Key Points
Question
Is nonmedical prescription opioid use associated with later heroin use initiation in adolescents?
Findings
In this 8-wave cohort study of 14-year-old and 15-year-old high school students in Los Angeles, California, who had never used heroin at baseline, youth reporting no, prior, and current nonmedical prescription opioid use during high school exhibited estimated cumulative probabilities of subsequent heroin use initiation by end of the 42-month follow-up of 1.7%, 10.7%, and 13.1%, respectively.
Meaning
Nonmedical prescription opioid use was prospectively associated with subsequent heroin use initiation in adolescents; future research is needed to evaluate whether this association is causal.
Abstract
Importance
There is concern that nonmedical prescription opioid use is associated with an increased risk of later heroin use initiation in adolescents, but to our knowledge, longitudinal data addressing this topic are lacking.
Objective
To determine whether nonmedical prescription opioid use is associated with subsequent initiation of heroin use in adolescents.
Design, Setting, and Participants
This prospective longitudinal cohort study conducted in 10 high schools in Los Angeles, California, administered 8 semiannual surveys from 9th through 12th grade that assessed nonmedical prescription opioid use, heroin use, and other factors from October 2013 to July 2017. Students were baseline never users of heroin recruited through convenience sampling. Cox regression models tested nonmedical prescription opioid use statuses at survey waves 1 through 7 as a time-varying and time-lagged regressor and subsequent heroin use initiation across waves 2 to 8 as the outcome.
Exposures
Self-reported nonmedical prescription opioid use (past 30-day [current] use vs past 6-month [prior] use without past 30-day use vs no past 6-month use) at each wave from 1 to 7.
Main Outcomes and Measures
Self-reported heroin use initiation (yes/no) during waves 2 to 8.
Results
Of 3298 participants, 1775 (53.9%) were adolescent girls, 1563 (48.3%) were Hispanic, 548 (17.0%) were Asian, 155 (4.8%) were African American, 529 (16.4%) were non-Hispanic white, and 220 (6.8%) were multiracial. Among baseline never users of heroin in ninth grade with valid data (3298 [97% of cohort enrollees]; mean [SD] age, 14.6 [0.4] years), the number of individuals with outcome data available at each follow-up ranged from 2987 (90.6%) to 3200 (97.0%). The mean per-wave prevalence of prior and current nonmedical prescription opioid use from waves 1 to 7 was 1.9% and 2.7%, respectively. Seventy students (2.1%) initiated heroin use during waves 2 to 8. Prior vs no (hazard ratio, 3.59; 95% CI, 2.14-6.01; P < .001) and current vs no (hazard ratio, 4.37; 95% CI, 2.80-6.81; P < .001) nonmedical prescription opioid use were positively associated with subsequent heroin use initiation. For no, prior, and current nonmedical prescription opioid use statuses at waves 1 to 7, the estimated cumulative probabilities of subsequent heroin use initiation by wave 8 (42-month follow-up) were 1.7%, 10.7%, and 13.1%, respectively. In covariate-adjusted models, associations were attenuated but remained statistically significant and current nonmedical prescription opioid use risk estimates were stronger than corresponding associations of nonopioid substance use with subsequent heroin use initiation.
Conclusions and Relevance
Nonmedical prescription opioid use was prospectively associated with subsequent heroin use initiation during 4 years of adolescence among Los Angeles youth. Further research is needed to understand whether this association is causal.
This cohort study explores the association between nonmedical prescription opioid use and an increased risk of future heroin use initiation among adolescents in Los Angeles, California.
Introduction
Many adolescents access prescription opioids from friends or relatives for nonmedical reasons.1,2,3 Concern is heightened if adolescent nonmedical prescription opioid use is associated with an increased risk of subsequently initiating heroin, a drug with substantial addiction potential that poses extensive medical, psychological, social, and legal consequences.4,5,6,7
Prescription opioids and heroin share neuropharmacological actions through a stimulation of endogenous opioid receptors and the activation of the brain’s reward circuit.8,9 Experiencing the euphoric effects of nonmedical prescription opioid use could be associated with an increased inclination for youths to try other opioid drugs, including heroin. Cross-sectional analyses of adolescents’ retrospective reports indicate an association between prior nonmedical prescription opioid use and later heroin use initiation.10,11,12 As cross-sectional research is limited, the absence of longitudinal, prospective data on this topic is an important gap in the literature. This prospective longitudinal cohort study estimated the association between nonmedical prescription opioid use and subsequent heroin use initiation during 42 months of follow-up in high school students in Los Angeles, California.
Methods
Participants and Procedures
Data were drawn from a longitudinal cohort survey of behavioral health that included students from 10 Los Angeles–area suburban and urban high schools recruited by convenience sampling and described previously.13 Approximately 40 public high schools in the Los Angeles metropolitan area were approached for participation. Schools were chosen because of their diverse demographic characteristics and proximity. Ten schools agreed to participate; of these, 8 (80%) were in urban areas and 2 (20%) were in suburban areas. Ninth-grade students not enrolled in special education at the participating schools in 2013 with written active student assent and parental consent were enrolled (N = 3396). The data collection involved 8 assessments conducted every 6 months from baseline (wave 1; fall 2013, 9th grade; 3383 [99.6%] surveyed; mean [SD] age, 14.5 [0.40] years) through 42-month follow-up (wave 8; spring 2017, 12th grade; 3140 [92.5%] surveyed; mean [SD] age, 17.9 [0.39] years). Paper-and-pencil surveys were administered in students’ classrooms. Students not in class completed surveys by telephone, internet, or mail (numbers of phone/internet/mail surveys across follow-ups ranged from 49-468). The University of Southern California institutional review board approved the study.
