The opioid crises in the United States and Canada continue to worsen (Lisa and Jessica 2018, Humphreys et al. 2022, Kariisa et al. 2022), while there is growing evidence of a rising problem in parts of Europe as well (Pierce et al. 2021, Kalkman et al. 2022, Kalkman et al. 2019). In the U.S., a “fourth wave” is emerging in which polysubstance use drives overdose deaths involving opioids to new, unfortunate heights (Ciccarone 2021, Jones et al. 2020). A large literature documents the recent rise of polysubstance use within the U.S. opioid crisis (Compton, Valentino, and DuPont 2021, Cicero, Ellis, and Kasper 2020), emphasizing the concurrent use of opioids with illicit drugs, especially stimulants (Goodwin et al. 2021, Seth et al. 2018, McCall Jones, Baldwin, and Compton 2017, Kariisa et al. 2021, Townsend et al. 2022).

While receiving less attention, the United States has also been experiencing a rise in the co-use of opioids with non-opioid pharmaceutical drugs, including gabapentinoids (gabapentin and pregabalin), “Z-drugs” (e.g., zaleplon, zolpidem, eszopiclone), and benzodiazepines (Pardo et al. 2019, Guy Jr et al. 2019, Zhang et al. 2022, Sun et al. 2017). The co-use of these drugs with opioids has been observed with both prescription opioids and fentanyl (Pardo et al. 2019, Guy Jr et al. 2019, Zhang et al. 2022, Sun et al. 2017). A similar increase in the co-use of opioids with non-opioid pharmaceuticals is also occurring in Europe (Hockenhull, Amioka, et al. 2021, Hockenhull, Black, et al. 2021, Macleod et al. 2019, United Nations Office on Drugs and Crime 2017) and Canada (Russell et al. 2023, Sharma et al. 2020), despite very different health care systems.

The prescribing and use of non-opioid drugs concurrently with opioids has garnered attention from the medical and pharmacy fields (Guy Jr et al. 2019, Macleod et al. 2019, Esechie et al. 2021, Szmulewicz et al. 2021, Oueini et al. 2022, Tardelli et al. 2022, Ray et al. 2021, Kuehn 2022, Bykov et al. 2020, Peet et al. 2023). Despite this attention, concerns of misuse of gabapentinoids, Z-drugs, and benzodiazepines remain, fueled by studies showing the mortality risks of these drug combinations. Warnings within the health care system, such as those issued by the FDA against the combined use of opioids with gabapentin or pregabalin (US Food & Drug Administration 2019) and, separately, with benzodiazepines and Z-drugs (US Food & Drug Administration 2018), appear successful at reducing the likelihood of joint prescribing (Zhang, Olfson, and King 2019). The continued surge of overdoses involving these substances, then, suggests that some of these drugs may be accessed through illicit markets. These trends are worrisome as prescribers are often less aware of illicit opioid use among their patients and may unknowingly prescribe gabapentinoids, Z-drugs, or benzodiazepines to someone concurrently using illicit opioids. Notably, non-opioid drugs are sometimes used to address opioid withdrawal symptoms; in particular, there is mixed evidence for gapabentinoids in relieving these symptoms (Kheirabadi et al. 2018, Salehi et al. 2011, Kheirabadi et al. 2008).

Our motivation for studying these drugs is due to the recorded dangers of each of them when used concurrently with opioids and the high rates at which they are prescribed. Gabapentin was the 7^th highest dispensed medication in the United States in 2019 (IQVIA 2020). Gabapentin is an anti-convulsant intended to control epileptic seizures, relieve pain associated with postherpetic neuralgia, and treat restless leg syndrome. Off-label prescribing, especially for pain-related diagnoses, is common (Goodman and Brett 2019, Peckham et al. 2018, Zhao et al. 2022). Pregabalin (e.g., Lyrica) treats fibromyalgia and is also a standard off-label pain treatment therapy (Bonnet and Scherbaum 2017, Calandre, Rico-Villademoros, and Slim 2016). Rates of gabapentinoid misuse among those also misusing opioids are relatively high (Evoy, Morrison, and Saklad 2017, Smith, Havens, and Walsh 2016). Similarly, “Z-drugs,” which treat insomnia, are commonly prescribed to prescription opioid users (Vozoris 2019, Oueini et al. 2022).

