“Doctor and Pharmacy Shopping”: A Fading Signal for Prescription Opioid Use Monitoring?

Abstract

Background:

The term “doctor and pharmacy shopping” colloquially describes patients with high multiple provider episodes (MPEs)-a threshold count of distinct prescribers and/or pharmacies involved in prescription fulfillment. Opioid-related MPEs are implicated in the global opioid crisis and heavily monitored by government databases such as U.S. state prescription drug monitoring programs (PDMPs). We applied a widely-used MPE definition to examine U.S. trends from a large, commercially-insured population from 2010 to 2017. Further, we examined the proportion of enrollees identified as “doctor shoppers” with evidence of a cancer diagnosis to examine the risk of false positives.

Methods:

Using a large, commercially-insured population, we identified patients with opioid-related MPEs: opioid prescriptions (Schedule II-V, no buprenorphine) filled from ≥5 prescribers AND ≥5 pharmacies within the past 90 days (“5×5×90d”). Quarterly rates per 100,000 enrollees (two specifications) were calculated between 2010 and 2017. We examined the trend in a recently published all-payer, 7 state cohort from the U.S. Centers for Disease Control and Prevention for comparison. Cancer-related ICD-9-CM codes were used.

Results:

Quarterly MPE rates declined by approximately 73% from 18.2 to 4.9 per 100,000 enrollee population with controlled substance prescriptions. In 2017, nearly one fifth of these commercially-insured enrollees identified by the 5×5×90d algorithm were diagnosed with cancer. Approximately 8% of this sample included patients with ≥ 1 buprenorphine prescriptions.

Conclusions:

Opioid “shopping” flags are a long-standing but rapidly fading PDMP signal. To avoid unintended consequences, such as identifying legitimate medical encounters requiring high healthcare utilization or opioid treatment, while maintaining vigilance, more nuanced and sophisticated approaches are needed.

1. Introduction

Multiple provider episodes (MPEs) refer to medical encounters with prescribers and/or pharmacies resulting in the receipt of controlled substances (CS) such as opioid analgesics. Identifying patients with exceptionally high MPEs, colloquially termed “doctor shoppers” and “pharmacy shoppers” (Ferries et al., 2017; Sansone and Sansone, 2012), is a cornerstone function of electronic prescription drug monitoring programs (PDMPs) in multiple countries (Chenaf et al., 2016). MPEs are commonly calculated from prescription dispensing data and often interpreted as signals of patient engagement in misuse, abuse, diversion, fraudulent or illegal drug acquisition behaviors (Biernikiewicz et al., 2019; Data Brief, 2018) and are associated with negative health outcomes in the medical literature, ranging from substance misuse to fatal overdose (Dhalla et al., 2009; Hall et al., 2008; Peirce et al., 2012). Furthermore, MPEs involving prescription opioids are frequently associated with receiving overlapping opioid prescriptions and other classes of CS (e.g. benzodiazepines), increasing the risk of transition to chronic use, adverse medication interactions, and diversion (Cuthbert et al., 2020; Holmgren et al., 2020; Schneberk et al., 2020; Simeone, 2017; Wang et al., 2019; Wilsey et al., 2010). In 46 states, prescribers automatically receive patient reports from state PDMPs on MPEs (“PDMP TTAC,” 2020b; Smith et al., 2019). PDMP model state legislation explicitly recommends sending reports when “a patient may be practitioner shopping or pharmacy shopping or at risk of abusing or misusing a CS or other monitored drugs” (PDMP TTAC, 2020a).