Measures
Nonmedical Prescription Opioid and Heroin Use
Past 6-month (yes/no) and past 30-day (forced choice with 9 options ranging 0-30 days) use of prescription opioids (described as “prescription painkillers to get high [eg, Vicodin, Oxycontin, Percocet, Codeine]”) and other substances were measured in separate questions derived from previously validated surveys.14,15 Nonmedical prescription opioid use statuses at each wave were coded into a trichotomous variable (past 30-day [current] use vs past 6-month [prior] use without past 30-day use vs no past 6-month use). Participants reported ever heroin use (yes/no) at baseline and past 6-month heroin use (yes/no) at each semiannual follow-up.
Covariates
Factors previously associated with nonmedical prescription opioid or heroin use considered peripheral to the putative risk pathway were included as a priori covariates. Each of the measures described hereafter have demonstrated adequate psychometric properties in youth.16,17,18,19,20,21,22,23,24,25,26
Nonopioid Substance Use and Sociodemographic and Environmental Factors
Marijuana, alcohol, cigarettes, and other substance (eg, cocaine, methamphetamine, inhalants, and nonmedical prescription stimulants) use were assessed and coded in the same fashion as nonmedical prescription opioid use as time-varying covariates. Baseline age, sex, highest parental education level, and family living situation were measured using investigator-defined forced-choice items (Table 1).17,20 Because opioid use may differ by race/ethnicity,11 self-reported race/ethnicity (American Indian/Alaska Native, Asian, black/African American, Hispanic/Latino, Native Hawaiian/Pacific Islander, white, multiethnic/multiracial, or other) was included. A family history of smoking, alcohol problems, or drug problems was also measured (yes/no). A 4-item parental monitoring questionnaire was administered at wave 3 (α = .82),21,22,23 yielding a composite score ranging from 1 (no monitoring) to 4 (regular monitoring).
Table 1. Sample Characteristics of Baseline Never Users of Heroin and Comparisons by Heroin Use Initiation Over Follow-upa.
| Characteristics | Total Analytic Sample (N = 3298)b | Comparisons by Heroin Use Initiation Over Follow-up | ||
|---|---|---|---|---|
| Never Used Heroin (n = 3228) | Initiated Heroin Use (n = 70) | P Value | ||
| Female sex, No. (%) | 1775 (53.9) | 1754 (54.4) | 21 (30.0) | <.001c |
| Age, mean (SD), y | 14.6 (0.4) | 14.61 (0.40) | 14.6 (0.4) | .50d |
| Race/ethnicity, No. (%) | ||||
| Hispanic | 1563 (48.3) | 1527 (48.3) | 36 (52.2) | .97c |
| Asian | 548 (17.0) | 537 (17.0) | 11 (15.9) | |
| African American | 155 (4.8) | 153 (4.8) | 2 (2.9) | |
| Non-Hispanic white | 529 (16.4) | 518 (16.4) | 11 (15.9) | |
| Multiracial | 220 (6.8) | 215 (6.8) | 5 (7.2) | |
| Othere | 218 (6.7) | 214 (6.8) | 4 (5.8) | |
| Parent(s) without high school diploma, No. (%) | 376 (13.2) | 367 (13.2) | 9 (14.3) | .71c |
| Living with both parents, No. (%) | 2080 (63.7) | 2032 (63.6) | 48 (69.6) | .37c |
| Family substance use history, No. (%) | 2190 (70.0) | 2149 (70.1) | 41 (64.1) | .18c |
| Parental monitoring, mean (SD)f | 3.06 (0.69) | 3.06 (0.69) | 2.84 (0.83) | .03d |
| Delinquent behavior, mean (SD)g | 1.43 (0.47) | 1.42 (0.45) | 1.81 (0.91) | <.001d |
| Depressive symptoms, No. (%)h | 1161 (35.7) | 1131 (35.5) | 30 (42.9) | .21c |
| Generalized anxiety symptoms, No. (%)i | 701 (22.4) | 683 (22.3) | 18 (27.7) | .30c |
| UPPS, mean (SD)j | ||||
| Negative urgency | 1.78 (0.60) | 1.77 (0.59) | 2.01 (0.72) | .002d |
| Positive urgency | 1.78 (0.61) | 1.77 (0.60) | 1.98 (0.75) | .01d |
Unless otherwise specified, wave 1 data reported. Variables depicted were also time-invariant covariates in the multivariable-adjusted regression model.
Of the 3298 baseline never users of heroin, the number of users with available data (and corresponding denominator for % values) ranged from 2845 (86.3%) to 3296 (99.9%).
P values from the χ2 test for comparisons of proportions by group.
P values from the analysis of variance test of mean scores by group.
American Indian/Alaska Native, Native Hawaiian/Pacific Islander, or other responses constituted the “Other” race/ethnicity category.
Score ranges from 1 to 4, with higher scores indicating greater perceived parental monitoring. The mean rating from 1 (no monitoring) to 4 (regular monitoring) across 4 items. Data from wave 3.
Score ranges from 1 to 6, with higher scores indicating a greater average frequency of engaging in 11 different delinquent behaviors. Each behavior is rated from 1 (never) to 6 (10 or more times) for 11 behaviors.
