Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2020 Feb 1.
Published in final edited form as: Exp Clin Psychopharmacol. 2018 Sep 27;27(1):87–95. doi: 10.1037/pha0000229

Underage drinking, alcohol dependence, and young people starting to use prescription pain relievers extra-medically: A zero-inflated Poisson regression model

Maria A Parker 1, James C Anthony 2
PMCID: PMC6355342  NIHMSID: NIHMS982675  PMID: 30265064

Abstract

Among young people who start using prescription pain relievers (PPR) for feeling-states such as ‘to get high’ or otherwise beyond boundaries intended by prescribers, the most recent epidemiological incidence estimates show 2%-to-9% with rapid-onset opioid dependence. In this work, we study recently active underage alcohol dependence as a susceptibility marker and estimate alcohol dependence-associated PPR rates of use, once use starts. In recent United States epidemiological samples, we identified 16,125 community-dwelling 12-to-20-year-olds with standardized assessments of both problem drinking and newly incident extra-medical PPR use. We applied zero-inflated Poisson (ZIP) regressions to estimate (a) alcohol dependence associations with susceptibility-to-persist after the very first occasion of extra-medical PPR use, and (b) the rate of PPR use, conditional on persistence. Underage drinkers with alcohol dependence were more susceptible to persistence in their extra-medical PPR use (p<0.001). In addition, given susceptibility-to-persistence, there was an alcohol dependence-associated excess rate of extra-medical PPR use (Risk Ratio = 1.3; 95% Confidence Interval = 1.1, 1.6). Using ZIP regressions, we can see that underage alcohol dependence signals membership in a susceptible-to-persistence class of extra-medical PPR users and excess rates of extra-medical use. Underage drinking can be an indicator of greater vulnerability to start and persist in extra-medical use of PPRs, particularly if presenting clinical features of alcohol dependence already are seen at or near time of first onset of such PPR use. For alcohol dependence-affected adolescent patients, non-drug pain management plans deserve consideration, with special surveillance if analgesic drugs are prescribed. Implications for genetic susceptibility research are discussed.

Keywords: opioids, adolescents, alcohol, dependence, prescription pain relievers

Studying patients or community residents who consume opioid prescription pain relievers (PPR), we can identify a subgroup of individuals who stay within the boundaries intended by the prescriber. On the other hand, there are individuals who start using PPRs to ‘get high’ or for other feeling states or in ways that fall outside the boundaries (i.e., ‘extra-medical users’). In an early article that presented United States (US) epidemiological estimates from 1990–92, roughly 1-in-11 adult extra-medical users of opioid PPR had developed an opioid dependence syndrome; corresponding estimates for ‘medical users’ who take PPR exactly as prescribed have not yet been published (Anthony, Warner, & Kessler, 1994). The same focus on extra-medical use has been sustained in recent community samples. For example, studying more than 8,000 US adults with extra-medical PPR use histories in 2005–2006, an estimated 14%–15% qualified as currently active cases of opioid dependence (Martins, Storr, Zhu, & Chilcoat, 2009).

In more recent studies, attention was shifted to newly incident users in US community samples of adolescents, 2002–2013. All of these young people had just started to use PPR extra-medically (within 1–12 months before assessment). According to estimates based on these newly incident PPR users, between 2%–9% had progressed to the point of opioid dependence within a 12-month interval after first use. Slightly larger opioid dependence rates were found when extra-medical PPR use had started in early adolescence. For example, peak risk for a transition from start of extra-medical PPR use to opioid dependence within 12 months was seen at mid-adolescence among 14-to-15-year-olds (Parker & Anthony, 2015) – one to two years earlier in adolescence than the peak risk for starting extra-medical PPR use (Meier, Troost, & Anthony, 2012).

A history of alcohol involvement, including alcohol dependence, seems to be important in whether opioid dependence develops once extra-medical use starts (Reyes-Gibby, Yuan, Wang, Yeung, & Shete, 2015; Schepis & Krishnan-Sarin, 2008; Sung, Richter, Vaughan, Johnson, & Thom, 2005). Among the unanswered questions is one about the timing of an alcohol dependence role. Namely, when does alcohol dependence start to function as a susceptibility marker or causal influence on progression from first extra-medical PPR use onward toward opioid dependence?

One possibility is that it has importance only in the earliest stages of extra-medical PPR involvement. For example, underage drinking might shape whether and when an adolescent has the first chance to experiment with PPR, as seems to be the case with chances to try cocaine (Wagner & Anthony, 2002). Another possibility is that underage drinking and adolescent-onset alcohol dependence are not at all salient early in this progression. Alcohol dependence might accelerate an opioid dependence process only after opioid dependence symptoms have started to become obvious.

In research on drug use, there is a precedent for using a zero-inflated Poisson (ZIP) regressions approach to investigate the earliest stages of drug-taking once it starts. ZIP models are useful in this context for two main reasons. First, the ZIP models are consistent with a precision medicine concept that individuals or population subgroups might vary in their susceptibility to engage in persistence of drug use, once a drug has been administered. For example, we can posit a conceptual model with two latent classes of drug users. One class includes individuals who are susceptible to continue to use a drug, once it has been tried for the first time (although some of these susceptible-to-persist individuals are observed with no use during the observation interval, perhaps because the chance to try a second time has not yet occurred). The other class includes individuals for whom this susceptibility to persist is at or near zero levels (and all are observed with zero days of use after the first-time use). Second, the ZIP models make it possible to estimate subgroup variation in the rates of use within the class with a susceptibility-to-persist (e.g., number of days or occasions of use during a specified interval of time after first use, with an allowance for susceptible-to-persist individuals observed with zero days of use during the observation interval). In work along these lines, Barondess and colleagues looked for subgroup variations in progressions beyond the first occasion of tobacco cigarette smoking. They turned to the ZIP regressions model and were surprised to discover no male-female differences of importance. Once tobacco smoking started, females were just as likely as males to persist in smoking, and there was no appreciable male-female variation in the smoking rate, given persistence (Barondess, Meyer, Boinapally, Fairman, & Anthony, 2010).

