Predictors of Opioid Misuse During Emerging Adulthood: An Examination of Adolescent Individual, Family and Peer Factors

Abstract

Introduction.

Opioid misuse has reached epidemic proportions among emerging adults in the U.S. To inform prevention efforts, this study examined adolescent factors related to alcohol and marijuana (AM) use that are associated with a higher or lower risk for opioid misuse during emerging adulthood.

Methods.

We used 11 waves of survey data from a diverse California cohort (N=6,509). Predictor variables from waves 1-7 (ages 11-17) included individual (resistance self-efficacy, positive expectancies) family (older sibling and important adult use), and peer (perceived norms, time spent with peers who use, peer approval) factors. Opioid misuse at wave 8 (mean age=18.3) and wave 11 (mean age=21.6) included heroin and nonmedical prescription drug use.

Results.

Initial latent growth models (LGMs) indicated that nearly all intercepts and slopes for individual, family, and peer AM factors predicted opioid misuse at waves 8 and 11. These associations were reduced to non-significance after adjusting for prior other substance use with the exception of three intercepts: positive expectancies, peer approval, and older sibling use predicted a higher probability of opioid misuse at wave 8.

Conclusions.

Stronger AM positive expectancies, perceived peer approval of AM use, and older sibling AM use during adolescence are associated with a higher likelihood of opioid misuse during the transition to emerging adulthood. However, most adolescent factors were no longer associated with subsequent opioid misuse after adjusting for history of other substance use, highlighting the importance of considering the larger context of substance use in studies of opioid misuse among young people.

1. INTRODUCTION

The opioid crisis in the U.S. has been declared a national public health emergency (USDHHS, 2017). Opioid misuse can include taking prescribed opioid analgesics in a manner not intended by the prescriber, as well as the use of illicit opioids such as heroin and synthetic opioids such as fentanyl (NIDA, 2019). The misuse of opioids particularly affects emerging adults between the ages of 18-25. In 2018, 5.6% of emerging adults (about 1.9 million) misused opioids in the past year, with the vast majority of misuse due to prescription opioids rather than heroin (Substance Abuse and Mental Health Services Administration (SAMHSA), 2017). The prevalence of prescription opioid use disorder (OUD) among 18-25 year olds increased 37% between 2002-2014 (Martins et al., 2017). By 2018, 0.9% of emerging adults (about 312,000) had an OUD in the past year (SAMHSA, 2019). Further, 12.4% of deaths among those ages 15-24 in 2016 could be attributable to opioid misuse (Gomes et al., 2018).

To inform efforts to reduce opioid misuse among emerging adults, it is critical to identify factors during key developmental periods, such as adolescence, that may contribute to opioid misuse as young people transition into adulthood. However, surprisingly few longitudinal studies have been conducted to address this important issue. For example, among 50 studies included in a 2015 systematic review of risk and protective factors associated with the nonmedical use of prescription drugs among youth in the U.S. (Nargiso et al., 2015), only one study used a longitudinal design and specifically focused on opioid misuse. This study by Catalano et al. (2011) involved 912 participants from the Raising Healthy Children study who were repeatedly assessed from elementary school through age 21. They found that nearly all emerging adults who reported prescription opioid misuse also indicated that they had used other substances such as alcohol, cigarettes, and marijuana. Further, results indicated the importance of examining risk factors and outcomes associated with opioid misuse within the context of other substance use; prescription opioid misuse explained little unique variance in most negative outcomes that were examined at age 21 (e.g., drug use disorder, mood disorder, school dropout, work-related problems, poor health, property crime) beyond that explained by participants’ other substance use. A more recent review of the 76 empirical articles identified just a handful that examined longitudinal trajectories of opioid misuse and called for more longitudinal research examining risk and protective factors (Bonar et al., 2020).