Screen results were positive (vs negative) for mild to moderate depressive symptoms or higher on the Center for Epidemiologic Studies Depression Scale.
Screen results were positive (vs negative) for subclinical or clinical generalized anxiety symptoms on the Revised Anxiety and Depression Scale.
Score ranges from 1 to 4, with higher scores indicating an impulsive tendency to act rashly during negative emotional (negative urgency) or positive emotional (positive urgency) states for respective subscales of the UPPS measure of impulsive personality.
Intrapersonal Factors
Baseline emotional symptoms were assessed using the Center for Epidemiologic Studies Depression Scale24 (α = .81) and Revised Child Anxiety and Depression Scale generalized anxiety disorder25,26 subscale (6 items; α = .91), which were dichotomously coded as “symptomatic” (scoring at or higher than clinical cutoff values) vs “non-symptomatic.” Subscales of the UPPS Impulsive Behavior Scale,18 a measure of impulsive personality traits assessing the tendency to act rashly during negative (negative urgency [12 items; eg, “I do impulsive things that I later regret”; α = .89]) and positive (positive urgency [14 items; eg, “I act without thinking when I am really excited”; α = .94]) emotional states, were included. Delinquent behavior was measured with mean frequency ratings for engaging in 11 behaviors (eg, stealing, lying to parents; each item was rated 1 [never] to 6 [≥10 times]; α = .79) in the past 6 months.19
Statistical Analysis
After descriptive analyses, Cox regression models tested nonmedical prescription opioid use statuses at waves 1 to 7 as having a time-varying and time-lagged association with heroin use initiation at waves 2 to 8 in baseline never–heroin users.27,28 This approach incorporated all waves of data on nonmedical prescription opioid use occurring before heroin use for every student (up to 7 waves for students who never used heroin by wave 7). For each available wave of nonmedical prescription opioid use data, heroin use initiation data were used at all ensuing waves spanning from the immediately subsequent wave (6 months later) to the last follow-up (up to 42 months of follow-up). Follow-up heroin use data during waves 2 through 8 were regressed on wave 1 nonmedical prescription opioid use data, follow-up heroin use in waves 3 to 8 was regressed on wave 2 nonmedical prescription opioid use, and so on. When a student had initiated heroin use, nonmedical prescription opioid use data at that wave and ensuing waves were not additionally incorporated into the model estimates. We tested a univariable unadjusted model that included time-varying past 6-month nonmedical prescription opioid use status as the sole regressor. We also tested a multivariable model that included nonmedical opioid, alcohol, cigarette, cannabis, and other substance use for waves 1 through 7 as simultaneous time-varying regressors that additionally adjusted 12 time-invariant covariates listed previously. Substance use regressors were modeled categorically, producing hazard ratios (HRs) and 95% confidence intervals for associations of current (vs no) and prior (vs no) use status contrasts with subsequent heroin use initiation. We also tested head-to-head comparisons of whether the magnitude of HRs for nonmedical prescription opioid use were significantly different from corresponding HRs for other substance use regressors from the multivariable model using the χ2 difference test based on the log likelihood values derived from the maximum likelihood robust estimator.
Analyses were conducted in Mplus, version 7 (Muthén & Muthén), including school random effects to account for clustering.29,30 Missing data were addressed using a full information maximum likelihood estimation (available sample sizes of each variable are presented in Figure 1; eTables 1 and 2 in the Supplement).31 Statistical significance was determined after Benjamini-Hochberg multiple-testing corrections to raw P values (2-tailed) of each substance use regressor estimate to control studywise false-discovery rates at .05.32
Figure 1. Study Accrual Flowchart.
Results
Sample
Of 4100 eligible 9th grade students, 3396 (82.8%) provided assent and parental consent for enrollment (Figure 1). Following prior strategies for minimizing invalid responses,33 students reporting questionable substance use patterns (ie, everyday use of 6 substances in the past 30 days) or biologically implausible body mass indexes (calculated as weight in kilograms divided by height in meters squared) were excluded (77 [2.3%]). Baseline ever users of heroin (21 [0.6%]) were excluded, resulting in an analytic sample of 3298. Cohort enrollees who were excluded from the analysis differed from the analytic sample on several characteristics (eTable 3 in the Supplement). The numbers of individuals with outcome data available at each follow-up wave ranged from 2987 (90.6%) to 3200 (97.0%).
Descriptive
The sample (3298 [97% of cohort enrollees]; 1775 (53.9%) adolescent girls; mean [SD] age, 14.6 [0.4] years) was sociodemographic diverse (Table 1). Seventy students (2.1%) initiated heroin use during the 42-month follow-up. Students who initiated heroin use were more likely to be male, report lower parental monitoring, and report higher baseline delinquent behavior, negative urgency, and positive urgency.
Across waves 1 through 7, 596 students (18.1%) reported nonmedical prescription opioid use at 1 or more waves. The mean per-wave percentages of prior and current nonmedical prescription opioid use during waves 1 to 7 were 1.9% (range, 25 [0.8%] to 114 [3.5%]) and 2.7% (range, 56 [1.7%] to 112 [3.5%]) (eTable 2 in the Supplement). Most students reporting past 30-day nonmedical prescription opioid use reported use for 9 days or fewer (eFigure 1 in the Supplement).