The regular Poisson model might be just as useful as ZIP modeling when an observed distribution of counts of drug-using days has a mean equal to its variance (i.e., without an inflation of the number of observations at the lowest count). However, in our research to date we have not seen drug use count distributions that fulfill the mean=variance assumption required for regular Poisson modeling.

A few investigators have split the count distribution into two parts and have studied the characteristics of users who used one time only and have estimated the rates of use for individuals using more than one time. This two-part modeling approach has a serious flaw in its estimation of rates because the susceptible-to-persist individuals who have not yet taken the drug a second time are excluded from the denominators of the rates in the second part and are considered only in the first part. As just mentioned, rate estimation based on ZIP modeling is set up to include these susceptible-to-persist individuals who are likely to go on and try the drug a second time but have not yet done so before the end of the stated observation interval.

Readers might know individuals who have taken a prescribed opioid PPR, soon experienced a dysphoric or other reaction, and decided never to repeat the experience. Conceptually, these individuals might belong in the first ZIP class – i.e., not susceptible to persist. For them, the first-time use is the first and last time the drug is used. Robust evidence of the existence of this class has been available since 1955, when Lasagna and colleagues described double-blind experiments with 20 opioid-naïve young men, who received injected solutions with placebo, heroin, and morphine administered in a random order on separate days. Still blinded to the identity of the drug, they answered a question about whether they would like to repeat the experience a second time. In the placebo trial, only one said he would not like to take that compound a second time (5%) and three said they would like a repeat. For heroin, the corresponding estimates were seven (35%) and four. The estimates for morphine were nine (45%) and two (Lasagna, Von Felsinger, & Beecher, 1955).

As for our hypothesis about alcohol dependence as a determinant of whether use of the PPR persists after an initial experience, we have no evidence from human experiments, but we can offer a basic behavioral perspective. The opioid dependence process can be conceptualized as an early count process with an accumulating count of PPR-using experiences. To the extent that there is a reinforcement process when the PPR is used extra-medically for the first time, a second-time use should be seen at some point; the PPR should be used again. Moreover, an alcohol dependence-affected subgroup of newly incident PPR users might be more likely to become engaged in what has been colloquially described as ‘chasing the high,’ rather than stopping at the first-time occasion of extra-medical PPR use (Keegan & Moss, 2008). An analogous theoretical formulation leads us expect the alcohol dependence syndrome to predict a higher rate of PPR use, conditional on persistence of such use.

This line of research has some potential clinical implications. Evidence on the alcohol dependence hypothesis might prove to be useful to pediatricians or other physicians prescribing opioids for pain relief. Knowledgeable about evidence of alcohol dependence-associated excess risk in the earliest stages of PPR experience, they might be prompted to prescribe non-drug or over-the-counter alternatives for pain relief in their patients with histories of problematic alcohol use. Or, office-dispensing might be constrained to few opioid PPR dosage units, with a trip to the pharmacy required if more units are needed.

Launching this study, we expected alcohol dependence cases to be under-represented among young people who engage in extra-medical use of PPRs a single time and then never try PPR use a second time (i.e., to be more likely to engage in repetitive drug use), and to be over-represented among those observed to persist. Second, among those who have engaged in extra-medical PPR use one time only, we conceptualized two latent classes: (a) one subclass of users who are ‘not-susceptible-to-persistence,’ by virtue of which they never use again, and (b) a second class of users who are ‘susceptible-to-persistence,’ and who already have been observed to progress beyond the first-time use or who might be expected to progress if follow-up assessments extend beyond the currently observed interval.

Methods

Research Design and Study Population

The research design involved an epidemiological cross-sectional survey approach, with ascertainment of newly incident extra-medical PPR users in adolescence – i.e., all use starting within 24 months prior to the survey assessment date. The study population consisted of non-institutionalized US civilian community residents aged 12-to-20-years-old, 2005–2014.

Samples

Multi-stage area probability samples were drawn and recruited for US National Surveys on Drug Use and Health (NSDUH) completed in each of the survey years. Consent/assent procedures governed by Institutional Review Board-approved protocols produced acceptable participation levels (generally >70%). The Substance Abuse and Mental Health Data Archive (SAMHDA) NSDUH public use datasets made available large sub-samples of each independently drawn survey sample (n~55,000 yearly). Detailed methods descriptions are available in online reports (e.g., last accessed April 2018, http://www.samhsa.gov/data/population-data-nsduh/reports). Data from 2015–16 have not been included due to major NSDUH methodology changes after 2014.

Figure 1 sums across 2005–2014, showing that aggregate NSDUH samples included 247,174 12-to-20-year-old participants, 31,182 who had ever used an extra-medical PPR, and 18,911 started their use within 24 months before assessment. A relatively small fraction (n=2,786) had missing data (e.g., on days of use) and were excluded. Remaining were 16,125 newly incident extra-medical PPR users in adolescence, all with extra-medical PPR use starting within the past 24 months.

Figure 1.