A recent study reported that MPE rates in 7 states (CA, DE, FL, KY, ME, OH, WV) declined by >60% between 2010 and 2016 (Strickler et al., 2020) where MPEs were defined as receiving Schedule II-IV opioid prescriptions from ≥5 prescribers AND ≥5 pharmacies within the past 90 days. Opioids with abuse deterrent formulations (ADF) may have contributed to this decline when they were introduced (Chilcoat et al., 2016). But, it remains unclear if MPE rates declined accordingly in the broader U.S. population. These downward trends are positive indicators assuming that patients are truly engaged in inappropriate drug acquisition (Simeone, 2017). However, prescription dispensing records, by themselves, offer limited insight on patient motivations or medical conditions because they lack diagnostic information. Thus, there is reasonable concern that MPE algorithms may inadvertently identify patients with justifiable reasons for high MPEs, such as patients diagnosed with cancer (Vitzthum et al., 2020). When diagnostic data is available (e.g., via insurance claims), data-driven MPE measures promulgated by multiple national organizations recommend a priori exclusions of patients with cancer to avoid “false positives” (Moyo et al., 2019; NCQA, 2020). Yet, inadvertent PDMP “spill over” effects resulting in decreased opioid prescribing to patients with cancer have been reported as have data errors affecting cancer patients (Graetz et al., 2020; Tang et al., 2020). Oncologists report that patients encounter difficulty accessing opioids due, in part, to programs like “lock-ins” meant to prevent MPEs (Page and Blanchard, 2019).

The purpose of this short communication is to examine and compare longitudinal trends in prescription opioid-related MPEs in the U.S. from 2010 to 2017 using a large commercial claims database with diagnostic medical claims available. Secondarily, we examine the proportion of those patients diagnosed with cancer.

2. Methods

We used quarterly de-identified health claims from a large, nationally representative, commercially-insured population of enrollees with a history of any CS prescription between 2010 and 2017 (Perry et al., 2019). We replicated a common MPE definition: enrollees with opioid prescriptions (Schedule II-V, Generic Product Identifier 65, buprenorphine excluded) from ≥5 prescribers AND ≥5 pharmacies within a previous 90-day period (“5×5×90d”)(Strickler et al., 2020). The MPE rate per 100,000 enrollees was calculated as the quarterly number of enrollees with an MPE divided by the total number of enrollees with CS (continuous enrollment not required). In addition, we calculated rates using enrollees dispensed opioids in the denominator to adjust for opioid prescribing changes.

The all-payer comparator was the averaged 5×5×90d MPE rates in 7 states between 2010 and 2016 (last year available) from the Prescription Behavior Surveillance System (PBSS), which compiles all-payer (cash inclusive), quarterly state-wide data on dispensed CS prescriptions tracked by participating PDMPs to evaluate high-risk prescribing behaviors.(Strickler et al., 2020) States contributed unequally to each quarter. Cancer was identified using International Classification of Diseases, 9th Revision, Clinical Modification codes.(SEER, 2014, p. 9) This study was approved by the University of Kentucky IRB #43511.

3. Results

Our sample, representing approximately 2.5% of the U.S. population (quarterly mean 7,989,904 enrollees), was relatively stable with a temporary decline of approximately 8% between 2014–2015, returning to expected levels thereafter (not shown). On average, 1,200,322 (~15% of sample) enrollees per quarter had an opioid prescription; this group increased by 19.3% over the study period.

On average, 1,074 enrollees (60% female, avg. 42 y.o.) met the definition of “doctor shopping” using the 5×5×90d algorithm on a quarterly basis over the 8-year period. The MPE rate per 100,000 enrollees with any CS steeply declined (73%) from 18.2 to 4.9 over the period (Panel A, Figure 1). We roughly calculated a quarterly decline of 43 “opioid shopping” enrollees per 10 million population (linear slope). These trends were not sensitive to the use of the number of enrollees with opioid prescriptions (Panel B, Figure 1). The 7-state, all-payer cohort trend is shown in Panel C of Figure 1. Sharper declines are observed in the immediate period following the introduction of ADF OxyContin compared to ER oxymorphone in all series (see reference lines, Figure 1).

Figure 1.

Multiple provider episode rates (MPEs; “doctor shopping”). Panel A: per 100,000 enrollees with any controlled substance (CS) prescriptions, Panel B: per 100,000 enrollees with opioid prescriptions, and Panel C: per 100,000 population all-payer cohort (CA, DE, FL, KY, ME, OH, WV)*. MPEs were defined as having an opioid prescription (Schedule II-V) from ≥5 prescribers AND ≥5 pharmacies within a 90-day period. Abbreviations: Abuse deterrent formulations (ADF), extended-release (ER). Note different scales resulting from alternative denominator specifications. *See (Strickler et al., 2020) for technical specifications on all-payer cohort.