The larger literature on nonmedical use of prescription drugs suggests that individual, peer, and family factors are important to consider as both risk and protective factors for opioid misuse among young people. Young et al. (2012) reviewed 25 studies, and found that individual-level characteristics such as school performance (e.g., poor academic achievement, low school bonding), poor self-reported health status, mental health problems, sexual victimization, and sensation seeking were associated with increased rates of nonmedical use of prescription drugs. Significant family- and peer-level correlates included parental monitoring and parental disapproval predicting decreased risk, and peer approval and peer delinquency increasing risk for pain reliever use. Importantly, other substance use was a consistent and positive correlate of nonmedical prescription drug use across studies. Similar results were found more recently by Nargiso et al. (2015), who used a social ecological perspective in their review of 50 articles to identify risk and protective factors associated with nonmedical use of prescription drugs among adolescents and young adults. Although informative, none of the studies reviewed by Nargiso et al. (2015) and Young et al. (2012) examined the extent to which individual, family and peer factors related to the most common substances used to get “high” during adolescence, namely alcohol and marijuana (AM), and how these substances may be prospectively related to the risk of opioid misuse during emerging adulthood. Further, little is known about the unique contribution of these factors to subsequent opioid misuse among young people after accounting for their use of AM and other substances.

Using 11 waves of data from an ongoing study of a racially/ethnically diverse California cohort, the present study addresses the important question of what adolescent AM-related factors from ages 11-17 are associated with a higher or lower likelihood of subsequent nonmedical prescription opioid and heroin use during the transition into emerging adulthood (age 18) and several years later (age 21-22). The adolescent AM-related factors focus on three predictor domains - individual, family, and peers – given their relevance to nonmedical prescription drug use among youth (Nargiso et al., 2015; Young et al., 2012). Within each domain, we examine AM-related factors which are known to be relevant to other forms of substance use among young people and are often targets of prevention and intervention efforts (e.g., D’Amico et al., in press; Stone et al., 2012): resistance self-efficacy (RSE) and positive expectancies about use (individual); use by older siblings and important adults (family); and perceived norms, time spent around peers who use, and peer approval for use (peers). We used latent growth modeling to: (a) identify which adolescent AM-related factors in each of the three different domains are most predictive of subsequent opioid misuse; and (b) how the timing of these factors [i.e., initial levels (intercept) and change (slope)] is related to opioid misuse at waves 8 and 11. Importantly, given the considerable overlap between opioid misuse and use of other substances to get “high” (Catalano et al., 2011), we examine these prospective associations between adolescent factors and emerging adult opioid misuse in the context of participants’ history of other substance use.

2. METHODS

2.1. Sample and Procedures

Participants were from two cohorts of 6^th and 7^th grade students enrolled in 2008 (wave 1: mean age 11.5; n=6,509) and subsequently followed through 2019 (wave 11: mean age 21.6; n=2,496). They were initially recruited from 16 middle schools across three school districts in Southern California, selected to obtain a diverse sample, as part of an evaluation of CHOICE, a voluntary after-school substance use prevention program (D’Amico et al. 2012). Adolescents who participated in CHOICE were representative of students in their middle schools in Southern California with respect to demographic and substance use risk (D’Amico et al. 2012). The CHOICE program, conducted over 10 years ago, showed effects on youths’ alcohol and other drug use one year after the program; however, no effects were observed beyond one year and intervention status at wave 1 is unrelated to substance use or retention across study waves. Study procedures are reported in detail elsewhere (D’Amico et al. 2012). Briefly, participants completed waves 1-5 (wave 1: Fall 2008; wave 2: Spring 2009; wave 3: Fall 2009; wave 4: Spring 2010; wave 5: Spring 2011) during PE classes at schools, with follow-up rates ranging from 74-90% (excluding new youth that could have come in at a subsequent wave) during this time period Following wave 5, participants transitioned from these middle schools to over 200 high schools. At that point they were re-contacted and re-consented to complete annual web-based surveys, 61% of the sample participating in wave 6 (Spring 2013-Spring 2014). Wave-to-wave retention rates between waves 6-11 range from 80-92% (waves 6-7: 80%; waves 7-8: 91%; waves 8-9: 89%; waves 9-10: 90%; and waves 10-11: 92%). The present analyses focus on predictors of opioid misuse for those who completed wave 8 (n=2,511) and those who completed wave 11 (n=2,496). Participants who did not complete a particular wave of data collection remained eligible to complete all subsequent waves. That is, they did not “dropout” of the study once they missed a survey wave; rather we fielded the full sample at every wave so that all participants had an opportunity to participate in each individual survey. Participants were paid $50 for completing each web-based survey. Substance use at wave 10 did not significantly predict retention at wave 11, similar to previous waves (D’Amico et al., 2018; Dunbar et al., 2018); however, retained participants were slightly more likely to be female (94% vs. 91%) and tended to be slightly younger at wave 10 (mean=20.6 years vs. 20.9 years). We did not find a significant difference in retention by race/ethnicity. All participants consented to the study, and procedures were approved by the RAND IRB.