Benzodiazepines (e.g., Valium, Xanax) treat insomnia, anxiety, and seizures and are also often prescribed with opioids (Hwang et al. 2016, Zhang et al. 2022) despite known risks (Jann, Kennedy, and Lopez 2014, Jones and McAninch 2015, Sun et al. 2017, Boon et al. 2020). Benzodiazepines are commonly found in overdoses involving opioids (Yarborough et al. 2016), and the rate of overdoses involving benzodiazepines has been growing, especially overdoses also involving opioids (Liu et al. 2021).

A series of supply-side interventions were implemented throughout the United States to manage the prescription opioid crisis in its early stages, but researchers have shown that some of these supply-side interventions generated a number of unintended consequences, including a massive shift of consumers to the illicit drug market (Mallatt 2022, Meinhofer 2018, Beheshti 2019, Kim 2021, Powell, Alpert, and Pacula 2019, Cicero and Ellis 2015). Recent work has suggested that the transition of the opioid crisis to illicit opioids was predominantly induced by the reformulation of OxyContin (Alpert, Powell, and Pacula 2018), a drug originally introduced in the U.S. pharmaceutical market in 1996 but reformulated in April 2010. The abuse-deterrent version uses physicochemical barriers to make the pill hard to break, crush, or dissolve. It does not eliminate the ability to misuse the drug (Cicero and Ellis 2015), but it makes it harder to misuse while maintaining the medical benefits of the drug. In August 2010, Purdue Pharma stopped distributing the original formulation of OxyContin to pharmacies. The removal of the original formulation represents one of the largest reductions in the supply of abusable prescription opioids to date, as OxyContin was widely abused as early as 2000 (Cicero, Inciardi, and Munoz 2005) and has been implicated as the key driver of the first phase of the opioid epidemic in the United States (Kolodny et al. 2015, Alpert et al. 2022). The massive shift of consumers from the medical OxyContin market to the illicit drug market subsequently sparked the rise in deaths involving illicit non-opioid substances, including cocaine (Powell and Pacula 2021).

Prior work has leveraged the reformulation of OxyContin to evaluate the impacts of illicit opioid market growth on rates of non-opioid drug overdoses by exploiting differential exposure to reformulation -- states with higher rates of OxyContin misuse prior to reformulation were more “exposed” to the switch to the new formulation. Higher pre-reformulation rates of OxyContin misuse are associated with sharper growth in deaths involving heroin or synthetic opioids, causing overall overdose rates to grow faster (Alpert, Powell, and Pacula 2018, Powell and Pacula 2021). Similar results were found for cocaine overdose deaths due to the mixing (intentional or unintentional) of cocaine with fentanyl (Pacula and Powell, 2021). Related work has found that OxyContin reformulation increased crime (Mallatt 2022), heroin overdose rates (Evans, Lieber, and Power 2019), infectious diseases (Powell, Alpert, and Pacula 2019, Beheshti 2019), foster care rates (Mackenzie-Liu 2021), and child neglect (Evans, Harris, and Kessler 2022) while harming economic conditions (Park and Powell 2021).

