The present study aimed to explore specificity in genetic and environmental risk for POM and heroin use within a multivariate twin framework. Most twin studies using this approach have utilized an aggregate opioid use phenotype that combines POM and heroin use, potentially obscuring unique influences on each opioid type (Dash et al., 2022; Karkowski et al., 2000; Kendler et al., 1999, 2003; Tsuang et al., 1998). Here, we aimed to identify whether and to what degree POM and heroin use are influenced by distinct etiologic influences and to explore how genetic and environmental influences on POM and heroin use are situated within a broader constellation of drug (mis)use behaviors. Results did not provide definitive support for the IPM v. the CPM, which is consistent with mixed findings regarding the relative fit of these models in multivariate twin studies of drug use (Kendler et al., 2003; Tsuang et al., 1998). However, the conclusions regarding general and specific genetic and environmental influences on POM and heroin use were quite similar across the IPM and CPM. Though their assumptions regarding the nature of shared influences differ, both models identified substantial drug-specific additive genetic variance in POM (26–39% of the total phenotypic variance; 69–71% of the total genetic variance) that was not similarly present for heroin use (0% of the total phenotypic variance in both models; 0% of the total genetic variance in both models), the latter of which was primarily accounted for by variance shared with other drug use (79–80% of the total phenotypic variance). This supports past findings of smaller genetic and environmental correlations between POM and heroin use than would perhaps be expected (Dash et al., 2022).

Also notable were the patterns of effect for prescription sedative misuse, which largely mirrored those of POM. Both POM and prescription sedative misuse emerged with relatively low loadings on general factors shared across drugs and more robust drug-specific influences (45–76% and 30–55% of the total phenotypic variance in POM and prescription sedative misuse, respectively). Such findings suggest that, despite a lack of support for differentiable prescription and illicit use factors, there is something unique about the etiology of prescription misuse behaviors, at least for drugs that act as ‘downers.’ It is not clear why prescription stimulant misuse did not adhere to this pattern, with only 18–21% of the phenotypic variance being drug-specific and predominantly attributable to unique environment. One reason may be differences in subjective drug effects that result in distinct instrumental reasons for misuse of prescription ‘downers’ such as opioids and sedatives v. prescription ‘uppers’ such as stimulants. For example, individual differences such as personality may facilitate use of particular drugs; studies have shown that neuroticism and hopelessness predict opioid and sedative use, while extraversion and sensation-seeking predict stimulant use (Dash, Martin, & Slutske, 2021b; Mahu et al., 2019). Prescription stimulants are also used more frequently in a manner consistent with illicit drug use (e.g. intranasal administration) (Butler et al., 2021; Wilens et al., 2008), which may indicate that prescription stimulant misuse aligns more closely with use of other illicit drugs than with other prescription misuse behaviors.

More broadly, the finding that variance shared across drugs could be modeled as a single factor, as opposed to smaller, separable clusters, is consistent with phenotypic and genotypic findings on the factor structure of substance use and use disorder – and psychopathology more broadly – that have identified an overarching factor accounting for overlap across phenotypes (Caspi et al., 2014; Caspi & Moffitt, 2018; Hatoum et al., 2022; Hicks, Schalet, Malone, Iacono, & McGue, 2011; Sanchez-Roige, Kember, & Agrawal, 2022). These findings are also consistent with a past twin study of drug use that found genetic variance shared between opioids, sedatives, cannabis, cocaine, stimulants, and hallucinogens to be captured by a single factor (Kendler et al., 2003). Though a twin study testing licit and illicit factors conducted by Kendler et al. (2007) identified a licit factor comprised of alcohol, caffeine, and nicotine dependence symptoms and an illicit factor comprised of cannabis and cocaine dependence symptoms, there are plausible explanations as to why this two-factor pattern did not replicate for the prescription-illicit configuration tested here. Alcohol, caffeine, and nicotine are more widely available, more commonly used, and, among adults, universally legal in the place and time of data collection (Australian Department of Health and Aged Care, 2021; Australia Department of Health and Aged Care, 2022; SAMHSA, 2021). These characteristics of ‘licit use’ as operationalized by Kendler et al. (2007) may have contributed to a starker contrast between licit and illicit use than would be observed between prescription misuse and illicit use. That is, differences between prescription misuse and illicit use may be more subtle so as to be insufficient to form clearly separable factors. However, the pattern of results for POM and prescription sedative misuse suggests that the present study may have been underpowered to detect the uniqueness of the prescription misuse phenotypes. Future research in larger datasets may be equipped to shed further light on this pattern.

Overall, the finding that POM may share little genetic influence with heroin use has several implications for opioid research and policy. First, such a pattern suggests that disaggregating POM and heroin use could increase the replicability of findings across samples with varying proportions of POM v. heroin use (Cheng et al., 2018; Dash et al., 2022). Second, the presence of POM-specific genetic influence unshared with heroin use has important implications for case–control selection in genomic studies of opioid use disorder (OUD), which typically do not differentiate POM-exposed individuals from heroin-exposed individuals, nor POM- v. heroin-based OUD (Cheng et al., 2018; Dash et al., 2022; Gelernter et al., 2014; Zhou et al., 2020). This is also salient considering that heroin-exposed individuals tend to progress to OUD at higher rates than POM-exposed individuals (Wu et al., 2011). Thus, POM- and heroin-involved individuals could be differentially represented among cases and exposed controls, thereby introducing potential uncontrolled confounds into such studies. Third, if POM and heroin use have little etiologic overlap, public health efforts seeking to mitigate the impact of heroin and other illicit opioid use by reducing opioid prescribing rates are unlikely to be an optimal approach. Indeed, overdose deaths due to opioids have not dropped in parallel with declines in opioid prescribing, and there are increasing calls to shift efforts toward addressing the root social and economic causes of the opioid crisis (Dasgupta, Beletsky, & Ciccarone, 2018; Mattson et al., 2021).