2.2. Measures

2.2.1. Past year use of opioids (waves 8 and 11).

Participants were asked separate questions about how many times in the past year they had used or tried heroin (horse, smack, H) and prescription narcotic medications to get high (like Vicodin, codeine, OxyContin, and Percocet), rated on a scale from 1=none to 6=more than 20 times. Most participants who reported heroin use also reported prescription drug misuse (96% at wave 8 and 83% at wave 11) and both items were highly skewed at the upper end of the response scale. As a result, we derived dichotomous indicators of any past year opioid misuse (0=no, 1 =yes) at waves 8 and 11.

2.2.2. Background covariates.

At wave 1, participants reported on their: age (in years), gender (female vs. male), race/ethnicity [non-Hispanic white (reference), non-Hispanic black, Hispanic, Asian, Multi-ethnic, and Other], mother’s education (as an indicator of family socioeconomic status; rated from 1=did not finish high school to 4=graduated from college), and family structure (i.e., intact nuclear family). To control for past substance use, participants were asked at Waves 3 through 7 how many times in the past year (1=none to 6=more than 20 times) they had used or tried: alcohol; marijuana; inhalants; other illegal drugs (cocaine, hallucinogens, methamphetamine); over-the-counter medicines to get high; and prescription medications to get high. At each wave we summed the number of substance types used in the past year (range: 0-6).

2.2.3. Individual time-varying covariates (waves 1-7).

Resistance self-efficacy (RSE) was assessed by asking: “Suppose you are offered alcohol [marijuana] and you do not want to use it. What would you do in these situations: 1) your best friend is drinking alcohol [using marijuana]; 2) you are bored at a party; and 3) all your friends at a party are drinking alcohol [using marijuana]?” (Ellickson et al., 2003). These six items were rated from 1=I would definitely drink [use marijuana] to 4=I would definitely not drink [use marijuana] and averaged (α=0.90). Positive expectancies for AM use were assessed with three items for each substance (e.g., using alcohol will relax you, let you have more fun, or helps you get away from your problems (D’Amico and Edelen 2007; Tucker et al. 2003). These six items were rated from 1=strongly disagree to 4=strongly agree and averaged (α=0.88).

2.2.4. Family time-varying covariates (waves 1-7).

Sibling substance use was assessed with the following items: “Do any of your older brothers or sisters drink alcohol [use marijuana] sometimes? Answers included “I don’t have any older brothers or sisters,” “yes,” or “no.” Participants without siblings were coded as ‘no’. Scores were summed across the two substances to form a single score, ranging from 0 to 2, indicating the number of different substances used. Adult substance use was assessed with respect to “the adult who is most important to you and that you spend time with.” This item was designed to focus on an influential adult figure, assumed to be a parent for many respondents, and asks how often this adult drinks alcohol [uses marijuana] (1=never, 2=less than once a week, 3=1–3 days a week, and 4=4–7 days a week). Responses were averaged across the two substances (α=0.39).

