Prescription drug abuse is the fastest-growing substance abuse problem in the United States (Centers for Disease Control and Prevention 2012). Since 2003, Nonmedical prescription drug use (NMPDU) of opiate analgesics resulted in more unintentional overdoses than heroin and cocaine combined (Centers for Disease Control and Prevention 2012). Given the rising prevalence rates and related significant public health implications (Huang et al. 2006; Blanco et al. 2007; Substance Abuse and Mental Health Services Administration 2008), NMPDU is becoming an increasingly important focus of research. NMPDU is often defined as use of prescription medication (1) without a prescription, or (2) in greater amounts, more frequently, or for longer than recommended by a prescribing professional.

NMPDU potentially differs from other types of illicit drug use in important ways. When prescription drugs are obtained legally for legitimate use, but then misused, exposure may be less associated with externalizing behaviors (e.g., the antisocial symptoms seen in other drug use disorders; Fenton et al. 2010), and more associated with internalizing problems such as anxiety. This is supported by research indicating that among those with anxiety disorders, those with a prescription are more likely than those without a prescription to engage in NMPDU, controlling for disorder severity. Thus, the prescriptions themselves confer risk among those with clinically significant anxiety (Fenton et al. 2010). Given that, legitimate drug prescriptions expose individuals to addictive substances who might not otherwise seek out illicit drugs, perhaps thereby activating an underlying genetic diathesis for addiction. This could alter the traditional pattern of comorbid disorders commonly seen with most illicit drug use.

By studying comorbidity in a pairwise fashion, others have shown that many psychiatric diagnoses are associated with increased risk for NMPDU, including mood, anxiety, substance use, and personality disorders (Fenton et al. 2010). Having an anxiety (OR = 1.46), mood (OR = 2.13), drug use (OR = 15.71), or personality disorder (OR = 2.57) was related to increased odds of NMPDU (Fenton et al. 2010). NMPDU has also been related to higher rates of sensation seeking (Arria et al. 2008) and comorbid alcohol use disorders (McCabe et al. 2006), as well as drunk driving (McCabe 2005) and other delinquent behaviors (Goldstein 2008). One study found that in individuals with prescription opioid dependence, 47.1 % had a comorbid mood or anxiety disorder (Gros et al. 2013). Although these pair-wise approaches offer important insights into comorbidity, they provide only a limited understanding, because they do not provide information about how NMPDU relates to multiple diagnoses in a more conceptually unified and parsimonious manner. Therefore, interest is growing in understanding the position of specific mental disorders within a comprehensive model of cormorbid mental disorders.

Extant research suggests that common mental disorder comorbidity reflects relations to higher-order internalizing and externalizing factors in a meta-structure of psychiatric comorbidity (Krueger et al. 1998; Eaton et al. 2012). The internalizing factor accounts for the shared variance amongst unipolar mood disorders and anxiety disorders, including major depressive disorder, generalized anxiety disorder, social and specific phobias, and panic disorder. The externalizing factor is defined by the disinhibitory and substance use disorders, including illicit drug, alcohol, and nicotine dependence; antisocial personality disorder and antisocial traits; pathological gambling; and conduct disorder. Within the higher-order internalizing factor, two highly correlated sub-factors are often found: (1) distress, representing mood disorders and some anxiety disorders (e.g., generalized anxiety disorder), and (2) fear, representing more paroxysmal anxiety disorders such as panic disorder and social and specific phobias (Eaton et al. 2013).

Further, although many prescription drugs have addictive potential, not all people are at equal risk of developing NMPDU problems. For example, there may be differential levels of risk for NMPDU by gender. Prescriptions are known to confer risk for NMPDU (Arria et al. 2008), and women are more likely than men to obtain a prescription for certain classes of prescription drugs (Simoni-Wastila 1998). However, findings on NMPDU by gender are mixed, with one study in a clinical sample showing that women are at significantly greater risk than men to engage in NMPDU when controlling for non-gender demographic and other diagnostic variables (Simoni-Wastila et al. 2004), while an epidemiological study shows higher prevalence and greater risk of NMPDU in men (Huang et al. 2006). However, overall, men have higher risk for NMPDU abuse and dependence (Huang et al. 2006).

