Secondary Logo

Journal Logo

Original Articles

Possible Opioid Shopping and its Correlates

Walker, Alexander M. MD, DrPH*; Weatherby, Lisa B. MS*; Cepeda, M. Soledad MD, PhD; Bradford, Daniel MS; Yuan, Yingli PhD

Author Information
doi: 10.1097/AJP.0000000000000483



When the term is applied to drugs with potential for abuse, “doctor shopping” refers to a patient’s practice of seeking prescriptions from multiple prescribers without their coordination or knowledge.1,2 It is not possible to assess patient intent or physician knowledge from insurance databases, so we use the term “possible shopping behavior” as general term to describe patterns of prescription fills that might suggest doctor shopping.

Previous work has shown that a useful definition of possible shopping behavior could incorporate the number of overlapping prescriptions written by different prescribers and the number of pharmacies that fill those prescriptions.3 An “overlap” here means that there is a new prescription fill before the expiry of a previous fill, calculated as a function of the earlier dispensing date and the nominal number of days of drug supplied. A prescriber-pharmacy-overlap definition of shopping clearly differentiated opioids from diuretics, which were taken as a negative control.3

Possible shopping behavior seems to be much less common in patients 65 years or older than in younger patients.4,5 Other documented correlates of possible shopping include concurrent dispensing of benzodiazepines as well as insurance claims diagnoses for mood disorders, back pain, and abuse of nonopioid drugs. Malignancy-related pain diagnoses seem to be negatively correlated with possible shopping behavior.5 Possible shoppers have been much more likely than others to pay fully on their own for prescription opioids, rather than go through their insurance plans.4,6

Circumstantial evidence using unvalidated measures suggests that opioid shopping may be associated with abuse. In persons who initiated oxycodone or tapentadol and whose insurance claims histories were recorded in the IMS PharMetrics Plus database, the occurrence of overlapping fills of prescriptions from ≥2 doctors at ≥3 pharmacies over 1 year was associated with 6.5% period prevalence of International Classification of Diseases 9th Revision claims for abuse, as opposed to only 0.7% in persons without such overlapping fills.5 In an Arkansas Medicaid population receiving chronic opioid therapy, increasing levels of a misuse score that incorporated the number of prescribers and the excess days supply were monotonically related to increasing prevalence of diagnostic codes for opioid abuse.7 After the introduction of a tamper-resistant formulation of extended-release (ER) oxycodone, overlapping prescriptions from ≥2 doctors and ≥3 pharmacies declined by 50% in national dispensing data from IMS Health.8 The same phenomenon of reduced occurrence of overlapping prescriptions occurred even more dramatically for ER oxymorphone after its reformulation.9 The decline in overlapping prescriptions from different prescribers and filled at pharmacies following the introduction of tamper-resistant formulations suggests that these formulations have been less appealing for abuse.

In the work reported here, we have examined indicators of possible opioid shopping in a large insurance claims database that was supplemented by a pharmacy dispensing database that captured out-of-plan dispensing, including self-paid prescription fills. We have derived a measure that strongly discriminates courses of opioid treatment from courses of diuretics (a negative control for shopping) over an 18-month observation period, and we have examined the correlation between possible shopping behavior and characteristics that could be ascertained in the claims data.

The work was carried out under a US Food and Drug Administration (FDA) postmarketing requirement, to manufacturers of ER and long-acting (LA) opioids, who supported the investigation reported here.10 The protocol and analysis plan were agreed with the FDA before variables comprising shopping behavior were extracted from the insurance data files. The postmarketing requirement includes studies involving chart review and self-reported abuse and misuse in relation to possible shopping behavior. Those studies are underway (see NCT02667158 and NCT02667210). The sponsor and the FDA had the opportunity to comment on this manuscript. The authors retained the final control over wording.



