there were no opioid prescribers during 2008 in our dataset; patients living in those counties may have traveled outside their county of residence to obtain prescriptions. (6) We do not include information about physician’s prescribing rates. The existence of high-rate prescribers in certain counties --who may draw disproportionate numbers of patients from other counties--may explain some of the residual variation not accounted for in the model.
Our finding of large residual variation is consistent with the majority of studies of geographic variation in medical practices, which report that between one-half to three-fourths of variation remains unaccounted for.9
How much variation is “too much” is also unsettled because at least some variation due to chance is to be expected.13
What may be special about opioid prescribing, however, is that the range of variation we find is wider than ranges reported for other types of medications and medical practices. We find COVs range from 1.06 for codeine to 4.21 for fentanyl, and 1.05 for all opioids combined. In contrast, among elderly Medicare beneficiaries nationwide during 2007, the COV for drug spending per beneficiary at the hospital referral region level was 0.08.52
The COV for total state level health care spending per capita in the U.S. during 2004 was 0.123 and 0.11 for Medicare spending per beneficiary.9, 17
The surgical procedure found to vary the most among the 306 hospital referral regions in the U.S.—lumbar fusion for Medicare enrollees—had a COV of 0.5.44
The existence of wide variation in practitioners’ prescribing practices is not surprising, given the recent shifts in the normative environment surrounding the use of prescription opioids. For most of the 20th Century, the norms that guided professional conduct supported a conservative approach to pain management. The Harrison Act of 1914, subsequent regulations, and prosecutions of physicians created a general reluctance to use narcotics to treat pain, especially neuropathic and chronic non-cancer pain.21, 38
Fear of investigation by authorities for excessive prescribing was commonplace among physicians.45
In many states, medical boards voiced recommendations and statements that discouraged using opioids for pain management, and state laws in some jurisdictions added additional restrictions.21
When opioids were used for pain relief, the most commonly prescribed were weaker ones—codeine and meperidine.27
Pain management was rarely taught in medical schools, and surveys of physicians’ knowledge of pain management principles found significant deficits.22,48
In the 1990s, the normative environment began to change. In 1998, the Federation of State Medical Boards promulgated model guidelines that authorized flexibility in pain management, including more expansive use of opioids,14
which was reinforced by policy statements in 2004,15
although some observers noted that medical boards in several states were slow to revise their policies.21
To promote more expansive treatment of pain in hospitals, the Joint Commission on the Accreditation of Healthcare Organizations proposed in 2001 that pain be considered the “fifth vital sign.”37
Some argued that the under-treatment of pain constituted a “prominent public health concern.”18
These developments reinforced a growing consensus that opioid therapy is appropriate not only for acute pain, cancer pain, and palliative care, but also for some patients with chronic non-cancer, and that potent opioids are appropriate. Chronic pain was being redefined as a disease deserving of treatment.1, 19
By 2008, the period studied here, the norms that had guided professional medical practice for decades had changed but had not been replaced by a new consensus. The analgesic effectiveness of short-term opioid treatment for severe pain has been well established and widely accepted, but the evidence base for long-term opioid treatment remains fragmentary to this day. Scientific studies have demonstrated that various chronic pain conditions, including neuropathic pain conditions, are opioid responsive, but the extent to which analgesic efficacy is maintained over the long term remains an open question. The evidence basis is also weak for making clinical decisions about other benefits of long term treatment, such as quality of life and functional improvements; the risks of misuse and iatrogenic addiction; optimal approaches to initiating, titrating, escalating, monitoring and terminating opioid therapy; the utility of opioid rotation; and the utility of informed consent and opioid management plans, among other matters. A review commissioned by the American Pain Society and the American Academy of Pain Medicine concluded in 2009 that “clinical decisions regarding the use of opioids for chronic non-cancer pain need to be made based on weak evidence."5
Various professional medical bodies have sought to revise and clarify guidelines for prescribing them for this purpose.6, 30, 49
Moreover, a tide of abuse of these potent opioids has arisen, with increasing overdose, accident and death rates, spurring a new public health crisis.12
Despite studies finding no evidence that increasing medical use of opioids is correlated with increases in health consequences of abuse,20
more recent research does find a direct association between rates of therapeutic use and of abuse.7
The extent to which these changes in professional practice norms contribute to wide geographic variation in opioid prescribing is difficult to determine with precision. Because no earlier published studies of county variation exist, we are unable to assess the extent to which regional variation changed as opioid prescribing became more prevalent and as practice norms became unsettled. It is likely, however, that the variation was significantly wider in 2008 than it was a decade earlier. One study of state-level variation in prescribing to Medicaid recipients reported that variation in opioid prescribing increased from 1996 through 2002.50
The introduction of extended release oxycodone in 1996 and subsequent efforts to market it no doubt contributed to greater variability in prescribing practice. Between 1997 and 2006, amounts of oxycodone dispensed from retail pharmacies in the U.S. increased by 678%, while the U.S. population increased only 11%.40
To market its extended release oxycodone, the pharmaceutical manufacturer mined prescription data to identify physicians who prescribed opioids with greater than average frequency (especially primary care physicians) and focused its sales efforts on them and the regions where they lived.41
Sales of this one drug alone increased from $48 million in 1996 to $1.1 billion in 2000. Our data show that by 2008, geographic variation in prescribing was greater for oxycodone than for the other most frequently prescribed opioids ().
But the aggressive marketing of extended release oxycodone cannot explain the overall increase in opioid prescribing and the widespread variation in practice. Sales of other opioids increased dramatically as well during this period: the amounts of methadone, fentanyl and hydrocodone dispensed from retail pharmacies in the U.S. grew by 1,053%, 591%, and 336%, respectively.40
In 2008, variation among counties in prevalence of prescribing these drugs was not as wide as for oxycodone but was still substantial ().
In summary, residual geographic variation in opioid prescribing may be explained by a number of dynamics, which are not exclusive of one another. One is that the collapse of an earlier consensus about using opioids for pain management resulted in individual prescribers setting their own practice pattern that reflects their own personal attitudes, values, knowledge, and other interests. Another is that practice variation may stem from different local medical subcultures—i.e., shared attitudes about what constitutes appropriate practice, which may be reinforced by policies and attitudes of licensing boards and other local regulators. Regional differences in how medical practice is organized and in the prevalence of managed care may explain some of the residual geographic variation, as there is evidence that patients with private HMO insurance are less likely to get opioids from their primary care providers in office visits.26
Patients’ demands for care that result in prescribing opioids--because of differences in injuries or other reasons we are unable to measure or that are not correlated with measured characteristics--may also vary regionally. Finally, geographic variation may reflect different rates of diverting opioids for non-medical use. Surveys find increasing numbers of U.S. residents reporting non-medical use of prescription drugs, including opioids,34
which has large economic, social, and health costs,3, 8, 27, 28, 35, 36
and studies have found correlations between amounts prescribed and various indicators of misuse, such as fatal opioid overdoses.29
Most opioids are prescribed and dispensed for legitimate medical reasons,7
so county differences in diversion rates probably contribute little to the observed variation in opioid prescribing.
More systematic attention to geographic variation in opioid prescribing is needed because the wide range of variation and the amount of variation that is not explained raise questions about the appropriateness and effectiveness of practice. The growing epidemic of opioid diversion and abuse has spurred consideration of a variety of legal and administrative controls,25
including even taking certain often-abused drugs off state formularies.33
Such drastic interventions are unlikely to be implemented, as they will make these drugs unavailable for legitimate medical use. Better understanding of effective opioid use and of prescribing practices would inform design of regulations so that they strike the right balance between the benefits and risks of opioids.