In the current study, we first aim to place NMPDU within research on the meta-structure of psychopathology. While Eaton et al. (2012) and others have modeled drug use more generally in the large epidemiological datasets, NMPDU specifically has never been modeled in any overall multivariate comorbidity context. By placing NMPDU into the meta-structure model, we compare its relationship to latent internalizing and externalizing factors, which will clarify if NMPDU is a separate phenomenon from illicit drug use, or if it is indeed another manifestation of the same diathesis that leads to illicit drug abuse and dependence. To date, bivariate studies of NMPDU comorbidity suggest NMPDU could be associated with either the internalizing domain (depression, anxiety), the externalizing domain (ASPD, drug use disorders), or both. Because prescription drugs are often prescribed to treat internalizing disorders and prescriptions confer risk for NMPDU (Fenton et al. 2010), NMPDU may load on internalizing or externalizing factors, and might be an unusual case of a substance use problem with a strong relationship to the internalizing dimension. By placing NMPDU in the metastructure model, we pursue an understanding of how patterns of comorbidities confer risk to NMPDU, and what underlying psychopathological factors may be primarily responsible.

We then elaborate on this basic structural question by comparing the meta-structural model including NMPDU in men and women. The question of gender invariance across the psychopathology metastructure was examined in a prior report (Eaton et al. 2013), but that study did not include NMPDU. Studying gender invariance with NMPDU is particularly important because NMPDU may be differentially related to the internalizing and externalizing spectrums in men and women, accounting for differences in rates of NMPDU. For example, many prescription drugs are prescribed to treat conditions in the internalizing spectrum, and women show higher levels of internalizing psychopathology (Eaton et al. 2013). Hence, women may have more exposure to NMPDU through prescriptions intended to treat internalizing psychopathology, potentially leading to gender differences in the way NMPDU relates to the comorbidity meta-structure. Further, the literature remains inconclusive with respect to the relative prevalence of NMPDU in men and women (Jamison et al. 2010; Simoni-Wastila et al. 2004; Fenton et al. 2010; Huang et al. 2006).

Method

Participants

This study utilized data from 43,093 individuals who participated in the first wave of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC). NESARC is a representative sample survey of adults 18 years or older living in the United States, conducted in 2001–2002, with an oversampling of African-Americans, Hispanics, and young adults. Data were weighted to be representative of the United States civilian population based on the 2000 Census. Weighted and unweighted demographic data can be found in Table 1. A more comprehensive description of the study design can be found in Grant et al. (2004). As reported therein, all potential respondents were informed in writing about the nature of the survey, uses of the survey data, voluntary nature of their participation, and legally mandated confidentiality of identifiable survey information. The research protocol received full ethical review and approval from the U.S. Census Bureau and the U.S. Office of Management and Budget.

Table 1 Demographics of NESARC wave 1 sample
Full size table

Assessment

Nonmedical Prescription Drug Use

Information on four classes of NMPDU was collected using the Alcohol Use Disorder and Associated Disabilities Interview Schedule-DSM-IV Edition (AUDADIS-IV; Grant et al. 1995). Participants were asked about “…medicines and other kinds of drugs that you may have used on your own—that is, either without a doctor’s prescription; in greater amounts, more often, or longer than prescribed; or for a reason other than a doctor said you should use them”. Participants were then asked about their use of the 4 prescription drug classes: sedatives, tranquilizers, opioids, and stimulants. They were reminded to endorse the item only if they used the drug “on their own”, to ensure endorsement only if they used without a prescription or differently than prescribed. The question is designed to include any use of a legitimately prescribed drug that is nonmedical in nature, and all use without a prescription. Respondents with prescriptions who use their medication as indicated would not endorse this item. All NMPDU drugs were assessed as lifetime variables. NMPDU items demonstrated “moderate” to “very good” reliability, with kappas ranging from 0.50 to 0.82 (Altman 1991).