This is a cross-sectional study in US commercial insurance claims data. The protocol called for defining and investigating the period prevalence of possible opioid shopping behavior over an 18-month observation period beginning with the first opioid dispensing in 2012 in relation to the period prevalence of characteristics of patients, prescribers, and dispensings. All patients meeting eligibility criteria were included.

Commercial Insurance Data

When accessed for this study, the IMS Health PharMetrics Plus database held pharmacy, provider, and facility claims for approximately 75 million persons enrolled in US health plans. Ninety-seven percent of the patients were commercially insured and almost all received insurance as an employee or a dependent of an employee of a firm or institution through whom the insurance was obtained. The health plans included in PharMetrics Plus had 35% of their membership from the South, 27% from the Midwest, 23% from the East, and 13% from the West. PharMetrics Plus also contained hospital discharge data and demographic information. About 54% of the members enrolled in plans contributing data to PharMetrics Plus in 2012 had mental health coverage.

National Dispensing Data

The IMS Health LRx database had information on 234 million patients through 2013 and at that time covered 86% of all retail dispensing in the United States, regardless of payment method (insurance or self-pay). LRx includes both in-person and mail order dispensing from pharmacy chains, food stores, mass merchandisers, and many independent pharmacies. The stores not included in LRx include small regional chains and independent pharmacies. Dispensing from nonretail sources, such as hospital and military pharmacies, are not well captured and were not included. We use the collective term “pharmacies” to denote all kinds of pharmacy outlets.

IMS Health had information on membership in shared medical practices for about 45% of prescribers. To avoid counting prescriptions from the same medical practice as having a different origin, we used practice as the unit of prescriber aggregation. We treated providers whose practice membership was unknown as if each provider were a single practice.

There were approximately 40 million persons in LRx who filled a prescription for opioids in 2012, of whom about 11% (4.2 million persons) were members of PharMetrics Plus in that year.

State Prescription Monitoring Programs

Each person in PharMetrics Plus had a known state of residence. To permit examination of the characteristics of state prescription drug monitoring programs (PDMPs) in relation to Shopping Behavior, Dr Peter Kreiner of Brandeis University assembled and provided characteristics of state prescription monitoring programs (PDMPs).11 He determined for each state, whether in 2012 the prescriber or the dispensing pharmacy had to be registered with the PDMP, whether the prescriber or the dispensing pharmacy was required to check with the PDMP before prescribing or dispensing, whether the PDMP used probabilistic or exact matches to link patients (known for 2014 only), and whether the PDMP was online in 2011.

Study Population

The study population consisted of persons aged 18 and over in the PharMetrics Plus database for whom a link could be established with IMS LRx and who had an immediate release (IR) or ER/LA opioid or a diuretic dispensing in 2012 plus another of the same within 18 months (548 d). The study population was further restricted to those for whom all opioid and diuretic dispensings that appeared in PharMetrics Plus were also found in LRx with no disagreement as to age or sex between the data sources, who were not receiving long-term care, who had mental health coverage, and who had at least 12 months of observation in LRx before their first opioid (or diuretic) dispensing in 2012 and at least 18 months observation afterward in both LRx and PharMetrics Plus, unless it appeared that their follow-up had been truncated by death.

The steps in cohort creation are shown in Table 1. There were 164,923 recipients of opioids who met these criteria (96% under the age of 65) and 99,281 recipients of diuretics. A total of 12,842 persons filled prescriptions for both opioids and diuretics, and contributed to both groups.

Steps in Creation of the Study Populations

Fills Evaluated

We explored measures of possible shopping behavior based on 1- and 10-day overlaps between successive fills, calculated for courses of both opioids and diuretics, and we examined the number of fills involved in overlap episodes. We also examined measures with no requirement for overlap, based on total usage of pharmacies and practices during the 18-month observation period. In those persons who contributed both opioid and diuretic treatment courses to the analysis, the evaluations of fills for opioids and diuretics were performed separately.