2.2.5. Peer time-varying covariates (waves 1-7).

Perceived norms were assessed by asking participants to think about a group of 100 youth their age, and indicate how many youth had 1) consumed alcohol at least once a month and 2) ever tried marijuana (Pedersen et al. 2013). Response options ranged from 1 (“none”) to 11 (“100”) with multiples of 10 as anchors. Responses were averaged across the two substances (α=0.79). Time spent around peers who use was measured by asking how often participants were around peers who drank alcohol [used marijuana] (1=never, 2=hardly ever, 3=sometimes, 4=often; D’Amico et al. 2008). Responses were averaged across the two items (α=0.79). Peer approval of substance use was assessed by asking how their friends would feel if they found out the adolescent sometimes drank alcohol [used marijuana] (1=they would disapprove and stop being my friends, 2=they would disapprove but still be my friends, 3=they would approve/wouldn’t care; Longshore et al., 2007). Responses to these two items were averaged (α=0.87).

2.3. Analyses

We estimated a series of latent growth models (LGM) from waves 1-7 in a structural equation modeling framework using Mplus v8 (Muthén and Muthén, 2012-2017). We used the weighted least squares with mean and variance adjusted estimator (WLSMV), which can accommodate categorical and ordinal data, missing data, and provide unbiased and consistent estimates (Asparouhov and Muthén, 2010). In LGM, the model intercept represents the predicted value of the outcome when the predictor is equal to zero, and thus represents a baseline level or probability, whereas the slope represents the change in level or in the probability over time.

We used a model building process in which the adolescent risk and protective factors were binned into one of three domains: individual, family, and peer. Within each domain we estimated two separate models. First, an initial latent growth model for each domain specific predictor was estimated for any opioid misuse (1=yes, 0=no) at wave 8 and wave 11. Second, the same model was estimated but, in addition to our baseline covariates, we included the LGM for prior substance use to determine what, if any, factors remained predictive of opioid misuse after controlling for prior use. All growth factors were allowed to correlate. Each domain model was evaluated for model fit using conventional fit criteria: χ², RMSEA, and CFI. For each domain model, a series of model constraints were imposed wherein non-significant paths were constrained to zero, and change in model fit evaluated for decrements in overall model fit. Nested models (models with and without constraints) were evaluated using the DIFFTEST model test function in Mplus given that with WLSMV estimation standard chi-square difference tests are not appropriate as the difference between nested models is not chi-square distributed (Asparouhov and Muthén, 2006). The model refining process was terminated once all non-significant associations were constrained or the DIFFTEST results indicated a significant decrement in model fit, thus resulting in the most parsimonious model. All models controlled for demographic characteristics (both Model 1 and Model 2) and other (non-opioid) forms of substance use (when entered into Model 2).

3. RESULTS

3.1. Descriptive analyses.

Table 1 provides information on demographic characteristics, as well as participants’ opioid and other substance use at wave 8 and wave 11. Prior to examining associations of adolescent risk and protective factors with opioid misuse at these two time points in emerging adulthood, we examined LGMs for all longitudinal measures for significant change from waves 1-7. All models tested the need for a random slope by constraining the slope variance to zero and conducting a model fit test using DIFFTEST function in Mplus. None of the domain-specific LGMs fit the data better when the slope variance was constrained to zero, thus a random slope was estimated for each. Each of these models fit the data well, based on conventional model fit criteria. There was significant change from waves 1-7 for each of the longitudinal measures; as such, each was examined in relation to subsequent opioid misuse. Table 2 presents the unconditional growth models for the adolescent longitudinal measures: AM resistance self-efficacy showed a significant decrease over time during adolescence, whereas all other measures showed a significant increase over time.

Table 1.