Psychiatric Disorders

The AUDADIS-IV assessed lifetime DSM-IV diagnoses including major depressive disorder, dysthymic disorder, generalized anxiety disorder, panic disorder, social phobia, specific phobia, alcohol dependence, nicotine dependence, marijuana dependence, and antisocial personality disorder. See Table 2 for a description of weighted disorder prevalences. We modeled diagnoses of alcohol, nicotine, and marijuana dependence (as opposed to other substances) so the substance dependence variables in the model involved use of different substances from those where nonmedical use of prescription drugs was queried. The AUDADIS-IV was administered by trained lay interviewers. It has shown moderate to very good diagnostic reliability, with kappas ranging from 0.42 to 0.84 (Altman 1991).

Table 2 Prevalence rates of lifetime NMPDU and other diagnoses
Full size table

Statistical Analyses

All analyses were conducted in Mplus version 6 (Muthén and Muthén 2010) using the Mplus defaults of delta parameterization and the WLSMV estimator for confirmatory factor analyses (CFA) and WLSM estimator and oblique Geomin rotation for exploratory factor analyses (EFA). To provide an additional and more sensitive index of the differential fit, additional CFAs were conducted using the MLR estimator. All analyses treated diagnoses as categorical variables and incorporated the NESARC design feature variables to ensure representativeness. To evaluate model fit in EFA and CFAs, we compared the comparative fit index (CFI), Tucker-Lewis index (TLI), and root mean squared error of approximation (RMSEA). Models with CFI/TLI values > 0.95 and RMSEA values < 0.06 suggest good model fit (Hu and Bentler 1999). When the indices did not clearly favor one model, the models were also fit using the MLR estimator, which yielded three additional measures of model fit; the Bayesian Information Criterion (BIC), the sample size corrected BIC, and the Akaike Information Criterion (AIC). Lower AIC and BIC values indicate closer model fit.

Exploratory factor analysis was used to examine whether it was reasonable to model all classes of prescription drugs as a single prescription drug use variable. Then, since previous work has shown the NESARC diagnostic data conform to the internalizing-externalizing structure of common mental disorders (Eaton et al. 2012), each diagnosis was parameterized to load on one of three factors previously identified, with (1) major depressive disorder, dysthymic disorder, generalized anxiety disorder, and posttraumatic stress disorder loading on the distress sub-factor of internalizing, (2) panic disorder, social phobia, and specific phobia loading of the fear sub-factor of internalizing, and (3) antisocial personality disorder, alcohol dependence, marijuana dependence, and nicotine dependence loading on the externalizing factor.

Confirmatory factor analysis was used to determine the fit of the basic internalizing-externalizing model, with distress and fear loading on higher-order internalizing. Distress and fear loadings on internalizing were constrained to equality to ensure model identification. Then, model fit was compared when NMPDU loaded on (1) distress, (2) fear, (3) externalizing, (4) distress and fear, (5) distress and externalizing, (6) fear and externalizing, and (7) distress, fear, and externalizing. Finally, gender invariance was tested by comparing constrained and unconstrained versions of the best fitting model using multiple groups CFA. In the constrained condition, thresholds and loadings were constrained to equality across gender, and factor means and scaling factors were fixed at 0 and 1 respectively in men, and free to vary in women. In the unconstrained models, loadings and thresholds were free across gender, while factor means and scaling factors were fixed to 0 and 1 respectively in both genders. Model fit was then compared between conditions. Cheung and Rensvold propose that a decrement in CFI of 0.01 or greater from unconstrained to constrained conditions should be considered a significant reduction in model fit when the data are constrained (Cheung and Rensvold 2002).

Results

Do Different Prescription Drug Classes Form a Single Nonmedical Prescription Drug Use Factor?