Tabulations of Possible Shopping Patterns

We formed exhaustive cross-classifications of possible shopping patterns, and collapsed adjacent categories (those sharing identical levels of 2 characteristics and differing by only 1 step in the level of the third characteristic) with an eye to medical plausibility to create 4 levels. We selected the cut-points between categories with the aims of preserving any natural clumping of values. We labeled the resulting categories of possible shopping behavior as “None,” “Minimal,” “Moderate,” or “Extensive.” Among persons who did not fall into the category of “None,” we also aimed to achieve a relative distribution of roughly 60%, 30%, and 10% for the categories “Minimal,” “Moderate,” or “Extensive.”

Characteristics of Drugs and of Patients

Appendix A (Supplemental Digital Content 1, lists the generic drug names in categories used to summarize drugs other than opioids and diuretics, prescriber specialty, and patient diagnoses. The major pain categories used in the analysis were arthritis, back pain, headaches, malignancy, musculoskeletal pain, neuropathies, and wounds/injury. Other diagnoses were grouped according to chapter headings of the International Classification of Diseases 9th Revision.

Descriptive Analysis

The distributions of opioid and of diuretic treatments across categories defined by numbers of prescribing practices, numbers of dispensing pharmacies, and numbers of prescription fills were compared to find categories that discriminated well between opioids and diuretics.

Discrimination Between Courses of Opioids and of Diuretics

To quantify the overall performance of the categories in the planned discrimination between opioid and diuretic use, a logistic regression was fit with opioid versus diuretic use as the dependent variable, and categorical levels of possible shopping behavior taken as the predictor. Some patients were users of both opioid and diuretics. The dispensing patterns for each were characterized and evaluated separately, and there was no need to account in the analysis for the fact that 2 sequences of fills (opioids and diuretics) might come from the same person. The c-statistic and the regression coefficients associated with various categories of possible shopping (vs. “None”) were taken as measures of discrimination. After categories were created, diuretic use did not enter further into any analysis.

Correlates of Possible Shopping Behavior

To examine the extent to which patient characteristics that could be derived from the insurance claims data were associated with levels of possible shopping behavior, this characteristic was reduced to a dichotomous outcome (Moderate-Extensive vs. None-Minimal) and taken as the dependent variable, against which all the insurance claims–derived patient characteristics were entered into a stepwise logistic regression, with the P-value for retention set at 0.01. The c-statistic summarized the overall prediction.


Prescribers, Pharmacies, and Fills

Definitions that involved ascertainment of overlaps of 1 or 10 days between successive fills captured relatively small numbers of patients and carried lower c-statistics for discrimination between opioids and diuretics than measures based on all fills in the 18-month observation period (data not shown). The number of fills was not a useful discriminator between opioids and diuretics for any degree of overlap, as multiple fills predominated among diuretic users.

Table 2 displays the categories of possible shopping behavior assigned as a cross-classification by numbers of pharmacies and numbers of practices used over the 18-month observation period. Table 2 also gives the number of individuals falling at each cross-classified level of pharmacy and practice counts. “Moderate” and “Extensive” levels together can be seen to be defined by having obtained prescriptions from ≥3 practices and filling them at ≥3 pharmacies during 18 months.

Categories of Possible Shopping and Numbers of Individuals Classified According to Number of Dispensing Pharmacies and Number of Prescribing Practices for Opioids Over 18 Months

Appendix B Table 3 (Supplemental Digital Content 1,, shows the coefficients and the c-statistic of the logistic regression that quantifies how levels of possible shopping behavior are associated with opioid versus of diuretic fills. There is a steady progression in discrimination moving through the higher levels. Each level above “None” has a higher relative likelihood of opioid use versus diuretic use. At the highest level, “Extensive,” the odds ratio for an opioid course of fills rather than a diuretic course is over 30-fold.

Geographic Region

The distribution of levels of possible shopping behavior by state is given in Appendix B Table 2 (Supplemental Digital Content 1, Among states with at least 500 individuals represented, the range in prevalence of “Extensive” ran from 13.1% in New Hampshire to 0.5% in Arkansas.