Associations Between Nonmedical Prescription Opioid Use and Subsequent Heroin Use Initiation
The univariable unadjusted Cox regression model found that prior vs no (HR, 3.59; 95% CI, 2.14-6.01) and current vs no (HR, 4.37; 95% CI, 2.80-6.81) time-varying nonmedical prescription opioid use statuses for waves 1 through 7 were associated with an increased likelihood of subsequent heroin use initiation for waves 2 to 8 (Table 2). For no, prior, and current nonmedical prescription opioid use statuses, the estimated unadjusted cumulative probabilities of subsequent heroin use initiation by the final 42-month follow-up (wave 8) were 1.7%, 10.7%, and 13.1%, respectively (estimated hazard curves can be found in Figure 2).
Table 2. Associations of Nonmedical Prescription Opioid and Nonopioid Substance Use With Subsequent Heroin Use Initiationa.
| Time-Varying Regressors, Waves 1 to 7 | Associations With Subsequent Heroin Use Initiation, Waves 2 to 8 | |
|---|---|---|
| Hazard Ratio (95% CI) | P Value | |
| Univariable Unadjusted Modelb | ||
| Prior (vs no) nonmedical prescription opioid usec | 3.59 (2.14-6.01) | <.001d |
| Current (vs no) nonmedical prescription opioid usec | 4.37 (2.80-6.81) | <.001d |
| Multivariable Adjusted Modele | ||
| Prior (vs no) usec | ||
| Nonmedical prescription opioid use | 2.09 (1.14-3.83) | .02d |
| Cannabis use | 1.54 (0.89-2.65) | .12 |
| Alcohol use | 1.76 (1.04-2.98) | .04 |
| Cigarette use | 0.93 (0.52-1.67)f | .82 |
| Other substance useg | 2.20 (1.45-3.33) | <.001d |
| Current (vs no) usec | ||
| Nonmedical prescription opioid use | 3.18 (1.68-6.02) | <.001d |
| Cannabis use | 1.68 (1.02-2.82)h | .04 |
| Alcohol use | 2.04 (1.10-3.92)h | .03 |
| Cigarette use | 0.75 (0.35-1.59)h | .45 |
| Other substance useg | 1.54 (0.92-2.62)h | .10 |
Baseline never users of heroin (N = 3298).
Cox regression hazards model including only nonmedical prescription opioid use as a time-varying (time-lagged) regressor with school random effects.
No use = no past 6-month use; prior use = past 6-month use without past 30-day use; current use = past 30-day use.
Statistically significant after Benjamini-Hochberg corrections for multiple testing to control the false-discovery rate at .05 (based on a 2-tailed corrected P value).
Cox regression hazards model that included 5 simultaneous time-varying (time-lagged) substance use regressors adjusted for time-invariant covariates (ie, baseline age, sex, race/ethnicity, parental education level, family living situation, family substance use history, delinquent behavior, depressive symptoms, anxiety, negative urgency, and positive urgency and parental monitoring at wave 3) with school random effects.
Statistically significant difference in the magnitude of the hazard ratio for the prior (vs no) use contrast of a respective substance and the hazard ratio for the prior (vs no) use contrast of nonmedical prescription opioid use from χ2 difference test using the log-likelihood values with the maximum likelihood robust estimator (detailed statistics from the tests are presented in eTable 5 in the Supplement).
Use of any substance to get high other than opioids, marijuana, cigarettes, or alcohol.
Statistically significant difference in the magnitude of the hazard ratios for the current (vs no) use contrast of a respective substance and the hazard ratio for the current (vs no) use contrast of nonmedical prescription opioid use from χ2 difference test using the log-likelihood values with the maximum likelihood robust estimator (detailed statistics from the tests are presented in eTable 5 in the Supplement).
Figure 2. Estimated Hazard Curves for Heroin Use Initiation by Nonmedical Prescription Opioid Use Status in Preceding Waves.
The horizontal axis depicts 8 semiannual assessments from wave 1 (fall 9th grade, 2013; mean [SD] age, 14.5 [0.40] years) to wave 8 (spring 12th grade, 2017; mean [SD] age, 17.9 [0.39] years). The vertical axis depicts the estimated unadjusted cumulative probability of heroin use initiation at each follow-up wave; the estimates were reported by opioid use status at the preceding waves. The cumulative estimated probabilities of heroin use initiation at the final assessment (wave 8) were 1.7%, 10.7%, and 13.1% for no, prior, and current nonmedical prescription opioid use statuses, respectively, in the preceding waves. No use = no past 6-month use; prior use = past 6-month use without past 30-day use; and current use = past 30-day use.
In the multivariable model adjusted for time-varying past 6-month use of nonopioid substances and time-invariant covariates, the associations of prior vs no (HR, 2.09; 95% CI, 1.14-3.83) and current vs no (HR, 3.18; 95% CI, 1.68-6.02) time-varying nonmedical prescription opioid use statuses with subsequent heroin use initiation remained statistically significant but were attenuated in association with the unadjusted results. In this model, the only time-varying nonopioid substance use status variable significantly associated with subsequent heroin initiation was the prior vs no use contrast for the “other substance” variable (HR, 2.20; 95% CI, 1.45-3.33; Table 2). Female sex and a family history of substance use were associated with heroin use initiation in the multivariable model (eTable 4 in the Supplement).
The HR estimates of the association of current vs no use statues with subsequent heroin use initiation was significantly stronger for nonmedical prescription opioids than for alcohol, cannabis, cigarettes, or other nonopioid drugs (∆χ2/df ≥ 3.91, P ≤ .03; Table 2; eTable 5 in the Supplement). For prior vs no use contrasts, nonmedical prescription opioid HRs were significantly stronger than the cigarette smoking HRs but not different than alcohol, cannabis, and other drug use HRs (eTable 5 in the Supplement).