The four classes of prescription drugs (sedatives, tranquilizers, stimulants, and opioids) were highly inter-correlated (average tetrachoric r = 0.82, range = 0.76 to 0.87). EFA analyses showed that the four classes were well represented by a single factor, as indicated by a nearly perfect fit of the one-factor model to the data (CFI = 0.999, TLI = 0.998, RMSEA = 0.015). All standardized loadings on the single factor were greater than 0.85. Drug misuse for any prescription drug class was therefore treated as a single variable in subsequent analyses. For thoroughness, the models described below were also run separately for each drug class to ensure that all classes behaved similarly. The optimal model differed only minimally in fit amongst the different drug classes. RMSEA values were identical for each of the four drug classes modeled separately and for all drug classes as a single variable, while CFI and TLI values differed by 0.004 at the most between all five models. Loadings were all within 0.053 of each other.

Where Does Nonmedical Prescription Drug Use Fit Within the Meta-Structure?

We then compared the fits of models where NMPDU loaded separately on the seven possible factor combinations. We ran these analyses separately for men and women, in order to understand any potential differences between groups in the way that prescription drug use fit into the model. Table 3 shows the model fit statistics for both men and women for all models. Fit statistics for models that placed prescription drug use on externalizing; fear and externalizing; distress and externalizing; and fear, distress, and externalizing; and externalizing alone, were similar. Using WLSMV estimated fit statistics, all models that included a loading on externalizing has very similar statistics, with low loadings on all factors other than externalizing. The three models that included an externalizing factor loading differed from each other only minimally (For women: RMSEA range = 0.012–0.013, CFI range = 0.992 for all models, TLI range = 0.989–0.990; For men: RMSEA range = 0.013 for all models, CFI range = 0.990 for all models, TLI range = 0.987 for all models). Because fit was so similar across models, the MLR estimator was also used to derive more sensitive indices of differential fit between the models. In women, the Fear and Externalizing model fit better than the model that included Externalizing alone, in terms of BIC, corrected BIC, and AIC. In men, the Fear and Externalizing model again outperformed all other models across fit indices. Standardized loadings on the Fear factor are low but significant, and appear to be consistent across gender groups and estimator types (MLR and WLSMV). Because this secondary Fear loading appears to provide a consistent advantage over the other models, while still providing a parsimonious solution, it was selected as optimal.

Table 3 Fit indices for NMPDU in models for women and men
Full size table

Figure 1 shows standardized loadings for the Fear and Externalizing model for all disorders. In the optimally fitting model where NMPDU loads on both Fear and Externalizing, the factor loading of NMPDU on externalizing was 0.83 in women and 0.81 in men, meaning that 69 % of the variance in NMPDU in women and 66 % in men was accounted for by factor variance shared with other externalizing disorders. The factor loading on Fear accounted for a small (albeit statistically significant) proportion of variance; loadings on fear were −0.10 in men and −0.14 in women, meaning that an additional 1 % and 2 % of the variances were accounted for by the Fear factor.

Fig. 1
figure 1

Figure displays standardized factor loadings of all diagnoses (listed below) and NMPDU on the latent Fear, Distress, and Externalizing factors. Loadings are provided separately for men (in boldface) and women (no boldface). All loadings are significant at p < 0.001 with one exception, i.e., for the loading indicated by *, p = 0.011. Arrows indicating unique variances, including error, are excluded to improve legibility. aMajor depressive disorder. bDysthymia. cGeneralized anxiety disorder. dPanic disorder. eSocial phobia. fSpecific phobia. gNonmedical prescription drug use. hNicotine dependence. iAlcohol dependence. jMarijuana dependence. kAntisocial personality disorder

Full size image

Is the Model Gender InvariantFootnote 1?