Table 3 shows the relation between the PDMP characteristics of the states and the prevalence of different levels of multiple prescribing practices and multiple pharmacies. Each characteristic was associated with a modest reduction in the prevalence of persons at the “Extensive” level. The range in the proportion of persons classed as “Extensive” at different levels of PDMP characteristics (1.2% to 1.6%) is much smaller than the range across states.

Distribution of Study Participants by Shopping Behavior and Characteristics of the PDMP Program in the US State of Residence

Personal Characteristics in Multivariate Analysis

Patient characteristics as identified in the drug dispensing and medical claims records are listed in Appendix B Table 2 (Supplemental Digital Content 1, cross-tabulated by levels of possible shopping behavior. Appendix B Table 2 (Supplemental Digital Content 1, also gives the P-value associated with each retained predictor from a stepwise multivariable logistic regression collapsed to Moderate-Extensive versus None-Minimal. A selection of these variables is shown in Table 4.

Selected Dispensing Characteristics in Relation to Possible Shopping Behavior (Excepted From Appendix B Table 2, Supplemental Digital Content 1,

All the results that were retained as significant in the stepwise regression and that are described below show the associations as described, even after adjustment for the effects of one another in the multiple regression analysis.


The proportion of persons with “Extensive” use of multiple prescribing practices and multiple pharmacies declined with age; persons aged 65 years and older and especially persons aged 75 and older exhibited very little such behavior. Men and women showed similar levels of possible shopping behavior, women being higher in the crude tabulations and men higher by 11% in the adjusted estimates. State of residence was an important predictor of Shopping Behavior, as indicated in Appendix B Table 1 (Supplemental Digital Content 1,

Types of Opioid Dispensed

Among persons dispensed both IR and ER/LA opioids during the observation period over 9% exhibited extensive possible shopping behavior, as opposed to about 1% of persons receiving either ER/LA or IR exclusively.

Other Drugs Dispensed

For all the drug groups examined (antidepressants, anxiolytics, antipsychotics, anxiolytics, hypnotics, and psychostimulants), higher numbers of fills, both during the observation period and in the year before, were associated with progressively higher proportions of the population exhibiting moderate and extensive possible shopping behavior. Drug dispensing in the 365 days before the observation period regularly dropped from the multivariable regression or, if they remained, had a significance level far less than the corresponding significance for the fills of the same agents during the observation period. An exception to this pattern was psychostimulants, for which fills before the observation period were as predictive of the “Extensive” level as were fills during the observation period.

The relation between the number of prescription fills and proportion of patients with possible shopping behavior was essentially monotonic for all drugs examined except for antipsychotics. For these agents, all nonzero levels of dispensing, both during and before the observation period, defined populations with higher levels of possible shopping behavior.

Diagnostic Groups

All persons with insurance claims noting one of the predefined diagnostic groups had modestly higher proportions of extensive possible shoppers than the overall average, generally on the order of 2%. Most of the associations between shopping and diagnosis were retained in the multivariable regressions. None of the diagnosis categories had a dramatically higher than average proportion of moderate-to-extensive possible shopping behavior. The highest proportion was among those with codes for wounds or injury, at 8.7%.

Dispensing Characteristics

The number of nonspecialist fills, the number of self-paid fills, and the total morphine milligram equivalents (MEQ) dispensed were all highly associated with possible shopping behavior. Larger numbers of self-paid fills were the most strongly associated with possible shopping behavior. Altogether a third of extensive possible shoppers were persons with self-paid fills. Persons with >5000 MEQ dispensed over 18 months accounted for half of all extensive shoppers.