Descriptive statistics for demographic and opioid misuse measures

Variable	M(SD) or n(%)
Gender (male)	3,271 (50.3%)
Age at wave 11	21.6 (0.8)
Mother’s education
Did not finish high school	1,002 (17.4%)
Finished high school	1,176 (20.4%)
Some college	794 (13.8%)
Graduated from college	2,785 (48.4%)
Race/Ethnicity
White	1,022 (15.7%)
Black	210 (3.2%)
Hispanic	3,495 (53.8%)
Asian	1,050 (16.2%)
Multiracial	619 (9.5%)
Other	104 (1.6%)
Domain average at wave 1
Resistance self-efficacy (RSE)	3.76 (0.55)
Positive expectancies	1.37 (0.68)
Sibling use	0.15 (0.42)
Adult use	1.35 (0.52)
Perceived norms	1.41 (1.01)
Time spent with peers	1.16 (0.48)
Peer approval	1.60 (0.70)
Past year opioid misuse n (%)
At wave 8 (mean age = 18.3)	109 (4.4%)
At wave 11 (mean age = 21.6)	96 (3.9%)

Note. RSE scale: 1 = definitely use to 4 = definitely not use. Positive expectancies scale: 1 = strongly disagree to 4 = strongly agree. Perceived norms scale: 1 = none to 11 = 100. Peer approval scale: 1 = they would disapprove and stop being my friends to 3 = they would approve/wouldn’t care.

Table 2.

Unconditional growth model for each longitudinal individual, family, and peer measure

Variable	Intercept	Slope
Individual
Resistance self-efficacy	3.76 (0.01)*	−0.09 (0.002)*
Positive expectancies	1.35 (0.01)*	0.13 (0.003)*
Family
Sibling use	0.13 (0.01)*	0.62 (0.01)*
Adult use	1.33 (0.01)*	0.02 (0.002)*
Peers
Perceived norms	1.30 (0.01)*	0.54 (0.01)*
Time spent with peers who use	1.13 (0.01)*	0.17 (0.003)*
Peer approval of use	1.56 (0.01)*	0.13 (0.002)*

Note.

^*Parameters denoted are significant at p < .05.

3.2. Individual models.

For AM RSE, both starting points (intercepts) and less steep declines over time (slopes) predicted lower probability of using opioids at waves 8 and 11. In the case of AM positive expectancies, starting points (intercepts) predicted a higher probability of opioid misuse at waves 8 and 11, and increase over time (slope) predicted a higher probability of opioid misuse at wave 11. After introducing prior substance use LGM, the effects for RSE weakened and were no longer significant. However, the intercept of positive expectancies remained a significant predictor of higher probability of opioid use at wave 8 and a marginally significant (p=0.05) predictor at wave 11. Further, increase over time in positive expectancies was a marginally significant predictor of higher probability of opioid use at wave 11 (p=0.05). Of note, when introducing prior substance use LGM, both intercepts and slopes predicted higher odds of opioid misuse.¹

3.3. Family models.

Prior to controlling for substance use during adolescence, the intercept of older sibling use predicted higher probability of opioid misuse at waves 8 and 11, and the slope predicted higher probability of misuse at wave 8. When introducing prior substance use, the intercept of older sibling use remained significantly predictive of opioid misuse at wave 8, and increases in sibling use from ages 11-17 emerged as a marginally significant (p=.05) predictor of lower probability of opioid misuse at wave 11. For AM use by the most important adult, the intercept predicted higher probability of opioid misuse at waves 8 and 11, and the slope predicted higher probability of misuse at wave 11. These associations were no longer significant when controlling for prior substance use, and increases in adult use from ages 11-17 emerged as a significant predictor of lower probability of opioid misuse at wave 8.

3.4. Peer models.

Prior to controlling for substance use during adolescence, all intercepts and slopes for AM peer norms, time with peers who use AM, and peer approval of AM use predicted a higher probability of opioid misuse at wave 8 (see Table 3). Similar results were found for opioid misuse at wave 11 (see Table 4; except intercept of peer norms was not predictive of opioid misuse). When introducing prior substance use, the only association that remained significant was between the intercept of peer approval of AM use and higher probability of opioid misuse at wave 8.

Table 3.