Sensitivity Analyses
Sensitivity analyses found that the primary results were not substantively changed if different approaches to handling missing data and data clustering were applied or if the results were retested, including invalid responders in the analytic sample. The hierarchical model adjusting for clustering within students, schools, and urban vs (sub)urban areas did produce slightly more conservative yet similarly significant estimates (eResults and eTables 6-8 in the Supplement). After eliminating 14 youths reporting opioid use before high school (eTable 9 and eFigure 2 in the Supplement), youth reporting no, prior, and current nonmedical prescription opioid use exhibited similar estimated cumulative probabilities of subsequent heroin use initiation to the primary analysis.
Discussion
This study provides new evidence of a prospective association between nonmedical prescription opioid use and an increased risk of future heroin use initiation among adolescents. Graded patterns of association with heroin use were observed for nonmedical prescription opioid use (ie, the risk was successively higher for nonuse, followed by prior use and then current use); graded associations were not observed for the use of nonopioid substances. To our knowledge, the only other published prospective longitudinal study examining the association of nonmedical prescription opioid use and later heroin use initiation was conducted in military veterans and also found evidence of a positive association.34
During the age period captured in this study (age 14-18 years), teenagers are exposed to new peers and often enter the workforce alongside adult coworkers, both of whom may provide access to licit and illicit drugs.35,36,37 Furthermore, this period is marked by imbalanced neural development in which brain pathways underlying pleasure seeking mature more rapidly than those underpinning decision-making skills.38 For these reasons, the risk of nonmedical prescription opioid exposure and heroin use initiation may escalate during mid to late adolescence.10
Prior studies of the association between nonmedical prescription opioid use and heroin use in adolescents used cross-sectional designs and retrospective reports of substance use.10,11,12 Such methods are subject to recall errors and reporting biases that may produce imprecise association estimates and inaccurate inferences regarding the ordering of nonmedical prescription opioid and heroin use. By using a prospective longitudinal design that excluded ever users of heroin at baseline, to our knowledge, this study is the first to establish the temporal precedence of adolescent nonmedical prescription opioid use in association with subsequent heroin use. Further, recall errors were reduced by repeated assessments and brief interwave intervals. Additionally, previous adolescent studies10,11,12 excluded salient covariates that may be associated with the risk of heroin use, including impulsive personality traits, parental monitoring, and delinquency. Finally, prior youth studies of this topic analyzed data collected in 2014 or earlier.10,11,12 Follow-up in this study spanned from 2013 to 2017, a recent historical period in which opioid-associated overdoses in US teenagers increased after several prior years of decline.39,40
The observed association is subject to 2 explanations: (1) a common liability to using nonmedical prescription opioids and heroin, or (2) nonmedical prescription opioid use directly increases the risk of heroin use initiation. Sources of common liability to using both drugs include an exposure to environments in which access to prescription opioids and heroin is high and parental restrictions are absent.41 An endogenous psychological or genetic disposition toward impulsive behavior or engaging in rebellious acts may also produce a common liability.42 Possible confounding influences were addressed by adjusting for covariates indicative of common liability. While adjusted estimates were reduced, suggesting common liability accounted for some of the association, current and prior nonmedical prescription opioid use remained significantly associated with 3.18 (95% CI, 1.68-6.02) and 2.09 (95% CI, 1.14-3.83) greater HRs of heroin use initiation. Although unmeasured confounding cannot be ruled out, the results suggest that a common liability may not entirely explain the association between nonmedical prescription opioid use and the subsequent heroin use initiation observed in this study.
Specificity in the study results also points toward a direct risk pathway from nonmedical prescription opioid use to heroin use initiation. As well as remaining robust to control for nonopioid substance use, the risk estimate for time-varying current nonmedical prescription opioid use was significantly stronger than the corresponding risk estimate for time-varying current use of nonopioid drugs in head-to-head comparisons. Thus, it is unlikely that the observed association is entirely a by-product of a nonspecific liability toward any drug use, including vulnerabilities that may vacillate across adolescence (ie, those that are time-varying). Additionally, the association between nonmedical prescription opioid use and heroin use initiation followed a graded (dose response–like) pattern, with larger HRs for current rather than prior use statuses. For the prior vs no use HRs, the HRs were comparable for other substances and opioids. For current use, the HR became larger for opioids but did not for other substances. Thus, graded patterns of association with heroin use were largely not observed for the use of other substances. As recency of use may be a proxy for the extent of exposure, these findings imply that adolescents with higher exposure to nonmedical opioid use are more likely to initiate heroin than those with lower exposure to nonmedical prescription opioids and imply that the recency of nonmedical prescription opioid use is also associated with later heroin use initiation. Finally, incidence of the reverse sequence was negligible; only 3 youths used heroin and later initiated prescription opioid use, precluding formal tests of a reverse association.