The c-statistic (0.82) associated with the logistic regression indicated that possible shopping behavior was strongly associated with the other patient characteristics listed in Appendix B Table 3 (Supplemental Digital Content 1,

Sensitivity Analysis in Persons With at Least 10 Opioid Dispensings

Persons with few opioid dispensings cannot qualify at the highest levels of multiple practices and multiple pharmacies. To check whether the results that obtained for the full population could be seen among relatively frequent users of opioids, we repeated the full analysis laid out in Appendix B Table 2 (Supplemental Digital Content 1,, including only the 29,960 persons with at least 10 opioid dispensings during the 18-month observation period. Correlates of possible shopping behavior at the level of P<0.0001 for frequent users of opioids in the logistic regression were: age; US state of residence; opioid type (IR only, ER/LA only, both); use during the observation period of antipsychotics, anxiolytics, and hypnotics; use before the observation period of antidepressants and psychostimulants; number of opioid prescriptions from nonspecialists; self-paid dispensings of opioids; total MEQ of opioid dispensed; and diagnoses of fractures, musculoskeletal disorders, wounds and injuries, and injury/poisoning. A logistic regression analysis in the full population with statistical control for number of opioid dispensings as a covariate again yielded a pattern of associations very similar to that of the main analysis.


Possible shopping behavior in 2012 to 2014 proved similar in this large national database to patterns of use of multiple prescribers and multiple pharmacies identified in previous work.1–6 Considering that many of the patients in this cohort may have experienced chronic pain and might consult multiple providers as part of their care, the overall prevalence of moderate-to-extensive levels of possible shopping behavior at 5.1% may not seem especially high. However, variations in the prevalence of possible shopping between people with different characteristics suggest that nonmedical factors may also be at play.

Use of multiple prescribers and multiple pharmacies varied dramatically across the US states. The single highest prevalence of moderate-extensive levels of this behavior was among persons with ≥2 self-paid fills for opioid prescriptions. A further important prescription characteristic that correlated with levels of possible shopping behavior was the total amount of opioid dispensed. The small fraction of the population with both ER/LA and IR use had a relatively high prevalence of moderate and extensive levels. Nonetheless, as almost all opioid use was among people who received only IR opioids, most high-level possible shopping behavior was observed among IR-only recipients.

As the multiplicity criterion that we examined as an outcome is necessarily correlated with the existence of multiple dispensings, it is important to note that all the significant correlations identified are mutually adjusted for one another and adjusted for the total amount of opioid dispensed and the existence of each of a variety of pain diagnoses. The adjustment was carried out in both the main analysis and in the sensitivity analysis restricted to frequent opioid users.

The relation with sex was complex, in that women had slightly higher crude rates of extensive possible shopping behavior, but men had adjusted rates that were 11% higher.

Together with more modest associations with the use of nonspecialist prescribers and with many classes of pain diagnoses and drug dispensings, characteristics identifiable in the insurance data yielded a very high c-statistic of 0.82 in predicting the possible shopping behavior.

There would be many limitations to the use of possible shopping behavior as a stand-in for patients’ possible drug-seeking behavior. Some forms of drug seeking may involve the use a single compliant prescriber or pharmacy (“pill mills”), classifying the individual into the “None” category of any measure based on multiplicity of prescribers and pharmacies. In addition, “pill mills” that required self-payment for all dispensings would not be captured in the national dispensing data source that we used. As fewer than half of the prescribers in this study could be assigned to practices, it is likely that the number of practices is overstated, meaning that we have sometimes counted multiplicity where really there was none. Using diuretics as a nonshopping control might have led to categories or cut-points that distinguish opioid from diuretic recipients, but that do not point to aberrant behaviors. Applied over an 18-month period of observation, for example, the definitions of possible moderate-to-extensive multiplicity might have pulled in people who legitimately had several injuries and who may have used pharmacies situated near to each of a series of different prescribers. Recurrent pain-causing medical events might explain why the diagnostic categories of “wound/injury” and “fracture” were associated with relatively high levels of use of multiple prescribers and multiple pharmacies.