After finding that the externalizing and fear model was optimal in both men and women, we next tested how similar these models were across gender by comparing model parameters—that is, whether the magnitude of parameters differed significantly across gender. Fit statistics were compared between constrained (gender invariant) and unconstrained (non-gender invariant) models to determine if the same model could characterize multivariate comorbidity equivalently well in each gender. We fit the unconstrained and constrained externalizing and fear models in men and women simultaneously via multi-group confirmatory factor analysis. Table 4 shows the constrained and unconstrained versions for the best fitting model that loads NMPDU on externalizing and fear. Model fit was evaluated using Cheung and Rensvold’s (2002) criteria in which a reduction in CFI of 0.01 or greater was considered a significant decrement in model fit when data are constrained. Parsimony and model fit statistics in the two models were also compared. The constrained model provided advantages over the unconstrained model in terms of freely estimated parameters (39 in constrained vs. 50 in unconstrained), and TLI (0.990 vs. 989), and identical RMSEA (0.012) and CFI values (0.991). Better model fit in the constrained model indicated that the general structure, as well as specific factor loadings and thresholds, were the same in men and women. The placement of NMPDU in the psychopathology meta-structure was therefore gender invariant.

Table 4 Fit statistics comparing unconstrained and constrained fear and externalizing models
Full size table

Discussion

Previous research has established that NMPDU co-occurs with specific psychiatric diagnoses more often than by chance. Bivariate approaches to modeling this comorbidity (e.g., odds ratios) show that a variety of disorders are associated with NMPDU, but studies have not yet modeled liability to NMPDU using methods that conceptualize relationships among multiple diagnoses simultaneously. While others have modeled different types of illicit drug use within the metastructure, NMPDU specifically has never been modeled in a multivariate comorbidity context. Because NMPDU could be a distinct phenomenon from general illicit drug use, with separate contributing factors, this paper provides an important extension to the literature by being the first to investigated the placement of NMPDU within a multivariate model of common mental disorders.

By fitting prescription drug misuse into the psychopathology meta-structure, we were able to identify NMPDU as a potent indicator of the externalizing spectrum. Sixty-nine percent of the variance in NMPDU in women and 66 % in men was accounted for by factor variance shared with other externalizing disorders. Another modest 1 % and 2 % of the variances in men and women respectively were accounted for by the Fear factor. The negative loading of NMPDU on Fear, although small, appeared to improve model fit and was consistent across gender and model estimator type. Although this finding should not be over-interpreted, it is interesting insofar as one might expect the opposite, in that disorders that load highly on fear (Panic Disorder, Specific Phobia, Social Phobia) are often conceptualized as risk (as opposed to protective) factors for NMPDU (Martins et al. 2012). Rather, our results suggest that it is perhaps the case that fearful individuals may fear the addictive or psychoactive properties of these medications and thus be more likely to refrain from NMPDU. Our results do not test this interpretation directly, and as there are many alternative interpretations, this interpretation is currently speculative, warranting further research.

This finding organizes the literature that generally shows a diffuse pattern of psychiatric comorbidities for NMPDU by providing a cogent liability model for the development of problems with prescription drugs. The finding focuses attention on the externalizing liability factor and suggests that NMPDU may be understood as part of a group or “spectrum” of interrelated externalizing behaviors. Because of its placement on the externalizing factor, NMPDU appears to share substantial variance with other drug use disorders, including illicit drug dependence, and is likely to share etiological antecedents with other forms of drug misuse as well. As such, the finding serves to focus both research and clinical attention on the externalizing spectrum and the ways in which NMPDU and other externalizing syndromes, are closely related phenomena, topics we will now explore in greater depth.

Nonmedical Prescription Drug Use as an Externalizing Disorder: Implications for Clinicians