Although the observational design of this study precludes definitive causal inferences, the pattern of the results warrants considering plausible mechanisms of risk from nonmedical prescription opioid exposure to subsequent heroin use initiation. Adults with long-term prescription opioid use who develop opioid dependence report transitioning to heroin to alleviate opioid withdrawal symptoms when access to prescription opioids becomes difficult or costly.5,6 In the current adolescent sample, the daily use patterns characteristic of severe opioid use disorder were uncommon, suggesting that the desire to alleviate opioid withdrawal mediated by opioid dependence was not a highly common mechanism of the transition to heroin use. Prescription opioids can produce powerful euphoric effects, particularly when used nonmedically and at higher doses.8,9 It is possible that youths who enjoy the euphoric effects from nonmedical prescription opioid use may become inclined to try heroin because of a desire to experience similar opioidergic intoxicating effects at a higher potency.5 If evidence of a direct risk pathway from nonmedical opioid use to heroin use initiation in adolescence were to be identified in future research, measures to prevent youths from accessing prescription opioids merit consideration as a public health priority.
Limitations
First, exposure was operationalized by survey questions concatenating the use of several opioid compounds, leaving unclear which opioid drug and whether polyopioid use was associated with heroin use initiation. Second, medical use of prescription opioids was not assessed; whether nonmedical use stemmed from medical use of opioids was not captured in this study. While nonmedical use was defined in the survey at several points, it is possible that some students may have misread the survey instructions. Although the most common prescription opioids were included as examples in the survey item, it is also possible that some of the opioid compounds students used were not mentioned (eg, tramadol). Third, characteristic of high school student samples,34 the prevalence of frequent nonmedical prescription opioid use was low, precluding analyses of whether daily vs nondaily use differentiates the likelihood of heroin use initiation. Future studies of nonmedical prescription opioid use in youth will need to be enriched for frequent users using a different sampling strategy (eg, clinical populations seeking substance use treatment). Fourth, the method of drug administration was also not assessed; whether different risk estimates are found for opioid smoking, inhalation, or injection is worthy of study.5
There may also be limitations to the generalizability of this study’s findings. The prevalence of nonmedical prescription opioid and heroin use in this study was larger than figures observed in some nationally representative cross-sectional surveys.10,11,12 In this study, all youth were followed up after 9th grade, including teenagers frequently absent from class or who later discontinued school and might be underrepresented in national surveys. The repeated assessment in this study may capture the incidence of reported substance use missed in single point surveys. Like any observational study, there is a potential risk of bias when considering participant dropout, making our estimates more conservative, as dropouts likely have a higher risk of substance use. Additionally, this Los Angeles area youth sample is a convenience sample that is more sociodemographically diverse than the overall US population, and the (sub)urban backdrop of this study diverges from some rural areas where opioid use is widespread. Although prior cross-sectional research has found that the association between nonmedical prescription opioid use and heroin use initiation does not differ by race/ethnicity or income,10 further investigation in geographically heterogenous samples is warranted.
Conclusions
Nonmedical prescription opioid use was prospectively associated with subsequent heroin use initiation during 4 years of adolescence among Los Angeles youth. Further research is needed to understand whether this association is causal.
eResults.
eTable 1. Available data for each time-invariant covariate
eTable 2. Prevalence of substance use statuses from waves 1 to 7 based on available data at each wave
eTable 3. Characteristics of students included versus excluded from the primary analytic sample
eTable 4. Associations of time-invariant covariates with heroin use initiation in multivariable model
eTable 5. Comparison of hazard ratios between nonmedical prescription opioid use and other substance use covariates
eTable 6. Associations of nonmedical prescription opioid use with subsequent heroin use initiation using alternate methods of handling missing data
eTable 7. Associations of nonmedical prescription opioid and other substance use with subsequent heroin use initiation including potentially invalid responders
eTable 8. Associations of nonmedical prescription opioid and other substance use with subsequent heroin use initiation using alternative method of addressing clustering effects
eTable 9. Nonmedical prescription opioid use at baseline
eFigure 1. Past 30-day nonmedical prescription opioid use frequency level distributions
eFigure 2. Estimated hazard curves for heroin use initiation by nonmedical prescription opioid use status in preceding waves, eliminating ever use of nonmedical prescription opioids at baseline
References
- 1.Fortuna RJ, Robbins BW, Caiola E, Joynt M, Halterman JS. Prescribing of controlled medications to adolescents and young adults in the United States. Pediatrics. 2010;126(6):-. doi: 10.1542/peds.2010-0791 [DOI] [PubMed] [Google Scholar]
- 2.Boyd CJ, McCabe SE, Cranford JA, Young A. Adolescents’ motivations to abuse prescription medications. Pediatrics. 2006;118(6):2472-2480. doi: 10.1542/peds.2006-1644 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.National Institute of Drug Abuse Drug facts: prescription and over-the-counter medications. http://www.drugabuse.gov/publications/drugfacts/prescription-over-counter-medications. Accessed May 28, 2018.