The quarterly number of enrollees in the sample with diagnosed cancer increased from 626,565 to 808,799 (29.1%). On a quarterly basis, the cancer prevalence ranged from 7.9% to 10.3%. The quarterly percentage of enrollees identified by the 5×5×90d algorithm with diagnosed cancer in medical claims ranged from 13.9% to 21.2% in Q1 2010 to Q4 2017, respectively. The MPE rate in this group, an order of magnitude lower than our primary population, declined by 57.7% between 2010 and 2017 (0.07 versus 0.03 MPE per 100,000 population, not shown).

4. Discussion

The absolute rate and magnitude in the decline of quarterly MPE rates from this U.S. commercial population was comparable to the all-payer cohort (73% v 80.3%, respectively) over the study period. Thus, MPE rates appear to have declined substantially in the entire U.S., likely due to several factors. First, opioid prescribing has declined nationwide (Guy et al., 2019). However, this cannot entirely explain our findings as the percentage of enrollees receiving opioid prescriptions (primarily from increased enrollment in Medicare Advantage plans) increased during the study period. Second, PDMPs, designed to help detect MPEs, have strengthened considerably (Smith et al., 2019). “Doctor shopping” is a commonly discussed between providers using PDMPs (Delcher et al., 2018) but whether PDMPs impact “doctor shopping” rates is an open question. A three state study (KY, OH, WV) showed that only KY’s MPE rate declined, significantly, immediately after PDMP use was mandated with no change to trend. (Strickler et al., 2019). Mandatory PDMP registration and proactive reporting had no immediate impact on MPE rates in California; similar to Oregon (Castillo-Carniglia et al., 2020; Deyo et al., 2018). Third, the introduction of ADF opioids appear associated with MPE reduction in the short-term. One study reported larger percentage declines in MPE rates for brand extended-release (ER) oxycodone after its ADF entry in late 2010 compared to ER oxymorphone entry in early 2012. (Chilcoat et al., 2016) We did not conduct formal tests, but visual trends during these periods are compelling and apparently several MPE rates rebounded to pre-ADF levels approximately one-year later. We note that ER oxymorphone never met FDA requirements for ADF labelling (Center for Drug Evaluation and Research, 2019).

In the literature, MPE prevalence varies widely due largely to inconsistencies in operational definitions and study populations, yet remaining rare as our findings show. A single year study in Ohio Medicaid identified only 231 out of 3.5 million beneficiaries as “doctor shopping” (Data Brief, 2018). The Office of Inspector General reported 8,796 beneficiaries as “doctor shoppers” (0.07% of 13.4 million beneficiaries with opioid prescriptions) and that “doctor shopping” had decreased by 61% between 2016 and 2018 (HHS OIG Data Brief, 2019). Using a back-of-the-envelope estimation, we multiplied the annual rate from the all-payer cohort (3 per 100,000 population) by the U.S. population at-risk of “doctor shopping” (≥1 opioid prescription, ~47 million people)(Schieber et al., 2020) to calculate that approximately 1,425 people (<30 people/state) had MPEs for opioids in 2016.

We found that, in 2017, nearly one-fifth of the commercially-insured enrollees identified by the 5×5×90d algorithm had cancer. A Pennsylvania Medicaid study found that cancer patients are likely be identified as “doctor shoppers” using common PDMP algorithms (Moyo et al., 2019). As previously noted, state PDMPs lack information on patients’ underlying health conditions (i.e., medical legitimacy) because they rely almost exclusively on transactional prescription dispensing data. While some state PDMPs take internal precautions and triangulate ancillary information (e.g., an oncology specialty in the patient’s prescribing history) to avoid misidentification, this is labor intensive, error prone and inconsistently determined (Prescription Drug Monitoring Program Center of Excellence, 2014; Young et al., 2018). In some states, patients with cancer are exempted from PDMP checks, but, in all likelihood, prescribers still receive unsolicited, automated reports on their patients with cancer (Hoppe, 2018). Depending on the nature of the provider/patient relationship, those reports may inadvertently affect patient care.

