Introduction

Opioid use disorder (OUD) has significant public health consequences including opioid-related overdose events and deaths [1]. An estimate of 2 million individuals in the US have OUD, with the majority being adults aged 18 to 64 years [2]. The prevalence of diagnosed OUD in adult populations has increased between 2006 and 2011 (from 0.07% to 0.19%) [3] and continued to rise each year thereafter [4].

Although the increased incidence of OUD and overdose has been largely linked to the increased availability of prescription opioids [5], recent evidence shows a decline in opioid prescribing since 2011 [6], which is inversely related to OUD and overdose trends [7]. Potential explanations include a possible transition from prescription opioids to heroin and fentanyl, which most recently have surpassed prescription opioids in their contribution to opioid deaths [4,8,9]. Indeed, a 2015 national survey showed 59.9% of adults with misuse of prescription opioids used them without a prescription, and 40.8% obtained their opioids from friends or relatives [10]. Importantly, the same survey also found that patients with opioid misuse or OUD identified uncontrolled pain as the predominant reason for opioid misuse, emphasizing that OUD remains one of the mainstream clinical problems [10].

Starting in 2010, some health plans and several state regulations limit opioid prescriptions by duration (e.g., maximum of 3- to 14-day supply for certain pain conditions) and by dose (i.e., maximum of 90 mg morphine equivalent dose [MED] per day) [11–13]. Both private and public insurance plans have launched programs that flag individuals at risk for OUD and overdose based on high-dose and chronic use of opioid prescriptions. To monitor inappropriate prescribing, almost all states (excluding Missouri) now have a prescription drug monitoring program to capture controlled substance prescriptions paid by third-party payers or in cash [14]. Yet, these policies and programs, which depend on prescription-dispensing data and do not capture opioids traded in street markets, have been criticized for their low sensitivity in identifying high-risk patients, especially in an era of increasing transition to illicit opioid use [15].

A concern has been raised as to whether current policies and programs related to prescription opioid restrictions may drive patients, especially the high-risk opioid users, to seek alternative prescription medications (e.g., stimulants) or illicit opioids to achieve pain control or to enhance euphoric effects [16]. Here, we conducted a descriptive study to understand changes in prescription opioid patterns among patients with OUD or overdose in the past decade, during which many policies and programs aimed at limiting opioid prescribing were implemented. Using a population-based health claims database from 2005 to 2016, we examined the trend in the proportion of patients across study years with a new diagnosis of OUD or overdose who had filled no opioid prescription during the year preceding the diagnosis. Among patients who filled opioid prescriptions prior to diagnosis, we examined prescribed opioid dose trajectories and described the associations of demographic characteristics with the various prescription opioid dose patterns. Empirical data on prescription opioid trajectories are important in understanding whether there are typical and atypical drug use patterns. This information may serve as a warning sign to alert clinicians in timely identification of high-risk opioid use that requires further management and intervention [17].

Methods

Results

Among 227,038 eligible adult patients with incident OUD or overdose between 2006 and 2016, 33.1% were aged 18 to 30 years, 52.9% were males, and 85.0% were metropolitan residents; 205,945 (90.7%) had a first OUD diagnosis, and 21,093 (9.3%) had a first overdose event (Table 1). In the year before OUD or overdose diagnosis, half (50.5%) of the eligible patients had a diagnosis of chronic pain, and 65.0% had musculoskeletal pain. One-third (32.7%) had depression, and one-fifth (20.3%) had anxiety. Among those with opioid prescription fills before the diagnosis, the majority (71.2%) received short-acting opioids only, and 26.7% received a combination of short-acting and long-acting formulations. One in 4 (30.0%) of patients with OUD or overdose had chronic use of prescription opioids. More than 1 in 3 (43.3%) received prescription opioids at daily doses higher than 50 mg MED, and 1 in 4 (28.8%) had 90 mg MED or higher at any month during the 12-month prediagnosis period. Within 3 months before the diagnosis, only 0.5% and 0.2% of the adult patients had a prescribed opioid dose greater than 50 mg and 90 mg MED, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics and prescription opioid receipt of adults, overall and stratified by OUD and overdose and prescription opioid receipt within 12 months before OUD or overdose diagnosis.

https://doi.org/10.1371/journal.pmed.1002941.t001

In the cohort of patients with incident OUD or overdose, over one-third (35.1%, 79,747/227,038) filled no opioid prescriptions within 12 months before the diagnosis (Fig 1). The crude (unadjusted) proportion increased from 28% in 2006 to a peak of 48% in 2015 (Fig 2). The overall increase was driven primarily by a 2-fold increase in the proportion among adults aged 31 to 40 years (19% in 2006 to 40% in 2016) and a 1.5-fold increase among younger adults aged 18 to 30 years (46% in 2006 to 74% in 2016). Data from the sensitivity analysis restricting the study period to the ICD-9-codes era only showed a similar consistent upward trend with a steeper slope in the younger age groups (S1 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of included patients.

https://doi.org/10.1371/journal.pmed.1002941.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Secular trends in the proportion of study adults without prescription opioid fills within 12 months before OUD or OD diagnosis, overall and by specific age group.