Individual level associations between growth factors and opioid misuse at wave 8 (mean age = 18.3), with and without controlling for past substance use

Variable		Not controlling for prior AOD use	Controlling for prior AOD use
Individual
Resistance self-efficacy	Intercept	−0.79 (0.11)*	−0.32 (0.29)
	Slope	−3.81 (0.34)*	0.46 (1.37)
Positive expectancies	Intercept	0.22 (0.10)*	0.32 (0.13)*
	Slope	−0.16 (0.11)	1.77 (0.91)†
Family
Sibling use	Intercept	0.71 (0.12)*	0.37 (0.15)*
	Slope	2.21 (0.63)*	−0.92 (0.76)
Adult use	Intercept	0.75 (0.12)*	0.10 (0.14)
	Slope	0.79 (0.99)	−2.80 (1.08)*
Peers
Perceived norms	Intercept	0.28 (0.05)*	0.11 (0.06)
	Slope	0.66 (0.18)*	−0.40 (0.26)
Time spent with peers who use	Intercept	0.74 (0.11)*	0.06 (0.26)
	Slope	2.80 (0.36)*	−1.69 (1.14)
Peer approval of use	Intercept	1.32 (0.12)*	0.68 (0.23)*
	Slope	6.37 (0.66)*	2.41 (1.45)

Note:

^*p < 0.05;

^†p = 0.05.

AOD = alcohol and other drugs.

Table 4.

Individual level associations between growth factors and opioid misuse at wave 11 (mean age = 21.6), with and without controlling for past substance use

Variable		Not controlling for prior AOD use	Controlling for prior AOD use
Individual
Resistance self-efficacy	Intercept	−0.78 (0.13)*	−0.44 (0.35)
	Slope	−3.35 (0.40)*	−0.34 (1.68)
Positive expectancies	Intercept	0.60 (0.10)*	0.29 (0.15) †
	Slope	3.64 (0.53)*	1.58 (0.97)
Family
Sibling use	Intercept	0.53 (0.14)*	0.16 (0.17)
	Slope	0.99 (0.75)	−1.76 (0.89) †
Adult use	Intercept	0.62 (0.14)*	0.08 (0.16)
	Slope	2.14 (0.93)*	−0.51 (1.01)
Peers
Perceived norms	Intercept	0.13 (0.08)	−0.05 (0.10)
	Slope	0.77 (0.21)*	−0.11 (0.30)
Time spent with peers who use	Intercept	0.40 (0.13)*	−0.43 (0.31)
	Slope	2.56 (0.37)*	−1.70 (1.31)
Peer approval of use	Intercept	0.88 (0.14)*	0.26 (0.29)
	Slope	4.98 (0.68)*	1.55 (1.64)

Note:

^*p < 0.05;

^†p = 0.05.

AOD = alcohol and other drugs.

4. DISCUSSION

Given the increasing misuse of opioids among young people in the U.S. (Jordan et al., 2017), it is critically important to identify early risk and protective factors for use that can inform prevention efforts. This study is the first to examine whether early experiences and change over time in certain AM-related individual, family and peer factors from ages 11-17 are associated with a higher or lower likelihood of subsequent opioid misuse during emerging adulthood. In general, results indicated that both initial levels and growth over time in the individual, family, and peer factors that we examined – including AM-related resistance self-efficacy, positive expectancies, perceived norms, older sibling and important adult use, time spent with peers who use, and peer approval of use - predicted subsequent opioid misuse both in the short-term at age 18 and several years later at age 21-22. Although some of these variables, such as peer approval and peer use, have been identified in the larger cross-sectional literature on nonmedical prescription drug use among young people (Nargiso et al., 2015; Young et al., 2012), our longitudinal findings help establish a temporal order of associations with substance use that is critical in determining prevention and intervention efforts, as well as their relevance to opioid misuse in particular.