Indiscriminately changing algorithms by lowering prescriber and pharmacy count thresholds (i.e., changing j_prescribers x k_pharmacies), including additional classes of CS, and/or extending observation windows casts a wider net but increases the potential for false positives, which can have serious consequences especially if action is taken against the wrong patient. The 2016 U.S. Centers for Disease Control and Prevention Guidelines emphasized that patients should not be dismissed from care based on PDMP information (Dowell et al., 2016) yet available evidence suggest that providers frequently do when they suspect “doctor shopping”. For example, in Kentucky (2015), 29% of PDMP users reported dismissing patients from practice after mandatory PDMP use legislation passed (Freeman et al., 2015). As MPEs become rarer (as we show), predictive associations will diminish. A Maryland study found that MPEs contributed no value to predictive models (Ferris et al., 2019). An 8-year, cross-sectional California PDMP study identified “doctor shoppers” (≥6 prescribers during past 6 months) and found no significant association with overdoses compared to a control group with similar opioid prescribing patterns (Schneberk et al., 2020). As MPE prevalence declines, more nuanced and yet implementable approaches are needed if we continue relying solely on data submitted on prescription dispensing records (Perry et al., 2019; Schneberk et al., 2020).

Furthermore, introspection may be necessary given that patients may engage in MPEs due to iatrogenic factors promoted by the healthcare system itself. For example, Stephenson et al. (2020) surveyed commercially-insured enrollees with MPEs and found that pharmacy convenience, drug pricing, change in patient/physician relationships (e.g. stigma) and illness characteristics were frequently reported reasons for MPEs (Stephenson et al., 2020). In Australia, MPEs occurred frequently in the early course of opioid therapy, perhaps associated with early treatment referral patterns (Adewumi et al., 2020). The American Medical Association has expressed concern that routine pharmacological adjustments to opioid therapy may be misinterpreted as “new starts” resulting in increased “doctor shopping” flags (AMA Press Release, 2020). Taken together, recent research suggests that MPEs may now signify continuity of care problems, described also as “touchpoint” opportunities, rather than prima facie evidence of opioid abuse (Larochelle et al., 2019). Even so, providers must still remain vigilant for signs of opioid misuse among their cancer patients (Yennurajalingam et al., 2021).

Our study has several limitations. Our estimate of the number of patients engaged in MPEs at the 5×5×90d intensity level is not absolute as the count would inevitably change with alternative threshold-based definitions. Patients could be counted more than once if they had the MPE in multiple quarters. Although excluded from our analysis, 7.6% of “doctor shoppers” were enrollees with buprenorphine for OUD which raises important questions about the OUD treatment status of patients with MPEs. Commercial claims do not include cash payments, or claims paid by other third parties, which would underestimate rates in this population. However, opioid utilization controls in pharmacies coupled with the fact that paying cash for a prescription when commercial insurance is available is often perceived as a ‘red flag’ by pharmacists may actually lead to lower MPEs. We cannot fully account for trends after 2017 but multiple state PDMPs and CMS publicly report a continuation well into 2019 (HHS OIG Data Brief, 2019; NC DHHS, 2020).

5. Conclusion

Given the historically-low levels and declining trends observed in opioid-related MPEs in the U.S., we ask whether the reliance on current methods of identifying episodes of “doctor and pharmacy shopping” outweigh the risks of misidentification. Providers relying on PDMPs should be aware of these population-level trends to weigh PDMP alerts of “doctor shopping” in a clinical assessment.

Highlights.

• “Doctor or pharmacy shopping” for opioids has declined precipitously in the United States

• Common data-driven algorithms may inadvertently identify patients with cancer

• Multiple provider episodes may signal problems with care rather than opioid abuse

• Improved methods of monitoring multiple provider episodes are needed