In each subfigure, lines represent trends, overall and within age groups, and for each line, each point represents annual percentage of patients with OUD or OD who had no opioid prescriptions. OD, overdose; OUD, opioid use disorder.

https://doi.org/10.1371/journal.pmed.1002941.g002

Table 2 gives the associations between demographics and no opioid prescription fill in the year before diagnosis of OUD or overdose. After adjustment, we observed a significantly increasing proportion of patients with OUD or overdose who had no opioid prescription fills, starting in 2010 (PRR, 1.08; 95% CI 1.03–1.12; P < 0.001 versus 2006) and continuing through 2015 (PRR, 1.97; 95% CI 1.90–2.05; P < 0.001 versus 2006), followed by a slight decrease from the 2015 level in 2016 (PRR, 1.86; 95% CI 1.79–1.93; P < 0.001 versus 2006). Patients who were male, younger, metropolitan residents, and living in the Northcentral or Northeast areas were more likely to have no opioid prescription in the year before diagnosis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Associations between patient characteristics and no opioid prescription fills within 12 months before OUD or overdose diagnosis (n = 227,038).

https://doi.org/10.1371/journal.pmed.1002941.t002

Among adults with incident OUD or overdose, we identified 5 groups with distinct 12-month prescribed opioid dose trajectories before the outcome (Fig 3). These 5 groups, categorized based on their mean daily MED use of prescription opioids per month, included patients with consistent low-dose use (<3 mg MED), consistent moderate-dose use (20 mg MED), escalating dose use (<3 to 20 mg MED), de-escalating dose (20 to 3 mg or less MED), and consistent high-dose use (150 mg MED). The dose trajectory groups differed significantly for all the demographics, as well as select pain and mental conditions (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Trajectories of mean daily MED in milligrams prescribed in each month for the 12 months preceding diagnosis of OUD or overdose OD in adults who filled at least 1 opioid prescription.

In each subfigure, lines represent types of dose trajectory group, and for each line, each point represents mean daily MED of prescription opioids per month. The scale on the left and right side of the figure was natural logarithm of MED and actual MED, respectively. MED, morphine equivalent dose; OD, overdose; OUD, opioid use disorder.

https://doi.org/10.1371/journal.pmed.1002941.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Characteristics of defined trajectories of prescribed opioid dose within 12 months before diagnosis of OUD or overdose, 2005–2016.

https://doi.org/10.1371/journal.pmed.1002941.t003

Across the 5 identified opioid dose groups in each cohort, there was variation in the proportion of patients with any monthly mean daily dose above 90 mg MED before OUD or overdose diagnosis (Table 3). The consistent low-dose use group showed the smallest proportion (2.2%) of patients who ever exceeded the high-dose threshold, whereas the consistent high-dose use group had the highest proportion (99.7%). Overall, 72.2% of patients with OUD or overdose who filled opioid prescriptions were never prescribed a mean daily dose of 90 mg MED or more during any month of the 12 months before OUD or overdose diagnosis.

Discussion

To our knowledge, the present study using US national, commercial insurance claims data is among the first to provide population-based data on prescription opioid use prior to incident OUD or opioid overdose. Our results provide important insight into the role of commonly used high-risk use criteria in identifying patients at risk for OUD or overdose. We found that more than one-third of patients with incident OUD or overdose had not filled any opioid prescriptions in the year before their diagnosis. More than half of those without opioid prescriptions were young males between 18 and 30 years of age. The proportion of patients whose new OUD or overdose diagnosis was not attributable to their own prescription opioids increased by 86% (after adjusting for patient characteristics) during the study period, resulting in 42% of patients with OUD or overdose without opioid prescriptions in 2016. Our findings are largely consistent with growing literature indicating that prescription opioid use is decreasing, which appears to extend to those populations who are at risk for OUD [6,28]. The findings also echo recent data reported by the US Centers for Disease Control and Prevention (CDC) revealing an increase in heroin overdose deaths starting in 2010 and deaths involving synthetic opioids (mainly fentanyl) in 2013 [29]. Taken together, these findings imply that an increasing proportion of patients with OUD or overdose might have used nonprescription opioids to achieve pain control or to enhance euphoric effects before the disease diagnosis.

There are different pathways regarding how prescribed opioids can lead to overdoses from heroin versus overdoses from prescription opioids. Many prescription opioid users transition to heroin injection after having nonmedical use of prescription opioids, such as misusing prescription opioids (e.g., Oxycontin) and using opioid pills not prescribed for them [9]. Transitioning to heroin is ostensible when patients’ tolerance to prescription opioids grows or they have difficulty in accessing prescriptions [30]. On the other hand, the pathways to prescription OUD and overdose appear to largely originate from inadequate pain control, followed by recreational use, nonmedical use of prescription opioids, and use of opioids for the relieving tension and emotional stress, rather than pain [10,31].