Figure 1

Open in a new tab

Flow chart showing the process used to enumerate newly incident extra-medical PPR users in the sample of 12-to-20-year-olds. Data from the National Surveys on Drug Use and Health (NSDUH), United States, 2005–2014.

Measures

NSDUH assessments were completed via audio computer self-interviews. Each interview covers multiple standardized multi-item modules on background variables (e.g., sex, age), drug use, and health topics. Newly incident users were defined by comparing each individual’s year of first extra-medical PPR use with the year of assessment for a value of < 24 months. To illustrate, the NSDUH survey asked, “Did you first use any prescription pain reliever that was not prescribed for you or that you took only for the experience or feeling it caused in [year-1] or [year]?” and “In what month in [year] did you first use any prescription pain reliever that was not prescribed for you or that you took only for the experience or feeling it caused?”

For this study, the outcome variable was the number of days of use of extra-medical PPRs during the 12-month interval before self-report assessment. Participants were asked, “Now think about the past 12 months, from [survey date] through today. We want to know how many days you've used any prescription pain reliever that was not prescribed for you or that you took only for the experience or feeling it caused during the past 12 months.”

The exposure of interest was recently active alcohol dependence, measured via the NSDUH multi-item module based on the Diagnostic and Statistical Manual for Mental Disorders, Fourth Edition (DSM-IV; American Psychiatric Association, 1994). We combined DSM-IV alcohol dependence responses with the alcohol recency survey question that asked, “How long has it been since you last drank an alcoholic beverage?” Respondents could answer, “Within the past 30 days”, “More than 30 days ago but within the past 12 months,” “More than 12 months ago,” or “Never used alcohol.” The recently active underage alcohol dependence cases, as observed, were contrasted with three other forms of underage alcohol involvement: (i) Never drank in lifetime (reference subgroup), (ii) Drank in lifetime, but not in the past year and no active alcohol dependence, and (iii) Drank in the past year but no active alcohol dependence.

Statistical Analysis

Standard Poisson ‘count’ regression modeling enables estimation of the degree to which co-occurring alcohol dependence might be associated with a young person’s rate of extra-medical PPR use during the initial one-year interval after onset of such use. When there is an inflation of the number of zeroes in a standard Poisson distribution, it is possible to specify a two-equation generalized linear model with an initial equation to express zero-inflation as a function of potentially explanatory covariates.

Our first analysis steps involved counting up the number of days of extra-medical PPR use during an interval of observation that started once newly incident extra-medical PPR use had occurred and plotting the frequency distribution of these counts. Doing so, we found that 39.3% of the newly incident users had engaged in extra-medical PPR use on one and only one occasion; thereafter zero use. This 39.3% value disclosed a zero-inflation, relative to the number of zeroes expected in a regular Poisson distribution, which prompted ZIP modeling using the two inter-related regression equations. In other words, the ‘persistent-use count’ was zero-inflated by the ‘susceptible-to-persistence’ users who were observed before they have tried a PPR on a second day after the day of first-time use. (These individuals might wish to try again, but there has been no local availability of PPR, and their use by the assessment date has been restricted to a single occasion.)

The ZIP ‘Inflate’ Equation

The first of two ZIP equations indexes subgroups or ‘subclasses’ that are more or less likely to be ‘not-susceptible-to-persistence’ users, and for whom the observed data indicates zero PPR-using days after the first trial extra-medical use of PPR. This ZIP equation can be called an ‘inflate’ equation, which serves as a reminder that this equation can be used to show which subgroups are contributing to an inflation of the count of zeroes (Hilbe, 1999). Membership in the ‘not-susceptible-to-persistence’ class versus the ‘susceptible-to-persistence class’ is binary, so the familiar logistic regression model provides slope estimates for each covariate of interest. These estimates can be exponentiated to yield a familiar odds ratio (OR) as a measure of association.

The ZIP ‘Count’ Equation

The second ZIP equation serves to disclose variation in rates of extra-medical PPR use, given membership in the ‘susceptible-to-persistence’ class. This equation resembles the standard Poisson ‘count’ regression equation. As with standard Poisson regression, this count equation serves to estimate event rate ratios (RR) in a contrast of alternative subgroups, such as alcohol dependence cases versus non-cases. In the ZIP context, RR estimation is conditional on ‘susceptible-to-persistence’ class membership. Solving the ‘count’ equation yields the estimates on the natural log scale. After exponentiation, the resulting familiar RR conveys the degree of excess or reduced rate of PPR use in each subgroup of interest versus a designated reference subgroup – here, the rate for alcohol dependence cases versus the rate for alcohol dependence non-cases.

Analysis Plan

We started with covariate terms for alcohol involvement and for survey year in the initial ZIP regression model. Then, we introduced sex and age covariate terms. Both age and age-squared were included because the relationship between age and number of PPR-using days was expected to be non-linear (i.e., as aged increased, the number of PPR-using days would taper).

Several post-estimation analyses were used to consider ancillary research questions about female-male differences and potential cannabis involvement (i.e., cannabis use before first PPR use). The predictive model was kept simple enough for use in clinical and public health practice with individual patients, and this exploration did not extend to include potentially explanatory variables that might not be routinely measured in the practice setting, such as school achievement, current employment, disposable income levels, or other drug use. [We note that these other variables actually might be determined by extra-medical PPR use and would represent model misspecification]. In addition, we wanted to consider the subpopulation of young adults who started using extra-medical PPR use within the past 12 months to examine whether the relationship with PPR-use days and alcohol involvement stayed consistent.