Although prescription drugs are often prescribed to treat internalizing disorders, and prescriptions are known to confer risk for NMPDU, internalizing disorders do not appear to confer risk for NMPDU. In fact, NMPDU actually showed a slightly negative relationship to the fearful subset of internalizing disorders. Although we must be cautious not to over-interpret these findings, it does appear that fear disorders do not pose additional liability for NMPDU, and may confer a slight protective effect against NMPDU, after accounting for externalizing. Because NMPDU loaded instead on the externalizing factor, clinicians should be on alert for externalizing characteristics when prescribing drugs with abuse potential. While clinicians who prescribe prescription drugs may be aware of the tendency of individuals with substance use histories to misuse their prescriptions (Arria et al. 2008), this study illustrates empirically that the general externalizing liability can be manifested as NMPDU. This work also expands clinical risk assessment beyond the standard indicators of externalizing liability, e.g., patients’ past history of illicit or other substance abuse, to include other externalizing indicators that clinicians may be less likely to consider when evaluating risk, such as antisocial personality disorder or traits. Clinicians may additionally find it useful to consider the broader spectrum of externalizing traits and disorders. Prescribers can better anticipate which patients are at highest risk of developing substance use disorders secondary to legitimate drug prescriptions if they conceptualize NMPDU as part of a related set of disorders that share antecedent risk factors.

We also found that the factorial structure and loadings of NMPDU was gender invariant. This finding suggests that observed gender differences in prevalence rates originate from differences in the latent internalizing and externalizing factor levels (Krueger et al. 2005). In the case of gender, clinicians can conceptualize men and women as being differentially susceptible to NMPDU problems only insofar as they exhibit different levels of externalizing on average. For example, men can be assumed to be more at risk for NMPDU generally because of higher documented rates of latent externalizing, but a man with no evidence of externalizing would be at lower risk than a woman high in externalizing behaviors. With regard to previous findings that women use more prescription drugs but men have higher prevalence of NMPDU disorders, these findings illustrate that it is likely not mere exposure to prescription drugs that confers risk, but exposure in conjunction with an externalizing diathesis.

Previous research (e.g., Simoni-Wastila et al. 2004) may have found increased liability to NMPDU in women because they controlled for other externalizing disorders in the relationship between gender and NMPDU, which occur at greater rates in men. One benefit of a modeling approach is that it allows us to formally test equality of measurement across groups, so that conclusions drawn from invariance testing cannot be due to differences in measurement, meaning we can conclude that we were able to equally measure and model externalizing in men and women. Additionally, the modeling approach provides a method of observing the shared liability between related disorders without artificially controlling for disorders that in reality co-occur because of that shared liability. As such, we conclude that the greater rates of NMPDU seen in men in large-scale epidemiological studies are due to higher levels of externalizing in men compared with women.

Implications for Substance Use and Nonmedical Prescription Drug Use Research

NMPDU is an important phenomenon to study, and merits attention separately from nonmedical prescription drug abuse or dependence, and from other kinds of drug use, because of its high prevalence, negative health outcomes, and high fiscal and societal costs. NMPDU’s high loadings on externalizing (as high as those of nicotine dependence and antisocial personality disorder in males) suggests that NMPDU is an important indicator of externalizing liability that should be incorporated into future externalizing research. This research also extends the concept of liability for externalizing beyond categorical diagnoses (i.e., drug abuse or dependence) by demonstrating that the liability is also present at the level of NMPDU, suggesting that manifestations of the externalizing factor extend beyond strictly defined DSM diagnostic categories (Krueger et al. 2005; Markon and Krueger 2005) into “sub-diagnostic” phenomena such as NMPDU.

Limitations and Future Directions

The survey question assessing NMPDU in these data does not directly distinguish between legal and illegal acquisition of prescription drugs. As a result, this study cannot separate misuse of legally-acquired prescription drugs and illegal use of prescription drugs. Future research would benefit from further refinement in survey methods to obtain more detailed information about specific episodes of NMPDU and corresponding specific pathways to NMPDU, to obtain a detailed history of legal and illegal means of acquisition for specific episodes. Although collecting this extensive information could add to respondent burden, the current research suggests this may be worthwhile in refining our understanding of NMPDU comorbidity. Future research could also explore the relationship between NMPDU and other disorders that lie along different continua using a metastructure approach, such as psychotic disorders or somatization, or include assessment of adaptive and maladaptive personality traits. This kind of inquiry might enrich the liability model and further inform risk factors for developing problems with NMPDU, and by extension, other externalizing psychopathologies.