Among patients who did fill opioid prescriptions during the year before their OUD or overdose diagnosis, the present study found 5 distinct trajectories of prescribed opioid doses. Some of these dose trajectories may be not necessarily clinically intuitive. For example, 34.6% of patients with OUD or overdose received only a low dose of opioids less than 3 mg/day MED, and 9.4% had de-escalated to the low dose of opioids before the diagnosis. Although some individuals can be susceptible to even small opioid doses (e.g., because of genetic differences in metabolizing enzymes) [32], literature has suggested that the minimum dose for increased risk of opioid overdose is 20 mg/day MED [33]. Thus, the observed dose trajectories reflect pathological pathways incompletely, suggesting a need for a better understand and capture of the drugs and substances that may have contributed to the OUD or overdose. This might include combinations of prescription drugs such as emerging reports of concomitant use of gabapentinoid leading to overdoses events [34–36], increasing manipulation of abuse-deterrent formulations such as observed with Opana [37], and rising supplementation of prescribed supply with prescription or illicit opioids from other sources. This might include heroin use, which has increased 5-fold in recent years [38].

Importantly, our study observed that 72% of patients receiving opioid prescriptions prior to OUD or overdose were prescribed a mean daily opioid dose below the current high-risk threshold of 90 mg MED. Our findings suggest that even among patients who are receiving opioids as part of their treatment plan, only a small proportion at risk for OUD or overdose will be recognized when relying on current risk stratification algorithms. Although the CDC has recommended avoiding increasing the daily dosage of prescribed opioids to 90 mg MED, there is a controversy over what constitutes a “high dose” [39]. Therefore, dose considerations alone provide a limited understanding of opioid risk without considering mental disorders, histories of trauma or substance use, and other risk factors. Our recent research suggested the Centers for Medicare and Medicaid Services (CMS) opioid overutilization criteria, which also include the 90 mg MED/day ceiling for risk identification, seemed not a good clinical marker for identifying Medicare patients at risk for OUD or overdose [15]. Replicating the similar design in a privately insured population will elucidate the accuracy of the high-dose criterion to assign patients into high-risk strata, especially in light of the increasing adoption and/or enforcement of the 90 mg MED ceiling in clinical practice [11,13].

Several strengths of this study are noteworthy. Our findings complement the existing literature by adding to the understanding of prior prescription opioid use before OUD or overdose. Use of administrative claims data yielded a large sample of adults with newly diagnosed OUD or overdose. Using more than a decade of accumulated data representing the US commercially insured population revealed secular trends in prescription opioid use before OUD or overdose diagnosis. Finally, the use of GBTMs allowed the identification of trajectories of prescription opioid doses among patients with OUD or overdose.

Several limitations warrant mention. First, reliance on reimbursed pharmacy dispensing events did not capture prescriptions paid out of pocket. Prescription drug monitoring program data may further inform our findings, even though there is limited evidence that privately insured populations with comprehensive drug coverage commonly resort to cash payment for opioids. Second, our claims data had no information on race/ethnicity, income and education, and other important risk factors for OUD or overdose, which may have further informed our multivariable models [10]. Third, we inferred the onset of OUD or overdose based on the first medical encounter with a diagnosis. Because OUD diagnosis occurs oftentimes delayed, the reported opioid dose trajectories should not be interpreted as causal risk factors for the development of OUD or overdose but rather understood as a historical snapshot of prescription opioid use that is available to clinicians when diagnosing OUD or overdose. Also, there is a limitation to the use of ICD codes for identifying OUD, as some clinicians may under- or overreport misuse/abuse in patients with opioid treatment [3]. To our knowledge, there are no studies to validate the ICD code–based algorithm for OUD. The limitation with the ICD code–based algorithm for OUD, as well as a limited 12-month washout period without OUD or overdose, might have prevented us from accurately identifying incident OUD cases. Thus, a proportion of our selected studied incident OUD population might include patients with prevalent cases of OUD that were known but underdiagnosed during the 12-month baseline or whose diagnosis was recorded prior to the baseline period. If the likelihood of being prescribed opioids differs between incident and prevalent OUD cases, our results of low prescription opioid use may be biased. Finally, our study results are only generalizable to privately insured populations and may not extend to patients covered by public insurance or those who are uninsured.

Conclusion

In this study of US commercially insured adults with incident OUD or overdose, 35.1% had no opioid prescription fills in the year before diagnosis, and the proportion increased in more recent years. Patients with OUD or overdose exhibited substantial heterogeneity in individual prescription opioid dose trajectories in the year preceding the diagnosis, with the majority receiving prescribed daily doses below the guideline-recommended high-risk threshold of 90 mg MED. Our findings suggest that the vast majority of patients at risk for OUD or overdose will not be identified if solely relying on currently employed prescription-based metrics. Further studies are needed that examine whether the absence of prescription fills was associated with overdoses from illicit opioids and the relationship between dose trajectories with incident OUD or overdose in this new era of decreasing access to prescription opioids.

Supporting information

Acknowledgments

Wei-Hsuan Jenny Lo-Ciganic, PhD, and Amir Sarayani, Pharm D, provided advice on group-based trajectory modeling and list of opioid medications included in the study, respectively. Both are located at the Department of Pharmaceutical Outcomes and Policy, College of Pharmacy, Univesity of Florida, Gainesville.