All analyses were conducted using Stata Version 14 software procedures for ZIP models and complex analysis-weighted survey data (Stata Corp, 2015). These steps addressed the NSDUH complex sample design, interdependence of survey observations, and analysis weights, with ‘subpopulation’ focus on 12-to-20-year-old newly incident extra-medical PPR users. We stress precision of the informative study estimates with a focus on 95% confidence intervals (CIs); p-values may aid interpretation and inference.

Results

Among US 12-to-20-year-olds with no prior extra-medical use of PPRs, an estimated 2.7% qualified as newly incident extra-medical PPR users. Table 1 describes study sample characteristics for these young newly incident users (unweighted n=16,125), with mean age at 17 years old and male-female balance. Overall, roughly 12% had never consumed alcohol, while an estimated 12% qualified as recently active cases of alcohol dependence. Race/ethnicity distributions are unremarkable. ZIP estimates reported below changed little with race/ethnicity adjustment (data not shown in a table).

Table 1.

Selected characteristics of 12-to-20-year-old newly incident extra-medical PPR users (n=16,125). Data from the National Surveys on Drug Use and Health, United States, 2005–2014.

Sample characteristics Newly incident users (n) Unweighted %a Weighted %a
Sex
 Female 8,297 51.4 49.7
 Male 7,828 48.6 50.3
Age in years (when interviewed)b 16,125 16.8 (0.02) 17.1 (0.02)
Race/ethnicity
 Non-Hispanic White 10,602 65.7 67.6
 Non-Hispanic Black 1,556 9.7 10.3
 Non-Hispanic Other 1,402 8.7 4.9
 Hispanic 2,565 15.9 17.1
Alcohol involvement
 Never drank in lifetime 1,909 11.8 11.6
 Drank at least once, but not in past year 1,139 7.1 7.0
 Drank in past year, no alcohol dependence 11,167 69.3 69.6
 Drank in past year, alcohol dependence 1,910 11.8 11.8
Number of days used extra-medical PPRs in the past yearb 16,125 20.3 (0.37) 19.5 (0.49)
Survey year, 2005–2014 ~1,600/year ~10%/year ~10%/year
Open in a new tab
a

Due to rounding, percentages may not add to 100%.

b

Mean (standard error).

Table 2 provides a display of covariate-adjusted estimates from the ZIP inflate equation and from the ZIP count equation with respect to our primary hypothesis about alcohol dependence and the persistence of extra-medical PPR use once it starts. Estimates from the unadjusted (Model 1) and adjusted models (Model 2) were similar. The inflate estimates speak to membership in the ‘not-susceptible-to-persistence’ class, with never-drinkers as the reference category. Alcohol dependent users and non-dependent recent drinkers were less likely to be in the ‘not-susceptible-to-persistence’ class (ORs = 0.5 and 0.8; 95% CIs = 0.4, 0.6 and 0.7, 0.9, respectively; Model 2). Underage drinkers who drank in the past but not recently were more likely to be in the ‘not-susceptible-to-persistence’ class versus never-drinkers (OR = 1.5; 95% CI = 1.2, 1.9; Model 2).

Table 2.

Estimated covariate-adjusted Zero-inflated Poisson model associations linking alcohol involvement with persistent extra-medical use of PPRs. Data from the National Surveys on Drug Use and Health, United States, 2005–2014.

Model 1a Model 2b
Inflate (not-susceptible-to-persistence class) ORc 95% CI p-value ORc 95% CI p-value
Alcohol involvement
 Never drank in lifetime (reference) 1.0 -- -- 1.0 -- --
 Drank at least once, but not in the past year 1.5 1.2, 1.9 <0.001 1.5 1.2, 1.9 <0.001
 Drank in the past year, no alcohol dependence 0.8 0.7, 0.9 <0.001 0.8 0.7, 0.9 <0.001
 Drank in the past year, alcohol dependence 0.5 0.4, 0.5 <0.001 0.5 0.4, 0.6 <0.001
Count (susceptible-to-persistence class) RRd 95% CI p-value RRd 95% CI p-value
Alcohol involvement
 Never drank in lifetime (reference) 1.0 -- -- 1.0 -- --
 Drank at least once, but not in the past year 1.2 0.9, 1.6 0.159 1.3 1.0, 1.7 0.063
 Drank in the past year, no alcohol dependence 0.8 0.7, 1.0 0.028 0.9 0.8, 1.1 0.344
 Drank in the past year, alcohol dependence 1.2 1.0, 1.4 0.099 1.3 1.1, 1.6 0.012
Open in a new tab
a

Model 1 included terms for alcohol involvement and survey year.

b

Model 2 included terms for alcohol involvement, age, age-squared, sex, and survey year. Neither sex nor age/age-squared were required for inclusion in the inflate part of the model (p-values >0.05).

c

Odds ratio (OR) estimate from the inflate part of the model (transformed log odds of being a member of the not-susceptible-to-persistence class).

d

Rate ratio (RR) estimate from the count part of the model, conditional on susceptible-to-persistence class membership.

The ZIP count equation of Table 2 provides evidence that the alcohol dependent extra-medical PPR users had a higher rate of extra-medical PPR use compared to never-drinkers, conditional on being ‘susceptible-to-persistence’ (RR = 1.3; 95% CI = 1.1, 1.6; Model 2). The corresponding RR estimate for past drinkers with no alcohol dependence also is 1.3, but the 95% CI included the null value of 1.0 (95% CI = 1.0, 1.7; Model 2). For recent drinkers with no alcohol dependence, the RR estimate is essentially null (95% CI = 0.8, 1.1; Model 2).