Parente et al12 used expert consensus panels to place the number of opioid prescribers and the number of dispensing pharmacies used by a single patient among criteria for identifying potential abuse in US health insurance data in 2004. Wilsey et al13 cast doubt on the use of such measures as screening tools. They pointed out that use of 2 to 5 prescribers for opioids in California defined a population somewhat different from those who used a single prescriber, but not so different as to indicate that patients were manipulating the health care system. By contrast, Gwira Baumblatt and her coworkers found in Tennessee in 2009 to 2010 that use of 2 to 3 prescribers or 2 to 3 pharmacies raised the risk of death from opioid overdose by about 5 times (estimated from fig. 2 in the Gwira Baumblatt publication) and that filling opioid prescriptions from ≥4 prescribers or at ≥4 pharmacies raised the risk of opioid overdose death by 15- to 20-fold by comparison with the risk associated with obtaining opioid prescriptions from a single prescriber.14

The measure used here incorporated multiple prescribers and multiple pharmacies over 18 months as a combined measure of possible shopping, taking cutoffs for these that discriminated best between opioid usage and diuretic usage. In dropping a requirement for overlap between successive opioid fills, we identified a larger number of shoppers than in our previous work.3–6

Use of multiple prescribers alone has permitted regional correlational analysis in earlier versions of the IMS LRx database. McDonald and Carlson15 developed a mixture model to describe the distribution of the number of prescribing physicians across patients receiving opioids in 2008. A count of 3 component populations fit the data best, and McDonald and Carlson took the inferred population with the highest mean number of prescribers, comprising 0.7% of opioid recipients overall, to be highly enriched for doctor shoppers. Calculating the size of shopping-enriched populations region by region, McDonald and Carlson found a 10-fold variation in prevalence of presumed shopping by state, with a ranking similar to that in Appendix B Table 2 (Supplemental Digital Content 1, Overwhelmingly, the most important county-level determinant of shopping prevalence was the amount of opioid sold, and the number of emergency department physicians per 1000 population was a further positive predictor. The prevalence of poverty was a strong negative predictor of shopping.

Over the period 2000 to 2006 in California, Gilson et al16 found no beneficial impact of PDMP policy changes on multiple-provider episodes of opioid prescription. We did find a benefit of various characteristics that might improve the communication between a PDMP and prescribers or dispensing pharmacies. The effects of PDMP characteristics on the prevalence of shopping were not as strong as other predictors. Because PDMPs are evolving rapidly, their characteristics in 2014 may be only partly informative about programs in the 2012 to 2014 observation period, much less about the current situation. Moreover, any conclusion about the relative impact of PDMP accessibility should be considered tentative considering the selected nature of the current study population and the purely structural features of PDMPs that were analyzed. It is possible that dispensing controls already implemented by insurers during the 2012 to 2014 study period were having an impact that blunted variations in shopping that might otherwise be attributable to state efforts. It is also possible that the >10-fold state-to-state variation in shopping prevalence points to regional differences in the societal context that overwhelmed the variation in direct effects of PDMP programs in those years.

Although the group we reviewed was minimally distinguished from the general population in its use of opioids (2 fills in 18 mo) and widely spread across the United States, it was not a representative sample of Americans. The population studied consisted of persons with commercial health insurance, who would have been mostly employed people and their dependents. We included in the analysis only persons with mental health benefits to capture predictors and correlates of shopping as completely as we could, but the restriction to insured persons whose benefits included mental health coverage may have further narrowed this group to those with “premium” insurance plans.

The unrepresentativeness of the population extends particularly to persons over 65, for whom commercial insurance recipients have a disproportionate number of Federal, state, and military retirees as well as retirees from certain large business. Considering this selection, the low prevalence of shopping in the elderly, though it is consistent with others’ findings, needs to be taken as a tentative finding.4,5

The prevalence of shopping and its strong dependence on medical and demographic patient characteristics in this national insurance sample support the idea that shopping is 1 manifestation of a widespread behavioral pathology that has many observable features. The direction of the causality is more complex than this cross-sectional analysis can tease apart.