In this study, we considered whether the shift to illicit opioids that occurred in the United States after the reformulation of OxyContin can explain the rise in polysubstance overdoses involving non-opioid prescription drugs. Prescribers are less likely to have knowledge about patient-specific illicit opioid use and alter prescribing patterns in response. This study examined whether the rise in illicit opioids is, in fact, driving higher rates of overdoses involving non-opioid prescription drugs. Gabapentinoids, “Z-drugs”, and benzodiazepines are typically provided via a prescription and not manufactured in illicit markets, though there are recent concerns about illicitly-manufactured benzodiazepines (Aldy et al. 2021, Sarker et al. 2022). We implemented an event study design, conditioning on state and year fixed effects, to study the temporal relationship between the rate of pre-reformulation nonmedical OxyContin use and overdoses involving gabapentinoids/Z-drugs or benzodiazepines. We also conditioned on pre-reformulation nonmedical pain reliever use rates and permitted this variable to have a different relationship with the outcome in each year, allowing us to isolate changes in outcomes unique to OxyContin exposure.

The study included (possibly overlapping) 564,543 deaths involving opioids, 131,472 deaths involving benzodiazepines, and 30,173 deaths involving gabapentinoids/Z-drugs. States with higher rates of OxyContin misuse had higher overdose death rates even prior to reformulation, motivating the use of an empirical strategy accounting for these pre-existing differences. See Table 1, Panel A for differences in misuse rates and overdose rates by the aggregated 2004–2009 state OxyContin misuse rate.

In 2004–2009, 0.7% of gabapentinoid/Z-drug overdose deaths involved heroin, 36.9% involved natural/semi-synthetic opioids, and 10.7% involved synthetic opioids. For benzodiazepine overdose deaths, 4.2% involved heroin, 46.7% involved natural/semi-synthetic opioids, and 10.5% involved synthetic opioids. See Table 1, Panel B. The high rates of natural/semi-synthetic involvement suggest substantial scope for reformulation to impact overdoses involving these substances.

Figure 1 shows national trends in deaths involving gabapentinoids/Z-drugs (Panel A) and benzodiazepines (Panel B). Both types of overdoses generally increased throughout the sample period, and these increases were driven by overdoses jointly involving opioids. There was a notable slope shift in deaths involving gabapentinoids/Z-drugs around 2013. Deaths involving benzodiazepines increased until reformulation when we observed a small decrease. We then observed a sharp increase beginning in 2013 until the FDA required a boxed warning in 2016. We do not observe a similar decline following the FDA warning regarding gabapentin in 2019.

Panel C shows opioid-related overdose deaths in comparison to (possibly overlapping) deaths involving gabapentinoids/Z-drugs and benzodiazepines. Death rates involving these substances are small when compared to opioid-related overdose death rates. We focused on them because these rising death rates are still independently important, and they illustrate broader consequences on deaths involving prescription non-opioid drugs.

For our main analysis, we first updated prior analyses (Powell and Pacula 2021) through 2020 to establish how reformulation impacted overdose deaths involving opioids through our time period. These results are provided in Figure 2. Panel A shows the relationship between pre-reformulation OxyContin misuse and per capita opioid-involved overdose deaths over time. We observed the beginning of a sharp increase starting after reformulation, suggesting that states more exposed to reformulation experienced large increases in overdose deaths relative to less exposed states. Table 2 shows the estimated increase in opioid-related deaths in 2020 using the parameterized specification – we estimated that a one percentage point increase in pre-reformulation OxyContin misuse predicted an additional 24.6 opioid-related overdose deaths per 100,000. This implies that a one standard deviation difference in OxyContin misuse led to an additional 5.6 opioid-related overdose deaths per 100,000 in 2020.

This relationship was driven by heroin deaths (Panel B) in the early years after reformulation followed by a staggering relative increase in synthetic opioid deaths (Panel C) in more recent years. Interestingly, in contrast to recent work, we found a relatively weak overall relationship with heroin mortality (see Table 2) as this relationship has weakened over time. We found a strong and growing relationship with synthetic opioids as fentanyl has increasingly replaced heroin. The Table 2 estimate suggests that a state with a one standard deviation higher rate of exposure to reformulation experienced additional growth in synthetic opioid deaths of 5.7 per 100,000.