After working through our alcohol dependence focus, we turned to a series of post-estimation exploratory analysis steps, which started by introducing a covariate for cannabis use that predated extra-medical PPR use (an attribute of ~4% of this study’s newly incident extra-medical PPR users). This cannabis covariate did not predict membership in the ‘not-susceptible-to-persistence’ class (p=0.085). Cannabis history did predict an increased conditional rate of extra-medical PPR use (p=0.025; data not shown in a table).

Then, we considered estimates from a ZIP model with no covariate adjustment for level of alcohol involvement, in which we found that being male was not a predictor of ‘susceptible-to-persistence’ class membership (Table 3). For the most part, the odds of membership in this class did not vary with age per se, but the statistical model showed a somewhat better fit with the age-squared term, which allowed odds of class membership to be modestly elevated at the larger age values (p=0.034; Table 3).

Table 3.

Estimated covariate-adjusted Zero-inflated Poisson model associations linking age and sex with persistent extra-medical use of PPRs. Data from the National Surveys on Drug Use and Health, United States, 2005–2014.

Inflate (not-susceptible-to-persistence class) ORa 95% CI p-value
Male 0.97 0.89, 1.06 0.509
Age (at assessment) 0.76 0.56, 1.03 0.079
Age-squared 1.01 >1.00, 1.02 0.034
Count (susceptible-to-persistence class) RRb 95% CI p-value
Male 0.90 0.81, 1.01 0.080
Age (at assessment) 0.86 0.67, 1.14 0.293
Age-squared 1.00 0.99, 1.01 0.426
Open in a new tab
a

Odds ratio (OR) estimate from the inflate part of the model (transformed log odds). Survey year was also included as a nominal variable.

b

Rate ratio (RR) estimate from the count part of the model, conditional on persistence of extra-medical PPR use. Survey year was included as a nominal variable.

Next, we estimated variation in number of days of extra-medical PPR use during 12-months, which had a mean of approximately 20 days accounting for all newly incident users (Table 1). Table 4 shows alcohol-related variation in days of extra-medical PPR use, with largest values for alcohol dependence cases (29 days versus ~18–19 days) . Setting a threshold of two days, we found that almost two-thirds (66%) of underage alcohol dependence-affected extra-medical PPR users exceeded that level, but only 52% of non-cases did so.

Table 4.

Days of extra-medical PPR use after initial use, stratified by levels of alcohol involvement. Data from the National Surveys on Drug Use and Health, United States, 2005–2014.

Number of extra-medical PPR use days (%)
Mean (SE)a,b 0 1 >1
Alcohol involvement
 Never drank in lifetime 19.4 (1.5) 43.5 7.9 48.6
 Drank at least once, but not in the past year 18.7 (2.2) 54.8 5.7 39.5
 Drank in the past year, no alcohol dependence 17.5 (0.5) 39.0 9.3 51.7
 Drank in the past year, alcohol dependence 28.8 (1.7) 27.8 6.4 65.8
Open in a new tab
a

Weighted mean with standard error. Note the range of extra-medical PPR use days in the past year for all levels of alcohol involvement spanned 0–365. The exception was ‘Drank at least once, but not in the past year,’ which had a range 0–363 days.

b

Unweighted marginal predicted means show that young people with AD have an average of 30 PPR-use days compared to 17–19 days for the other alcohol involvement use categories shown above.

In a final step, we limited our sample to the 6,435 young people who had started using extra-medical PPR within the past 1–12 months (rather than 1–24 months). The purpose was to learn whether some individuals might have started use more than 12 months before assessment, persisted with repeated uses during the next proximal months, but subsequently used just one time in the 12 months prior to assessment, and then stopped altogether. As such, these individuals would be counted as PPR users who used once and only once during during the 12 months prior to assessment. We suspect that this is a relatively small number of individuals, especially if they are affected by active alcohol dependence. Nevertheless, the set might exist because the survey items on days of use during the 12-month interval before assessment are ‘gated’ and are not asked unless there has been at least one occasion of use during that 12-month interval. If the set is large, the result would be an even greater than expected zero inflation in the count distribution, and the cause of this component of zero-inflation would be individuals who actually were susceptible-to-persist but failed to continue to use more than one day during the 12-months prior to assessment.

This sensitivity analysis confirmed our initial hypothesis about alcohol dependence and being susceptible-to-persist. Its results show alcohol dependent users, as well as non-dependent recent drinkers, as less likely to be in the ‘not-susceptible-to-persistence’ class (ORs = 0.4 and 0.8; 95% CIs = 0.3, 0.6 and 0.6, 0.9, respectively). However, given a reduced statistical power due to the more restricted numbers of cases in this sensitivity analysis sample, the estimated conditional rates of extra-medical PPR use did not vary across the alcohol involvement categories (all p>0.05; data not shown in a table).

Discussion

For this work, in order to focus attention on intermediate steps that occur in-between the first-time extra-medical PPR use and the occurrence of opioid dependence, we turned to the ZIP regressions approach. In order to evaluate the proposition that underage drinkers with alcohol dependence might be more susceptible to persist in their extra-medical PPR use once it starts, we studied nationally representative US samples of 12-to-20-year-olds who had just started their extra-medical PPR use. We discovered that recently active alcohol dependence in underage drinkers is a susceptibility marker and may help account for persistence in extra-medical PPR use, once it has started – that is, at a quite early stage in the process that can lead toward opioid dependence. In addition, the alcohol dependence-affected underage drinkers are seen with higher rates of extra-medical PPR use given their susceptibility to persist beyond the first occasion of extra-medical PPR use.