- 4.Jones CM, Logan J, Gladden RM, Bohm MK. Vital signs: demographic and substance use trends among heroin users—United States, 2002-2013. MMWR Morb Mortal Wkly Rep. 2015;64(26):719-725. [PMC free article] [PubMed] [Google Scholar]
- 5.Lankenau SE, Teti M, Silva K, Jackson Bloom J, Harocopos A, Treese M. Initiation into prescription opioid misuse amongst young injection drug users. Int J Drug Policy. 2012;23(1):37-44. doi: 10.1016/j.drugpo.2011.05.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lankenau SE, Teti M, Silva K, Bloom JJ, Harocopos A, Treese M. Patterns of prescription drug misuse among young injection drug users. J Urban Health. 2012;89(6):1004-1016. doi: 10.1007/s11524-012-9691-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Compton WM, Jones CM, Baldwin GT. Relationship between nonmedical prescription-opioid use and heroin use. N Engl J Med. 2016;374(2):154-163. doi: 10.1056/NEJMra1508490 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Maher CE, Martin TJ, Childers SR. Mechanisms of mu opioid receptor/G-protein desensitization in brain by chronic heroin administration. Life Sci. 2005;77(10):1140-1154. doi: 10.1016/j.lfs.2005.03.004 [DOI] [PubMed] [Google Scholar]
- 9.Kreek M. Molecular and cellular neurobiology and pathophysiology of opiate addiction In: Davis KLE, ed. Neuropsychopharmacology: The Fifth Generation of Progress. Philadelphia, PA: Lippincott Williams & Wilkins; 2002:1491-1506. [Google Scholar]
- 10.Cerdá M, Santaella J, Marshall BDL, Kim JH, Martins SS. Nonmedical prescription opioid use in childhood and early adolescence predicts transitions to heroin use in young adulthood: a national study. J Pediatr. 2015;167(3):605-12.e1, 2. doi: 10.1016/j.jpeds.2015.04.071 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Palamar JJ, Shearston JA, Dawson EW, Mateu-Gelabert P, Ompad DC. Nonmedical opioid use and heroin use in a nationally representative sample of us high school seniors. Drug Alcohol Depend. 2016;158:132-138. doi: 10.1016/j.drugalcdep.2015.11.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Muhuri PK, Gfroerer JC, Davies MC Associations of nonmedical pain reliever use and initiation of heroin use in the United States. https://www.samhsa.gov/data/sites/default/files/DR006/DR006/nonmedical-pain-reliever-use-2013.htm. Accessed May 28, 2018.
- 13.Leventhal AM, Strong DR, Kirkpatrick MG, et al. Association of electronic cigarette use with initiation of combustible tobacco product smoking in early adolescence. JAMA. 2015;314(7):700-707. doi: 10.1001/jama.2015.8950 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Eaton DK, Kann L, Kinchen S, et al. ; Centers for Disease Control and Prevention (CDC) . Youth risk behavior surveillance—United States, 2009. MMWR Surveill Summ. 2010;59(5):1-142. [PubMed] [Google Scholar]
- 15.Miech RA, Johnston LD, O’Malley PM, Bachman JG, Schulenberg JE Monitoring the future: national survey results on drug use, 1975-2014. http://www.monitoringthefuture.org/pubs/monographs/mtf-overview2014.pdf. Accessed May 28, 2018.
- 16.Leventhal AM, Cho J, Stone MD, et al. Associations between anhedonia and marijuana use escalation across mid-adolescence. Addiction. 2017;112(12):2182-2190. doi: 10.1111/add.13912 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Schiavon S, Hodgin K, Sellers A, et al. Medical, psychosocial, and treatment predictors of opioid overdose among high risk opioid users. Addict Behav. 2018;86:51-55. doi: 10.1016/j.addbeh.2018.05.029 [DOI] [PubMed] [Google Scholar]
- 18.Whiteside SP, Lynam DR. The Five Factor Model and impulsivity: using a structural model of personality to understand impulsivity. Pers Individ Dif. 2001;30:669-689. doi: 10.1016/S0191-8869(00)00064-7 [DOI] [Google Scholar]
- 19.Thompson MP, Ho CH, Kingree JB. Prospective associations between delinquency and suicidal behaviors in a nationally representative sample. J Adolesc Health. 2007;40(3):232-237. doi: 10.1016/j.jadohealth.2006.10.016 [DOI] [PubMed] [Google Scholar]
- 20.Larance B, Gisev N, Cama E, et al. Predictors of transitions across stages of heroin use and dependence prior to treatment-seeking among people in treatment for opioid dependence. Drug Alcohol Depend. 2018;191:145-151. doi: 10.1016/j.drugalcdep.2018.03.056 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.DiClemente RJ, Wingood GM, Crosby R, et al. Parental monitoring: association with adolescents’ risk behaviors. Pediatrics. 2001;107(6):1363-1368. doi: 10.1542/peds.107.6.1363 [DOI] [PubMed] [Google Scholar]
- 22.Fletcher AC, Steinberg L, Williams-Wheeler M. Parental influences on adolescent problem behavior: revisiting Stattin and Kerr. Child Dev. 2004;75(3):781-796. doi: 10.1111/j.1467-8624.2004.00706.x [DOI] [PubMed] [Google Scholar]
- 23.Kodl MM, Mermelstein R. Beyond modeling: parenting practices, parental smoking history, and adolescent cigarette smoking. Addict Behav. 2004;29(1):17-32. doi: 10.1016/S0306-4603(03)00087-X [DOI] [PubMed] [Google Scholar]
- 24.Radloff LS. The CES-D scale. Appl Psychol Meas. 1977;1:385-401. doi: 10.1177/014662167700100306 [DOI] [Google Scholar]
- 25.Chorpita BF, Yim L, Moffitt C, Umemoto LA, Francis SE. Assessment of symptoms of DSM-IV anxiety and depression in children: a revised child anxiety and depression scale. Behav Res Ther. 2000;38(8):835-855. doi: 10.1016/S0005-7967(99)00130-8 [DOI] [PubMed] [Google Scholar]