However, it is important to note that a number of studies have now shown that young people who engage in nonmedical prescription drug use more generally (Faraone et al., 2020; Stewart & Reed, 2015), and opioid misuse in particular (Catalano et al., 2011), are also likely to report use of other types of substances. Further, results from Catalano et al.’s (2011) study found that associations between opioid misuse and subsequent outcomes were generally non-significant after adjusting for prior substance use, indicating that opioid misuse did not have unique predictive value in understanding outcomes. That is why we also examined which individual, family, and peer factors were predictive of future opioid misuse after accounting for the adolescent’s prior substance use. When doing so, most associations between initial levels and changes over time in these factors were no longer predictive of emerging adult opioid misuse. In other words, with a few exceptions, the most robust predictor of opioid misuse was prior substance use rather than adolescent psychosocial factors. However, three notable exceptions were found: initial levels of positive expectancies for AM use, older sibling AM use, and peer approval of AM use remained significant predictors of opioid misuse at age 18. This finding suggests that universal prevention programs that target other forms of substance use during adolescence may have secondary effects on the likelihood of opioid misuse as these young people transition to adulthood by, for example, reducing adolescents’ positive expectancies for AM use and their perceptions of peer approval for AM use (see Spoth et al., 2013).

Another interesting finding from the current study had to do with adolescent family influences on later opioid misuse. Using cross-sectional data from the National Surveys on Drug Use and Health, Griesler et al. (2019) found that associations of parental lifetime AM use with adolescent opioid misuse were not significant after controlling for parental and adolescent other-drug use and other factors. In contrast, our longitudinal results indicated that steeper increases in AM use by an important adult (presumably a parent or guardian, in most cases) during adolescence was predictive of a lower risk of opioid misuse at age 18, and a steeper increase in AM use by an older sibling over time was marginally predictive of a lower risk of opioid misuse at age 21-22, after controlling for the youth’s own use of other substances. Given that family is a primary source of prescription drugs that are misused (Beyene et al., 2014; McCabe et al., 2019), one interpretation is that family members who increasingly engage in AM use may be less likely to misuse opioids, therefore limiting a young person’s access to these drugs. However, more longitudinal research is needed to better understand family influences on opioid misuse.

Several limitations should be noted. First, results are based on a predominantly California sample, which may limit generalizability to young people in other geographic regions in which the opioid crisis has evolved differently. In addition, it is a limitation that the information we have available on nonmedical prescription drug use during adolescence is general, and does not specifically address the use of opioids. Another limitation is the exclusive reliance of self-reported substance use data, as external validation was not feasible. However, research on young adults has shown, for example, that self-reported alcohol use can be corroborated by biomarkers (Simons et al., 2015). In addition, rates of substance use in our sample have been similar to those reported for national samples such as Monitoring the Future (Johnston et al., 2012). Finally, our analyses focused on a limited number of individual, family, and peer factors available in the dataset and consistently assessed from ages 11-17. It is possible that other variables within these three domains (e.g., impulsivity, parental monitoring), as well as other domains (e.g., neighborhood factors), are relevant to opioid misuse in emerging adulthood, and should be considered in future longitudinal research.

Results have implications for both future prevention and research efforts. Prevention efforts may be strengthened by the early identification of adolescents with a history of other drug use, as this was a strong predictor of future opioid misuse in this study. Further, our results suggest that AM positive expectancies and perceived peer approval of use are important to address in early substance use prevention programs, as they emerged as robust predictors of opioid misuse during emerging adulthood, even after controlling for prior substance use. From a research standpoint, findings also emphasize the importance of accounting for use of other substances in studies identifying antecedents (and consequences) of opioid misuse. Finally, most studies of opioid misuse among adolescents and young adults have been cross-sectional and descriptive; more longitudinal research is needed to fully understand risk and protective factors for opioid misuse among young people in order to enhance ongoing prevention efforts.

HIGHLIGHTS.

• Past year opioid misuse reported by 4.5% of sample at age 18 and 3.9% at age 21-22

• Early expectancies about alcohol and marijuana (AM) predict age 18 opioid misuse

• Early peer approval and older sibling AM use also predict age 18 opioid misuse

• Adjusting for prior drug use, opioid misuse unrelated to other adolescent factors

• Targeting early individual/peer/family factors may help prevent later opioid misuse