Ancillary discoveries might be of importance. Evidence suggests no male-female differences for those newly incident 12-to-20-year-old extra-medical PPR users, consistent with prior evidence on this topic (Seedall & Anthony, 2013). Nevertheless, this work also disclosed a cannabis implication not previously identified. Namely, young people who had used cannabis before their initial PPR use were more likely to be members of the susceptible-to-persistence class. Here, it is possible that cannabis use reflects social or neighborhood circumstances mentioned in our introduction, such as local area availability of a supply of PPR, in that antecedent cannabis use was not associated the conditional rate of extra-medical PPR use. This evidence extends early cannabis-PPR sequencing (‘gateway’) studies that were reported decades ago (Yamaguchi & Kandel, 1984b, 1984a).

We encountered an unanticipated ancillary result – namely, an increased odds of membership in the ‘not-susceptible-to-persistence’ class for the subgroup of past drinkers who drank at least once but not in the past year, with an elevated rate of PPR use given persistence. We can offer little more than speculation about this finding. One speculation that might be investigated involves student athletes who sign pledges not to drink, but whose injuries set up chances for escalating PPR doses beyond what a prescriber intended (Veliz, Epstein-Ngo, Austic, Boyd, & McCabe, 2015). Another speculative possibility is that these are individuals who tried alcohol, did not like the experience, and would not like to repeat drinking again, with something similar happening when they started using opioid PPR extra-medically. These also could be adolescents living in what might be characterized as ‘desert’ conditions with respect to drug-taking experiences. They tried alcohol one time and they later tried extra-medical PPR use one time, and they simply have not experienced another chance to repeat the experience, even though they might like to do so. Yet another speculation is that they progressed fairly rapidly from first drink to alcohol problems, stopped drinking, then tried PPR and realized that persistence of PPR use might mean new and unwanted PPR problems akin to the alcohol problems they have been trying to avoid by not drinking. Of course, the most parsimonious explanation is that this is an unexpected fluke statistical finding of the study, never to be reproduced in any future study of this type. Before taking any of the speculations seriously, we think it will be important to show reproducibility. In addition, given membership in the susceptible-to-persistence class of extra-medical PPR use, there was no significant excess rate of extra-medical PPR use for past drinkers. In constraining the analysis to young people whose onset of extra-medical PPR began in the past year, we also did not find a higher rate of PPR use based on alcohol involvement. It may be that one year is not enough time to capture an acceleration in extra-medical PPR use for recent drinkers.

Before detailed discussion, some strengths and limitations should be noted. These new epidemiological findings are from a large well-measured sample of young people with newly incident extra-medical PPR use. All newly incident extra-medical PPR users under study were identified via cross-sectional nationally representative surveys. Longitudinal extension of this research, building from these initial estimates, will be useful. However, longitudinal studies will face complexities not encountered here, such as sample attrition and measurement reactivity that sometimes complicate repeated measures research (Meier et al., 2012; Seedall & Anthony, 2015). We are mindful that this was a study of relationships observed in one-time survey assessment data. To check on the possibility that extra-medical PPR use actually pre-dated onset of drinking in this adolescent sample, we compared the onset ages and found that 97% of newly incident extra-medical PPR users started drinking before extra-medical PPR onset.

A potential complication for both cross-sectional and longitudinal research is left-truncation such that some newly incident extra-medical PPR users in the population are not observed in the sample. For example, in any population’s newly incident users, fatal overdose at first ‘trial’ use implies missed non-persistent users. However, fatal overdose seems to be a fairly remote event relative to frequency of extra-medical PPR initiation. Another primary origin of left-truncation can be rapid incarceration in a correctional institution or the start of long-stay hospitalization (possibly connected with rapid onset of severely disabling opioid dependence). These instances are quite rare in a community sample of young people year by year, as are rapid ‘institutionalization process’ as well as non-participation of those who are included in the sample but do not participate (i.e., missed persistent users). Furthermore, self-report measurement artifacts might prompt over-reporting of zeroes when respondents recall drug use over a 12-month interval (United States, 2010).

ZIP models are useful for a count outcome with excess zeroes, and have been used to study patients seeking treatment in a given year, including estimation of treatment-seeking rates conditional upon the initial treatment episode (Bohning, Dietz, Schlattmann, Mendonca, & Kirchner, 1999; Mullahy, 1998). As mentioned, the most recent tobacco applications used ZIP models to estimate smoking rates conditional on membership in the latent class of those ‘susceptible-to-persistence’ of smoking (Barondess et al., 2010; Hardin & Hilbe, 2007; Hilbe, 1999; Liu & Powers, 2007). In our population of newly incident adolescent extra-medical PPR users, the ZIP model handled the excess probability of zeroes better than alternatives we evaluated, such as the negative binomial model.

With respect to our findings, we note an implication that underage alcohol involvement, including alcohol dependence, might be a signal of future hazards when PPRs are prescribed. Given an observed excess in the conditional probability of becoming dependent on extra-medical PPRs when alcohol is involved, the origins of the dependence process might be traced back to the earliest stage of extra-medical PPR use (Vsevolozhskaya & Anthony, 2015). Moreover, in newly incident users, genetic susceptibility traits might be important as soon as the first opioid use occurs and might govern acceleration in the trajectory of PPR use (i.e., with an increased rate of extra-medical PPR days of use).