- 26.Chorpita BF, Moffitt CE, Gray J. Psychometric properties of the Revised Child Anxiety and Depression Scale in a clinical sample. Behav Res Ther. 2005;43(3):309-322. doi: 10.1016/j.brat.2004.02.004 [DOI] [PubMed] [Google Scholar]
- 27.Fisher LD, Lin DY. Time-dependent covariates in the Cox proportional-hazards regression model. Annu Rev Public Health. 1999;20:145-157. doi: 10.1146/annurev.publhealth.20.1.145 [DOI] [PubMed] [Google Scholar]
- 28.Bellera CA, MacGrogan G, Debled M, de Lara CT, Brouste V, Mathoulin-Pélissier S. Variables with time-varying effects and the Cox model: some statistical concepts illustrated with a prognostic factor study in breast cancer. BMC Med Res Methodol. 2010;10:20. doi: 10.1186/1471-2288-10-20 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Muthén LK, Muthén BO. Mplus: Statistical Analysis With Latent Variables. 6th ed Los Angeles, CA: Muthén & Muthén; 2010. [Google Scholar]
- 30.Vittinghoff E, Glidden DV, Shiboski SC, McCulloch CE. Regression Methods in Biostatistics: Linear, Logistic, Survival, and Repeated Measures Models. New York, NY: Springer; 2011. [Google Scholar]
- 31.Cassatt JC, Marini CP, Bender JW. The reversible reduction of horse metmyoglobin by the iron(II) complex of trans-1,2-diaminocyclohexane-N,N,N,n-tetraacetate. Biochemistry. 1975;14(25):5470-5475. doi: 10.1021/bi00696a014 [DOI] [PubMed] [Google Scholar]
- 32.Benjamini YHY. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Series B Stat Methodol. 1995;57:289-300. https://www.jstor.org/stable/2346101?seq=1#page_scan_tab_contents. Accessed November 18, 2018. [Google Scholar]
- 33.Miech RA, Johnston LD, O’Malley PM, Bachman JG, Schulenberg JE, Patrick ME Monitoring the future: national survey results on drug use, 1975-2017. http://www.monitoringthefuture.org/pubs/monographs/mtf-overview2017.pdf. Accessed May 28, 2018.
- 34.Banerjee G, Edelman EJ, Barry DT, et al. Non-medical use of prescription opioids is associated with heroin initiation among US veterans: a prospective cohort study. Addiction. 2016;111(11):2021-2031. doi: 10.1111/add.13491 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Needle R, McCubbin H, Wilson M, Reineck R, Lazar A, Mederer H. Interpersonal influences in adolescent drug use—the role of older siblings, parents, and peers. Int J Addict. 1986;21(7):739-766. doi: 10.3109/10826088609027390 [DOI] [PubMed] [Google Scholar]
- 36.Mason MJ, Zaharakis NM, Rusby JC, et al. A longitudinal study predicting adolescent tobacco, alcohol, and cannabis use by behavioral characteristics of close friends. Psychol Addict Behav. 2017;31(6):712-720. doi: 10.1037/adb0000299 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Wu LT, Schlenger WE, Galvin DM, Reineck R, Lazar A, Mederer H. The relationship between employment and substance use among students aged 12 to 17. J Adolesc Health. 2003;32(1):5-15. doi: 10.1016/S1054-139X(02)00447-0 [DOI] [PubMed] [Google Scholar]
- 38.Doremus-Fitzwater TL, Varlinskaya EI, Spear LP. Motivational systems in adolescence: possible implications for age differences in substance abuse and other risk-taking behaviors. Brain Cogn. 2010;72(1):114-123. doi: 10.1016/j.bandc.2009.08.008 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Curtin SC, Tejada-Vera B, Warner M Drug overdose deaths among adolescents aged 15–19 in the United States: 1999–2015. https://www.cdc.gov/nchs/data/databriefs/db282.pdf. Accessed May 28, 2018. [PubMed]
- 40.Gaither JR, Shabanova V, Leventhal JM. US national trends in pediatric deaths from prescription and illicit opioids, 1999-2016. JAMA Netw Open. 2018;1(8):e186558. doi: 10.1001/jamanetworkopen.2018.6558 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Schriber RA, Guyer AE. Adolescent neurobiological susceptibility to social context. Dev Cogn Neurosci. 2016;19:1-18. doi: 10.1016/j.dcn.2015.12.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Belsky J. Differential susceptibility to rearing influence: an evolutionary hypothesis and some evidence In: Ellis B, Bjorklund D, eds. Origins of the Social Mind: Evolutionary Psychology and Child Development. New York, NY: Guilford; 2005:139-163. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eResults.
eTable 1. Available data for each time-invariant covariate
eTable 2. Prevalence of substance use statuses from waves 1 to 7 based on available data at each wave
eTable 3. Characteristics of students included versus excluded from the primary analytic sample
eTable 4. Associations of time-invariant covariates with heroin use initiation in multivariable model
eTable 5. Comparison of hazard ratios between nonmedical prescription opioid use and other substance use covariates
eTable 6. Associations of nonmedical prescription opioid use with subsequent heroin use initiation using alternate methods of handling missing data
eTable 7. Associations of nonmedical prescription opioid and other substance use with subsequent heroin use initiation including potentially invalid responders
eTable 8. Associations of nonmedical prescription opioid and other substance use with subsequent heroin use initiation using alternative method of addressing clustering effects
eTable 9. Nonmedical prescription opioid use at baseline
eFigure 1. Past 30-day nonmedical prescription opioid use frequency level distributions
eFigure 2. Estimated hazard curves for heroin use initiation by nonmedical prescription opioid use status in preceding waves, eliminating ever use of nonmedical prescription opioids at baseline