Figure 2 Panel D suggests that reformulation may have induced some reduction – relative to trend – in deaths involving natural/semi-synthetic opioids, such as OxyContin. While there is little evidence of differential pre-existing trends for heroin or synthetic opioids, Panel D provides some evidence of pre-existing trends for natural/semi-synthetic opioids: states with higher rates of OxyContin misuse had rising levels of deaths involving prescription opioids. This relationship makes sense given that states with higher rates of misuse would tend to have had rising misuse rates and, therefore, increasing rates of overdose deaths during the same time period. However, this differential upward trend appears to stop around the time of reformulation, providing some evidence of reductions relative to a rising counterfactual trend. The Table 2 estimate, however, is positive (though small and statistically significant). The Figure 2 results, as a whole, suggest that reformulation induced substitution from natural/semi-synthetic opioids (likely, OxyContin) to heroin and, later, synthetic opioids (likely, fentanyl).

Our main results are presented in Figure 3. In Panel A, we observed a strong relationship between exposure to reformulation and overdose deaths involving gabapentinoids/Z-drugs. This relationship emerged only after reformulation. There was little evidence of any systematic pre-existing trends, and the relationship grew throughout the post-period. The estimates were statistically significant from 0 at the 5% level in every year beginning in 2014. The estimates implied that a state with a one standard deviation higher rate of exposure to reformulation incurred an additional 0.7 overdose deaths involving gabapentinoids/Z-drugs per 100,000 in 2020 relative to baseline (in 2010). The parameterized specification (Table 2) provides a similar result, implying an additional 0.6 overdose deaths involving gabapentinoids/Z-drugs per 100,000. These results are consistent with a growing propensity to concurrently use gabapentinoids/Z-drugs and illicit opioids after reformulation.

In contrast, the benzodiazepine results in Figure 3, Panel B more closely mirrored the natural/semisynthetic results (Figure 2, Panel D). As with overdose deaths involving natural/semisynthetic opioids, we found that exposure to reformulation may have led to a reduction – relative to trend – in deaths involving benzodiazepines (see Table 2). Notably, however, the trend did not flatten as much in this case. Panels C and D of Figure 3 provide results for the same substances plus co-involvement of opioids. The estimates and patterns are similar.

Focusing only on deaths involving natural/semi-synthetic opioids (rather than all opioids), the results were also similar (Figure 4). These results indicated that exposure to reformulation produced an increase in overdose deaths involving both gabapentinoids/Z-drugs and natural/semi-synthetic opioids after reformulation. The Table 2 estimates imply that a one standard deviation higher rate of exposure to reformulation predicted an additional 0.2 deaths involving both natural/semi-synthetic opioids and gabapentinoids/Z-drugs per 100,000 in 2020. However, deaths involving benzodiazepines and natural/semi-synthetic opioids declined, relative to trend, following reformulation. The Table 2 estimate is negative, though not statistically different from zero.

Figure 5 shows the relationship between reformulation and overdose deaths involving heroin or synthetic opioids and gabapentinoids/Z-drugs (Panel A) and benzodiazepines (Panel B). We grouped heroin and synthetic opioids together given a presumption that, during this time period, these predominantly represented illicitly-manufactured opioids. Here, we found that states more exposed to reformulation were more impacted by the heroin and fentanyl crises, increasing deaths involving gabapentinoids/Z-drugs (Panel A) and benzodiazepines (Panel B). While we did not observe evidence of reformulation increasing overdose deaths involving benzodiazepines and any opioid above (Figure 3) or overdose deaths involving benzodiazepines and natural/semi-synthetic opioids (Figure 4), it appears that reformulation increased the prevalence of deaths involving benzodiazepines and heroin/synthetic opioids. The Table 2 estimates imply that a state with a one standard deviation higher rate of pre-reformulation misuse experienced an additional 0.4 deaths per 100,000 involving illicit opioids and gabapentinoids/Z-drugs and an additional 0.7 deaths per 100,000 involving illicit opioids and benzodiazepines.