Our findings set the stage for future research on the important topic of extra-medical PPR use, but also might be useful when applied in research on drugs such as cannabis and cocaine. With shorter follow-up intervals (e.g., under one year, possibly monthly), it will become possible to approximate what might be found in both nationally representative samples and health plan settings to encompass both medically prescribed and extra-medical PPR use. Future research will clarify the hypothesized roles of underage drinking and alcohol dependence, as well as susceptibility traits such as chronic and acute pain, or characteristics such as peer and family histories of extra-medical PPR use, and SNPs or gene influences and gene-environment interactions. Applied here for the first time in research on alcohol dependence and underage drinking as potential susceptibility markers in the earliest stages of progression from the first extra-medical PPR use toward opioids dependence, the ZIP regressions approach also should prove to be useful in studies that lead toward precision medicine and evaluation of genetic susceptibility traits when clinicians who treat adolescents must make a selection from among alternative prescriptions for pain relief.

Conclusions

There now is more intensive clinical surveillance to detect extra-medical PPR use, as well as down-regulation of prescribing PPRs for young people, in an effort to shape modifiable practical public health practices in response to the opioid overdose epidemic in complement with epidemiological evidence of the type presented here (Jones, Paulozzi, & Mack, 2014). Current guidelines for clinical diagnosis, therapeutics, and case management permit opioid prescribing for pediatric patients in some carefully judged contexts of pain relief. However, these guidelines do not yet explicitly address underage drinking, alcohol dependence, or other potential vulnerabilities of adolescents susceptible to persistent extra-medical PPR use. Some research suggests underage drinking is not a prominent issue in these prescribing decisions (Chou et al., 2009; Dowell, Haegerich, & Chou, 2016). This study’s evidence suggests that underage drinking should not be neglected when clinicians, teachers, parents and other caregivers learn that an adolescent has started to engage in extra-medical use of PPRs. When pain relievers are prescribed for an adolescent patient, the patient’s history of underage drinking and alcohol dependence deserve consideration.

Public Significance.

Underage drinking should not be neglected when clinicians, teachers, parents and other caregivers learn that young people have started to engage in use of prescription pain relievers outside a prescriber’s boundaries.

Acknowledgments

This work was supported by the National Institute on Drug Abuse under Grant T32DA021129; National Institute on Drug Abuse Senior Scientist and Mentorship Award under Grant K05DA015799; and by Michigan State University. The content is the sole responsibility of the authors and does not necessarily represent the official views of Michigan State University, National Institute on Drug Abuse, or the National Institutes of Health.

Authors Parker and Anthony designed the study. Author Parker conducted literature searches, provided summaries of previous research studies, and conducted the analyses. All authors contributed to and have approved the final manuscript.

We would like the thank the United States Substance Abuse and Mental Health Services Administration Center for Behavioral Health Statistics and Quality for sponsoring the National Survey on Drug Use and Health and making the datasets available for public use to allow research of this nature.

Footnotes

Some ideas and data appearing in the manuscript were presented at the 76th Meeting, College on Problems of Drug Dependence, San Juan, Puerto Rico. A subset of the data and concepts were also a part of author Parker’s doctoral dissertation.

Disclosures

The authors report no conflicts of interests.

Contributor Information

Maria A. Parker, Michigan State University, Department of Epidemiology & Biostatistics

James C. Anthony, Michigan State University, Department of Epidemiology & Biostatistics