The many correlates of possible shopping seen in these data further suggest that we might expand on the insurance-screening agenda proposed by Parente and colleagues more than a decade ago.11 Not just drug usage patterns, but larger complexes of health insurance claims may be useful in identifying persons who need intervention and help. They might also identify both prescribers and pharmacies that would benefit from special interventions to assure safe prescribing and dispensing, either through state PDMP programs or by the actions of insurers.


1. Katz N, Panas L, Kim M, et al. Usefulness of prescription monitoring programs for surveillance—analysis of Schedule II opioid prescription data in Massachusetts,1996-2006. Pharmacoepidemiol Drug Saf. 2010;19:115–123.
2. Peirce GL, Smith MJ, Abate MA, et al. Doctor and pharmacy shopping for controlled substances. Med Care. 2012;50:494–500.
3. Cepeda MS, Fife D, Chow W, et al. Assessing opioid shopping behaviour: a large cohort study from a medication dispensing database in the US. Drug Saf. 2012;35:325–334.
4. Cepeda MS, Fife D, Chow W, et al. Opioid shopping behavior: how often, how soon, which drugs, and what payment method. J Clin Pharmacol. 2013;53:112–117.
5. Cepeda MS, Fife D, Kihm MA, et al. Comparison of the risks of shopping behavior and opioid abuse between tapentadol and oxycodone and association of shopping behavior and opioid abuse. Clin J Pain. 2014;30:1051–1056.
6. Cepeda MS, Fife D, Yuan Y, et al. Distance traveled and frequency of interstate opioid dispensing in opioid shoppers and nonshoppers. J Pain. 2013;14:1158–1161.
7. Sullivan MD, Edlund MJ, Fan MY, et al. Risks for possible and probable opioid misuse among recipients of chronic opioid therapy in commercial and Medicaid insurance plans: the TROUP study. Pain. 2010;150:332–339.
8. Coplan PM, Chilcoat HD, Butler SF, et al. The effect of an abuse-deterrent opioid formulation on opioid abuse-related outcomes in the post-marketing setting. Clin Pharmacol Ther. 2016;100:275–286.
9. Chilcoat HD, Coplan PM, Harikrishnan V, et al. Decreased diversion by doctor-shopping for a reformulated extended release oxycodone product (OxyContin). Drug Alcohol Depend. 2016;165:221–228.
10. US Food and Drug Administration. “Updated ER/LA opioid analgesic postmarketing requirements.” 2016. Available at: Accessed March 25, 2016.
11. Peter Kreiner, personal communication to Soledad Cepeda. February 12, 2015.
12. Parente ST, Kim SS, Finch MD, et al. Identifying controlled substance patterns of utilization requiring evaluation using administrative claims data. Am J Manag Care. 2004;10(pt 1):783–790.
13. Wilsey BL, Fishman SM, Gilson AM, et al. An analysis of the number of multiple prescribers for opioids utilizing data from the California Prescription Monitoring Program. Pharmacoepidemiol Drug Saf. 2011;20:1262–1268.
14. Gwira Baumblatt JA, Wiedeman C, Dunn JR, et al. High-risk use by patients prescribed opioids for pain and its role in overdose deaths. JAMA Intern Med. 2014;174:796–801.
15. McDonald DC, Carlson KE. The ecology of prescription opioid abuse in the USA: geographic variation in patients’ use of multiple prescribers (“doctor shopping”). Pharmacoepidemiol Drug Saf. 2014;23:1258–1267.
16. Gilson AM, Fishman SM, Wilsey BL, et al. Time series analysis of California’s prescription monitoring program: impact on prescribing and multiple provider episodes. J Pain. 2012;13:103–111.

shopping; drug seeking; opioids; health insurance

Supplemental Digital Content

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.