To further understand the driving forces behind these results, we also separated deaths involving heroin from deaths involving synthetic opioids. The results are provided in the Appendix. We estimated that reformulation increased deaths involving gabapentinoids/Z-drugs and heroin throughout the post-reformulation period. See Figure S2 Panel A. States with higher pre-reformulation rates of OxyContin misuse also experienced increases in overdose death rates involving heroin and benzodiazepines, but this relationship peaked in 2016 before declining. See Figure S2 Panel B. However, deaths involving synthetic opioids disproportionately increased for both gabapentinoids/Z-drugs and benzodiazepines for most of the post-reformulation period in high OxyContin misuse states. See Figure S3.

This study found evidence that the reformulation of OxyContin in the United States in late 2010, in addition to inducing a major shift towards illicit opioid abuse, also helps explain the rise in polysubstance overdoses involving some non-opioid prescription drugs. Using an event study design, we found that states with higher rates of OxyContin misuse prior to the reformulation of OxyContin experienced statistically significant and large increases in overdose deaths involving gabapentinoids and/or Z-drugs after reformulation.

As an extrapolation exercise, we can imagine comparing the national average (OxyContin misuse rate of 0.567 per 100) to a hypothetical state that was unexposed to reformulation (i.e., a misuse rate of 0). This exercise predicts the national level implications of reformulation on deaths involving gabapentinoids/Z-drugs. The 2020 estimate implies that reformulation increased overdose deaths involving gabapentinoids/Z-drugs by an average of 1.7 / 100,000, suggesting that the entire rise in these deaths can be explained by the transition to illicit opioids. Extrapolation exercises are necessarily “out-of-sample” so standard caveats apply, but the main estimates of the paper generally suggest that OxyContin reformulation explains a large share of the increase in deaths involving gabapentinoids/Z-drugs.

However, the impact of OxyContin reformulation on polysubstance overdoses varies across drugs and combinations of drugs. While overdose deaths involving gabapentinoids/Z-drugs increased following reformulation, overdoses involving benzodiazepines appeared relatively unaffected with some visual evidence that they even declined relative to trend. This result would be consistent with overdoses involving benzodiazepines prior to reformulation tending to also include OxyContin. Reformulation then reduced the rate of these overdoses and, at least in the short-term, did not induce substitution towards combinations of benzodiazepines with more potent illicit opioids. One study, using data submitted to the manufacturer, reported that 34% of deaths involving OxyContin also involved benzodiazepines, the highest rate of the listed substances in the study (Sessler et al. 2014). Our results are consistent with this statistic.

For benzodiazepines and gabapentinoids/Z-drugs, it is possible that the FDA warnings reduced the impact of reformulation. The time fixed effects should account for the independent effects of these warnings; however, if physicians became hesitant to prescribe these substances, then that would reduce the scope for reformulation to independently change the overdose rates involving these drugs. We see little visual evidence of a meaningful reduction in the relationship between OxyContin misuse and overdose deaths involving gabapentinoids/Z-drugs. On the other hand, we observe a plateauing and even small decreases in the relationship for deaths involving benzodiazepines after the FDA warnings. This evidence, however, is only suggestive.

The main findings also indicate that the fentanyl crisis drove gabapentinoids/Z-drugs and benzodiazepine overdose deaths higher in more recent years. We found that OxyContin reformulation increased the prevalence of overdose deaths involving (1) gabapentinoids/Z-drugs and heroin/synthetic opioids, and (2) benzodiazepines and heroin/synthetic opioids. Applying the same extrapolation exercise as before, the estimates imply that reformulation explains all of the rise in overdose deaths involving both gabapentinoids/Z-drugs and heroin/synthetic opioids, and 77% of the rise in overdose deaths involving both benzodiazepines and heroin/synthetic opioids.

Although the reformulation of OxyContin was not unique to the United States, the specific timing of the introduction of the reformulated product was. Because Canada and Europe are also experiencing an increase in the co-use of opioids with non-opioid pharmaceuticals, it will be important to understand if supply restrictions limiting access to medical pharmaceuticals prescribed in large doses in other countries would generate similar shifts in the illicit market. It would also be useful to understand whether the impacts of medical and pharmaceutical regulations spill over into neighboring countries.

This study has some limitations. First, there are concerns about the accurate coding of substances involved in overdose deaths (Ruhm 2018, 2017). Our assumption is that states more exposed to reformulation did not experience relative improvements in reporting at the time of reformulation. Powell and Pacula (2021) provide supportive evidence that this assumption is valid.

Second, we do not observe gabapentinoids or Z-drugs specifically and, further, we cannot distinguish between gabapentin and pregablin. However, a recent study using toxicology reports for 23 states (plus D.C.) for 2019–2020 reports 2975 deaths involving gabapentin specifically (Mattson, Chowdhury, and Gilson 2022). We observed 4091 overdoses involving gabapentinoids/Z-drugs in those same state-years, implying that 73% of these deaths were gabapentin.

Third, we do not observe whether any of the drugs involved in a death were acquired legally or through illicit channels. Instead, we selected prescription drugs in which there is limited evidence of illicit production. We assumed synthetic opioids were primarily illicitly-manufactured fentanyl, though this category also includes deaths involving tramadol and other prescription synthetic opioids. This study identified an important relationship between the reformulation of OxyContin and overdose deaths involving licit non-opioid substances.

Fourth, any difference-in-differences analysis is potentially confounded by secular trends and concurrent policy and behavioral changes. We found little evidence of pre-existing trends for many (but not all) of our main outcomes. Further, we separately accounted for the independent predictive power of pain reliever misuse more generally. This approach permitted us to isolate the impact of changes related to OxyContin specifically.

Fifth, we performed our analyses at the state-level though states are unlikely to represent appropriate borders for medical or illicit drug markets. Our measures of interest, however, were only available by state. The analysis relies on variation in OxyContin misuse rates, not correctly modeling drug markets. Additional geographic information would provide more variation and improve power.

The shift to illicit opioids radically changed the nature of the opioid crisis, renewing old drug overdose crises involving cocaine and psychostimulants. Drug overdose deaths involving substances not historically manufactured illicitly have also started to increase, but this rise has not yet been tied to changes in the opioid crisis and the reformulation of OxyContin. The existence of such a relationship is critical for understanding how prescribing patterns should adapt. Prescribers are less likely to be aware of illicit drug use. If overdose deaths involving non-opioid prescription drugs are increasing because of their co-use with illicit opioids, prescribers need to be more proactive at obtaining information about potential non-medical use of other substances and policymakers must consider the implications when evaluating policy responses.

We found that the response of overdoses involving non-opioid prescription drugs depends on the substance. The estimates related to overdoses rates involving gabapentinoids/Z-drugs expressed a pattern similar to those observed for illicit opioids and other illicit substances such as cocaine (Powell and Pacula 2021). This result suggested that the striking growth in overdose rates involving gabapentinoids/Z-drugs (Kuehn 2022, Tardelli et al. 2022, Enos 2022) is driven by the illicit opioid crisis and initiated by conditions due to the reformulation of OxyContin.

On the other hand, the estimates for deaths involving benzodiazepines suggest that reformulation may have reduced these deaths initially, consistent with high rates of OxyContin involvement in benzodiazepine overdose deaths (Sessler et al. 2014). However, we find a strong relationship with overdose deaths involving benzodiazepines and illicit opioids.

In each case, we observed increases in overdoses involving the substance and synthetic opioids, which would suggest that as fentanyl’s penetration continues to expand across the country that we will witness more deaths involving these prescription drugs. Given that prescribers are typically less capable of monitoring illicit drug acquisitions, these trends are concerning since they imply that prescribers may unintentionally enable co-use of opioids with these substances.