References

  1. American Psychiatric Association. DSM-IV: Diagnostic and Statistical Manual of Mental Disorders. 4. Washington, DC: American Psychiatric Association; 1994. [Google Scholar]
  2. Anthony JC, Warner LA, Kessler RC. Comparative epidemiology of dependence on tobacco, alcohol, controlled substances, and inhalants: Basic findings from the National Comorbidity Survey. Experimental and Clinical Psychopharmacology. 1994;2(3):244–268. doi: 10.1037/1064-1297.2.3.244. [DOI] [Google Scholar]
  3. Barondess DA, Meyer EM, Boinapally PM, Fairman B, Anthony JC. Epidemiological evidence on count processes in the formation of tobacco dependence. Nicotine & Tobacco Research: Official Journal of the Society for Research on Nicotine and Tobacco. 2010;12(7):734–741. doi: 10.1093/ntr/ntq073. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bohning D, Dietz E, Schlattmann P, Mendonca L, Kirchner U. The Zero-Inflated Poisson Model and the Decayed, Missing and Filled Teeth Index in Dental Epidemiology. Journal of the Royal Statistical Society. Series A (Statistics in Society) 1999;162(2):195–209. [Google Scholar]
  5. Chou R, Fanciullo GJ, Fine PG, Adler JA, Ballantyne JC, Davies P, … Miaskowski C. Clinical Guidelines for the Use of Chronic Opioid Therapy in Chronic Noncancer Pain. The Journal of Pain. 2009;10(2):113–130e22. doi: 10.1016/j.jpain.2008.10.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Dowell D, Haegerich TM, Chou R. CDC Guideline for Prescribing Opioids for Chronic Pain—United States, 2016. MMWR. Recommendations and Reports. 2016;65(1):1–49. doi: 10.15585/mmwr.rr6501e1. [DOI] [PubMed] [Google Scholar]
  7. Hardin JW, Hilbe JM. Generalized Linear Models and Extensions. 2. Stata Press; 2007. [Google Scholar]
  8. Hilbe JM. Zero-inflated Poisson and negative binomial regression. Stata Technol Bull. 1999;8:233–236. [Google Scholar]
  9. Jones C, Paulozzi L, Mack K. Sources of prescription opioid pain relievers by frequency of past-year nonmedical use: United states, 2008–2011. JAMA Internal Medicine. 2014;174(5):802–803. doi: 10.1001/jamainternmed.2013.12809. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Keegan K, Moss H. Chasing the High: A Firsthand Account of One Young Person’s Experience with Substance Abuse. 1. Oxford University Press; 2008. [Google Scholar]
  11. Lasagna L, Von Felsinger JM, Beecher HK. Drug-induced mood changes in man. I. Observations on healthy subjects, chronically ill patients, and postaddicts. Journal of the American Medical Association. 1955;157(12):1006–1020. doi: 10.1001/jama.1955.02950290026009. [DOI] [PubMed] [Google Scholar]
  12. Liu H, Powers DA. Growth Curve Models for Zero-Inflated Count Data: An Application to Smoking Behavior. Structural Equation Modeling: A Multidisciplinary Journal. 2007;14(2):247–279. doi: 10.1080/10705510709336746. [DOI] [Google Scholar]
  13. Martins SS, Storr CL, Zhu H, Chilcoat HD. Correlates of extramedical use of OxyContin versus other analgesic opioids among the US general population. Drug and Alcohol Dependence. 2009;99(1–3):58–67. doi: 10.1016/j.drugalcdep.2008.06.013. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Meier EA, Troost JP, Anthony JC. Extramedical use of prescription pain relievers by youth aged 12 to 21 years in the United States: national estimates by age and by year. Archives of Pediatrics & Adolescent Medicine. 2012;166(9):803–807. doi: 10.1001/archpediatrics.2012.209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Mullahy J. Much ado about two: reconsidering retransformation and the two-part model in health econometrics. Journal of Health Economics. 1998;17(3):247–281. doi: 10.1016/s0167-6296(98)00030-7. [DOI] [PubMed] [Google Scholar]
  16. Parker MA, Anthony JC. Epidemiological evidence on extra-medical opioid prescription use: Rapid transitions from newly incident use to dependence among 12–21 year olds in the United States using meta-analysis, 2002–2013. PeerJ. 2015;3:e1340. doi: 10.7717/peerj.1340. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Reyes-Gibby CC, Yuan C, Wang J, Yeung S-CJ, Shete S. Gene network analysis shows immune-signaling and ERK1/2 as novel genetic markers for multiple addiction phenotypes: alcohol, smoking and opioid addiction. BMC Systems Biology. 2015;9(1) doi: 10.1186/s12918-015-0167-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Schepis TS, Krishnan-Sarin S. Characterizing Adolescent Prescription Misusers: A Population-Based Study. Journal of the American Academy of Child. 2008;47(7):745–754. doi: 10.1097/CHI.0b013e318172ef0d. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Seedall RB, Anthony JC. Risk estimates for starting tobacco, alcohol, and other drug use in the United States: Male–female differences and the possibility that ‘limiting time with friends’ is protective. Drug and Alcohol Dependence. 2013;133(2):751–753. doi: 10.1016/j.drugalcdep.2013.06.035. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Seedall RB, Anthony JC. Monitoring by Parents and Hypothesized Male-Female Differences in Evidence from a Nationally Representative Cohort Re-sampled from Age 12 to 17 Years: An Exploratory Study Using a “Mutoscope” Approach. Prevention Science. 2015:1–11. doi: 10.1007/s11121-014-0517-8. [DOI] [PMC free article] [PubMed]
  21. Stata Corp. Stata Statistical Software: Release 14 (Version 14) College Station, TX: Stata Corp LP; 2015. [Google Scholar]
  22. Sung HE, Richter L, Vaughan R, Johnson PB, Thom B. Nonmedical use of prescription opioids among teenagers in the United States: Trends and correlates. Journal of Adolescent Health. 2005;37(1):44–51. doi: 10.1016/j.jadohealth.2005.02.013. [DOI] [PubMed] [Google Scholar]
  23. United States. Reliability of Key Measures in the National Survey on Drug Use and Health. Substance Abuse and Mental Health Services Administration; 2010. Office of Applied Studies, Methodology Series M-8, HHS Publication No. SMA 09-4425. [PubMed] [Google Scholar]
  24. Veliz P, Epstein-Ngo Q, Austic E, Boyd C, McCabe SE. Opioid Use Among Interscholastic Sports Participants: An Exploratory Study From a Sample of College Students. Research Quarterly for Exercise and Sport. 2015;86(2):205–211. doi: 10.1080/02701367.2014.983219. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Vsevolozhskaya OA, Anthony JC. Transitioning from First Drug Use to Dependence Onset: Illustration of a Multiparametric Approach for Comparative Epidemiology. Neuropsychopharmacology. 2015 doi: 10.1038/npp.2015.213. [DOI] [PMC free article] [PubMed]
  26. Wagner FA, Anthony JC. Into the World of Illegal Drug Use: Exposure Opportunity and Other Mechanisms Linking the Use of Alcohol, Tobacco, Marijuana, and Cocaine. American Journal of Epidemiology. 2002;155(10):918–925. doi: 10.1093/aje/155.10.918. [DOI] [PubMed] [Google Scholar]
  27. Yamaguchi K, Kandel DB. Patterns of drug use from adolescence to young adulthood: II. Sequences of progression. American Journal of Public Health. 1984a;74(7):668–672. doi: 10.2105/ajph.74.7.668. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Yamaguchi K, Kandel DB. Patterns of drug use from adolescence to young adulthood: III. Predictors of progression. American Journal of Public Health. 1984b;74(7):673–681. doi: 10.2105/ajph.74.7.673. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES