FormalPara Key Points

The total amount of prescription opioid analgesics dispensed from retail pharmacies has a strong relationship with clinically significant adverse outcomes reported to poison centers.

There is a 30-fold difference in adverse outcomes reported to poison centers among commonly prescribed opioid analgesic tablets, with much of the difference in variation explained by the relative drug potency.

1 Introduction

The current burden of misuse and abuse of prescription opioids in the United States has reached epidemic proportions. Over a 15-year period, the total number of prescriptions for opioids dispensed at US retail pharmacies tripled [1]. The majority of prescriptions emanate from primary care physicians, internists, dentists, and orthopedic surgeons [2]. Increasing opioid availability was mirrored by rising nonmedical use [3] and overdose morbidity and mortality [4], as well as increased social and healthcare costs with an estimated annual total economic burden of US$78.5 billion in the US for 2013 [5].

Numerous factors likely contribute to the current epidemic; however, it has been suggested that drug availability may influence abuse behavior [6]. Not surprisingly, increasing opioid prescription rates correlate with increased diversion and a rise in the magnitude of opioid-related mortality [7]. While hundreds of opioid-containing products of varying formulations are available by prescription, the vast majority (85%) dispensed from US retail pharmacies contain either hydrocodone or oxycodone [2]. Expectedly, the burden of overdose deaths is highest in the category containing these products—the natural and semisynthetic opioids [8]. While parallel increases in availability of opioid analgesics and their associated negative outcomes have been observed [9], drug-specific characteristics and formulation are not completely understood and are of particular interest.

1.1 Goals of This Investigation

While it is commonly presumed that increasing potency among opioid analgesics is linked to higher rates of adverse outcomes, this claim has not been fully substantiated. Potency has been shown to predict ED visits for drug abuse utilizing the Drug Abuse Warning Network (DAWN) database and IMS Health pharmacy dispensation estimates [10]. However, little work has been done to establish comparative rates of clinical morbidity and mortality among opioid analgesics. Poison center (PC) surveillance data can provide insight into this question as they track exposure and outcomes for commonly prescribed opioid formulations. We hypothesized that major outcomes in PC cases occur at different rates among commonly prescribed opioid analgesics, and are a function of drug potency.

2 Methods

2.1 Study Design and Setting

This retrospective observational study reviewed cases of prescription opioid exposures reported to US PCs participating in the Researched Abuse, Diversion and Addiction-Related Surveillance (RADARS®) System Poison Center Program. The study period was a convenience sample spanning 84 months from January 1, 2010 through December 31, 2016.

2.2 Data Source

2.2.1 Opioid-Related Exposures and Medical Outcomes

The RADARS System is a non-profit prescription drug abuse and misuse surveillance program and a part of the Rocky Mountain Poison and Drug Center, a division of the Denver Health and Hospital Authority. The RADARS Poison Center Program comprises 50 PCs accredited by the American Association of Poison Control Centers. These PCs provide poison management advice for 48 states, which includes over 90% of the US population. Each center receives spontaneous calls from caregivers, patients, and healthcare providers regarding potentially toxic exposures. PC specialists trained in toxicology assist in the care of the individual and document numerous aspects of each case, including medical outcome and the place where individuals with an exposure-related call are managed. In a case with multiple drug exposures, priority (or rank) of each substance is recorded by relative contribution to the patient’s clinical condition at the discretion of the PC specialist, with the acknowledgement that drug–drug interactions would be assessed by individual specialists. Participating PC records are uploaded to a central database where case review and quality control are conducted, and exposure mentions are recorded for analysis. Quality control includes a review of exposure reasons, removal of identifying information, and to verify exposure substances [11]; however, the exposure is self-reported and not systematically confirmed with laboratory testing. The data extraction process has been previously described (supplemental appendix reference).

2.2.2 Opioid Dispensing

National outpatient pharmacy dispensing data were obtained from the National Prescription Audit Plus maintained by IMS Health (Danbury, CT, USA now known as IQVIA, Research Triangle Park, NC, USA), a common data source used in health services research for opioid research [12,14,14]. Data are obtained electronically from outpatient pharmacies for every fill of a prescription medication. Approximately 90% of all retail pharmacies are included in the sample [15] and a proprietary projection methodology is used to extrapolate for the entire US. In this analysis, we used national estimates of kilograms dispensed, prescriptions filled, number of dosage units dispensed, and number of individuals filling prescriptions for opioid analgesics.

2.3 Definitions

Serious adverse event (SAE) was a composite outcome including death, hospitalization, or ‘major effect.’ The medical outcome of ‘major effect’ is recorded when the patient had life-threatening symptoms or significant residual disability resulting from the exposure [16].

2.4 Methods and Measurements

All cases regarding exposures to a prescription opioid medication in tablet form and resulting in death, major medical effect, or hospitalization were included for analysis. Schedule II opioid analgesics were included for analysis given their high abuse potential and risk of severe psychological or physical dependence. Tramadol (schedule IV) was included to expand the range of potency for comparison. Fentanyl was excluded due to its primary parenteral route of administration. Codeine was excluded because RADARS did not monitor it during the entirety of the study period.

2.5 Outcomes

The primary outcome of this study was SAEs as a result of exposure to tablets containing seven commonly prescribed opioid analgesics: oxycodone, hydrocodone, morphine, hydromorphone, oxymorphone, tapentadol, and tramadol. No distinction was made between formulation (e.g., extended or immediate release, abuse deterrent formulation, etc.) of each individual opioid in this study, but has been examined previously [17]. Route of administration for exposures was not examined; however, prior work has evaluated effect of relative risk of non-oral routes of opioid administration among the RADARS dataset [18].

2.6 Analysis

Rates of primary outcome were derived using four denominators: 100,000 grams dispensed, 100,000 prescriptions filled, 100,000 dosage units dispensed, and 100,000 individuals filling prescriptions. Linear regression was performed with rate of SAEs regressed on morphine milligram equivalents (MME), with strength of relationship reported as R2 value. Relative opioid potency was estimated using the CDC conversion chart estimating MME [19]. To evaluate the relative contribution among exposure to multiple products, a sensitivity analysis was performed by restricting analysis to cases of exposure to a single product (one opioid product without any other opioid or non-opioid co-ingestants).

3 Results

3.1 Characteristics of Study Subjects

There were 64,538 cases reported to PCs during the study period where one of the index opioids was mentioned. Of all cases with exposure to one or more index opioids, 19,480 cases (30.1%) met criteria for SAEs. Among cases resulting in SAEs, 52.5% were male, and 78.1% of patients were between the ages of 20–59 years (Table 1).

Table 1 Age and gender distribution of exposures with serious adverse event, January 2010 to December 2016
Full size table

3.2 Main Results

There were 19,480 cases involving an opioid of interest associated with an SAE during the 84-month study period, of which 148 were associated with death, 2955 with a major medical effect, and 18,713 with a hospitalization (Table 2). Hydrocodone ranked first in each category and was involved in 49% of all opioid-related deaths reported to PCs. Dispensing estimates of each of the seven opioid products are reported in Table 3. Hydrocodone represented the most prescriptions, dosage units dispensed, and number of individuals filling a prescription, while oxycodone had the largest mass of drug dispensed. The rate of death, major medical effect, or hospitalization varied substantially (0.27–8.02 SAEs per 100 kg dispensed) depending on the individual opioid exposure (Table 4).

Table 2 Poison center cases of opioid analgesic exposure involving the outcomes of death, major medical effect, or hospitalization, January 2010 to December 2016
Full size table
Table 3 Nationwide US retail pharmacy prescription dispensing data associated with commonly prescribed opioid analgesics, January 2010 to December 2016
Full size table
Table 4 Rate of death, major medical effect, hospitalization, or SAE (composite outcome) for commonly prescribed opioid analgesics, January 2010 to December 2016
Full size table

Comparing rates of outcome by relative potency, a hierarchy was observed with hydromorphone (8.02 SAEs/100 kg of drug dispensed) and tapentadol (0.27 SAE/100 kg of drug dispensed) as the highest and lowest rates, reflecting a 30-fold difference among individual opioid products. The relation of SAE rate and potency was linear with an increase of 2.04 cases (death, major medical effect, or hospitalization) per 100 kg of drug dispensed for each 1 unit change in MME (p < 0.001). Linear regression of SAE/100 kg of drug dispensed and drug potency (Fig. 1a) revealed that 96% of the variation was accounted for due to MME. In contrast, prescriptions dispensed, dosage units dispensed, and the number of individuals filling a prescription identified only 42%, 38%, and 33% of variation, respectively (Fig. 1b–d). Association between potency and individual rates of death, hospitalization, and major medical effect are reported in Table 4 and Fig. 2. There was a stronger association between potency and rate of hospitalization (R2 = 0.96) and major medical effect (R2 = 0.83), compared with death (R2 = 0.43).

Fig. 1
figure 1

a–d Correlation of serious adverse event rate and relative opioid potency, January 2010 to December 2016. Grey areas indicate 95% CI range. Serious adverse event—composite outcome resulting in death, hospitalization, or meeting criteria for major medical effect as defined by the American Association of Poison Control Centers coding manual [16]. MME morphine milligram equivalents, URDD unique recipients of  dispensed drug

Full size image
Fig. 2
figure 2

a–d Correlation of drug dispensation and relative opioid potency by outcome, January 2010 to December 2016. Grey areas indicate 95% CI range. Serious adverse event—composite outcome resulting in death, hospitalization, or meeting criteria for major medical effect as defined by the American Association of Poison Control Centers coding manual [16]. MME morphine milligram equivalents

Full size image

A sensitivity analysis was performed by restricting analysis to the 7169 cases of single product exposures (accounting for 36.8% of cases resulting in SAE), yielding a similar pattern of SAE rate and potency to drug dispensation but weaker correlation (R2 values of 0.91, 0.21, 0.25, 0.24 for grams dispensed, dosage units dispensed, prescriptions dispensed, and individuals filling a prescription, respectively).

4 Limitations

PCs capture reported incidents associated with opioid use, representing only a small portion of total adverse events from opioid use, and are subject to reporting bias. Estimating the fraction of all cases reported to, and factors that predict PC involvement is challenging. Early work using PC surveillance data sets suggested poor capture of toxicological fatalities [20]; however, subsequent work has demonstrated a strong relationship between deaths associated with PC cases and deaths reported by coroners and medical examiners on death certificates [21]. A recent analysis of methadone poisonings using RADARS suggested that cases were predictive of deaths despite the limited capture among PCs [22].

Another limitation is that the sources of opioids, whether obtained from diversion or from pharmacy dispensation, were unknown. Further, reported exposures were not confirmed cases and there is no quantitative verification of any opioid exposure. It is unknown if patients were hospitalized because of opioid-related exposure or because of additional effects of other possible co-ingestants. While our sensitivity analysis attempted to address this issue, the PC data cannot attribute causality.

This analysis reports a composite outcome of clinically relevant, yet disparate severity. The relatively small sample size of cases resulting in death and major medical effect limit our interpretation. The majority of cases reported to PCs meeting criteria for inclusion in this analysis were hospitalizations. As such, this individual outcome may weight the outcome disproportionately (as evidenced by Fig. 2). Additionally, the route of administration is unknown. Pharmacokinetic properties of individual opioids vary substantially based on route of administration (e.g., enteric, inhalation, or injection) and portend substantially different risk of mortality [23].

Methadone was not included in our analysis due to the potential for misclassification bias. For the exposure, dispensing data from IMS Health do not include methadone dispensed onsite for treatment of opioid dependence in federally regulated addiction treatment clinics, which would have resulted in an undercount in the denominators of rate calculations. Similarly for the outcomes, the accuracy of differentiating between formulations of methadone used in pain versus addiction is unknown in PC data. Previous analysis by our research group has examined rates of methadone calls to PCs in detail [22].

5 Discussion

Among commonly prescribed opioid analgesics, we found wide variability in the number of SAEs involved in PC cases, ranging from hydrocodone (7872) to tapentadol (97). When adjusting SAEs for total drug dispensing in kilograms, the rate of SAEs was highest for hydromorphone (8.02/100 kg) and lowest for tapentadol (0.27/100 kg), representing a 30-fold difference. The descending rank order was hydromorphone, oxymorphone, hydrocodone, oxycodone, morphine, tramadol, and tapentadol. A highly linear correlation was observed when comparing rates of SAEs and opioid analgesic potency with 96% of the variation observed due to MME alone.

Previous investigations have identified specific opioid analgesics implicated in overdose mortality. Hall reported methadone, hydrocodone, and oxycodone as the opioid analgesics contributing to the greatest number of unintentional overdose deaths in West Virginia, USA [24]. Similarly, a North Carolina cohort identified oxycodone, methadone, and hydrocodone as the worst offenders by number of unintentional overdose deaths [25].

Our findings revealed hydrocodone and oxycodone to be associated with the highest number of cases resulting in SAEs (40% and 37%, respectively) among the studied opioid analgesics, presumably due to their relative abundance and availability. Previous estimates suggest that hydrocodone and oxycodone were the most commonly prescribed opioid analgesics among patients discharged from the emergency department [4], and together account for 85% of all opioid prescriptions dispensed from US retail pharmacies [2]. We report a similar rate of dispensing either hydrocodone or oxycodone (79.6%) among schedule II opioid analgesics included in this study. Interestingly, tramadol accounted for the third most SAEs, despite being the least potent opioid analgesic included in this study. A possible explanation may lie in the additional pharmacokinetic effects of tramadol to inhibit serotonin and norepinephrine reuptake activity, leading to unique toxic effects such as adrenergic hyperactivity, seizures, and serotonin depletion syndrome [26, 27].

We found a robust relationship between relative opioid analgesic potency and rate of SAE per the total amount of the drug dispensed from retail pharmacies. The observed positive linear relationship for increasing potency and rate of outcome was found only for the total amount of drug dispensed (R2 0.92). Interestingly, denominators of prescriptions filled, tablets dispensed, and individuals filling prescriptions nationally did not behave in a linear fashion and showed weak correlation (Fig. 1, R2 0.29–0.36) between rate of outcome and potency. A similar pattern of variation has been observed with rates of relative abuse-related ED visits among opioid analgesics when adjusting for several measures of drug availability [28]. There are several possible explanations for this observation. Prescriptions filled and tablets dispensed are heavily influenced by drug formulation, where extended-release products contain a large amount of active drug per unit dispensed compared with immediate-release products that contain a much smaller amount of drug. For example, the category of hydrocodone would include primarily immediate-release products in the range of 5–10 mg compared with morphine extended release, which is composed primarily of extended-release products in the range of 15–200 mg. Thus, hydrocodone has a very large number of prescriptions with small pill size and often small number of doses due to its use primarily for acute, short-term pain control. In contrast, extended-release morphine is often prescribed for a month with a larger number of units for treatment of chronic pain. The profile of each opioid in relation to the proportion of prescribing various tablet strengths and quantities creates large variation. We posit that both prescriptions filled and tablets dispensed involve an important intrinsic difference based on drug formulation (immediate versus extended release), whereas total drug dispensed removes this inconsistency.

The observed correlation with total amount of drug dispensed has important public health implications. Prior work has illustrated a dose–response relationship associated with opioid overdose-related morbidity and mortality. Dose escalation due to tolerance further complicates matters, as higher daily MME doses are associated with an increased risk of opioid overdose morbidity [29] and mortality [30]. Similarly, it is documented that a majority of prescription opioids implicated in overdose deaths are obtained via diversion [25, 29]. Thus, as more prescription opioids are dispensed (and therefore diverted and available for nonmedical use), it is plausible that opportunities for larger dosing become more abundant and imply greater hazard. To mitigate this risk, physicians should prescribe the lowest potency opioid, for the shortest duration, and exhaust non-opioid alternatives in order to minimize opioid needs and adequately manage a patient’s pain [31].

While the vast majority (89%) of all toxicologic exposures reported to PCs involved a single product, cases resulting in death frequently involve multiple products (58%) [32]. To better isolate the opioid effect alone, we performed additional sensitivity analyses restricting analysis to cases of single opioid product exposure; the results were similarly robust for grams dispensed (R2 = 0.91) and poor for other denominators (R2 = 0.21–0.25), and thus did not alter findings or interpretation.

Epidemiologic data have shown a relationship between prescription opioid sales and opioid-related morbidity and mortality [33]. Paulozzi and Ryan described a linear relationship between total opioid analgesic sales (grams per 100,000 population) and drug poisoning mortality rates (per 100,000 population) among US states [34]. These authors evaluated population-adjusted dispensing and mortality rates and found substantial variability by specific prescription opioid. However, the variation in mortality as explained by each individual drug was limited and other contributing factors, such as potency, were not examined. Dasgupta et al. reported that morbidity associated with non-medical use of opioids is predictable based on potency and prescriptive use [10]. Their model identified higher rates of emergency department ‘mentions’ associated with opioid abuse sequelae when more potent opioid analgesics were involved. Our findings show a congruent relationship with potency, demonstrating higher rates of important clinical outcomes among more potent opioid analgesics. Our data further solidify the predictability of opioid-associated morbidity and mortality in relation to drug potency.

The application of epidemiologic data suggesting that more potent opioid analgesics impart a disproportionate risk of an SAE has clinical importance. A recent 2015 CDC report described marked increases in the strength of opioid analgesics prescribed between 2002 and 2012 [9]. Our study suggests that increasing potency (per one unit MME) correlates with an increase of 1.7 in the rate of SAEs involving a PC, and therefore a smaller number needed to harm (NNH; number of grams dispensed to cause one death, major medical effect, or hospitalization). The conclusions we can draw upon any derived NNHs would be very limited, as individual providers do not prescribe opioids by the kilogram, and PCs capture but a small portion of adverse outcomes related to opioid use, misuse, and abuse. However, mounting evidence that user profiles, total drug dispensation, and potency correlate with clinically significant outcomes suggests that judicious prescribing of opioid analgesics according to clinical guidelines and policies should remain a priority of public health stewardship.

Our results also suggest that adjustments currently used in postmarketing surveillance may not be optimal. The two most common adjustments currently used in analyses used in FDA analgesic advisory committees are prescriptions dispensed and number of dosage units dispensed. However, these measures had poor correlation with the outcomes reported by PCs in our study. In contrast, the total amount of opioid dispensed correlated well. Further work is needed to assess the appropriate adjustments to use in the analysis of postmarketing surveillance data.

6 Conclusions

Several questions remain, particularly regarding various formulations of individual opioid analgesics. Future research may be required as new analgesics come to market that have agonism beyond the μ-opioid receptor pathway. Further, many other drugs like benzodiazepines or gabapentin complicate our understanding of prescription opioid abuse. Future direction for research could elucidate the roles of benzodiazepine co-ingestion, combination drug products, route of administration, and dose–response associated with important clinical outcomes.

In summary, potency of a prescription opioid analgesic demonstrates a significant, highly positive linear relationship with SAEs per 100 kg dispensed that are reported to PCs. Our work suggests that opioid potency should be carefully considered from both individual provider and public health perspectives and supports tighter regulation of more potent opioid drugs.